diff --git a/.gitattributes b/.gitattributes index 5d79390a2..0c35cf7c5 100644 --- a/.gitattributes +++ b/.gitattributes @@ -1,2 +1,19 @@ +# Font-build test fixtures are compared byte for byte: a checkout must not convert them. +crates/fonts-build/tests/fixture/** -text + +# The pinned font index is hashed and compiled in as committed: a checkout must not convert its +# line endings, or `Index::bundled` fails its checksum on Windows. +crates/font-store/index/sources.json -text +crates/font-store/index/indexes/*.gz binary + # Copies of the vendored krilla source diff, one per patch (see vendor/krilla/README.craft.md). vendor/krilla/patches/*.patch linguist-generated=true + +# Has a pinned digest and is embedded as bytes by craft-font-store. +crates/font-store/index/sources.json -text + +# The build fixture verifies these license files are copied byte-for-byte. +crates/fonts-build/tests/fixture/craft-fonts/fonts/**/OFL.txt -text + +# DXF drawing goldens assert exact writer bytes, including LF on every platform. +crates/dxf/tests/data/drawing_*.dxf -text diff --git a/.github/workflows/actions.lock b/.github/workflows/actions.lock new file mode 100644 index 000000000..02cdf71c8 --- /dev/null +++ b/.github/workflows/actions.lock @@ -0,0 +1,188 @@ +# This file is machine-generated by `gh actions-lock`. +# Do not edit by hand; run `gh actions-lock` to update. +# Docs: https://gh.io/actions-lockfile +version: 'v0.0.3' +workflows: + '.github/workflows/ai-core.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/atomic.yml': + - 'actions/checkout@v4.4.0' + - 'actions/checkout@v7.0.1' + - 'dtolnay/rust-toolchain@stable' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/audio.yml': + - 'actions/cache@v6.1.0' + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/rust-cache@v2.9.2' + '.github/workflows/cms.yml': + - 'actions/checkout@v4.4.0' + - 'dtolnay/rust-toolchain@stable' + '.github/workflows/crw.yml': + - 'actions/cache@v6.1.0' + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/rust-cache@v2.9.2' + '.github/workflows/dialogs.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/doc.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/docpath.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/ecproj-format.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/font-store.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/formats.yml': + - 'actions/checkout@v7.0.1' + - 'dtolnay/rust-toolchain@stable' + - 'step-security/harden-runner@v2.22.4' + - 'swatinem/rust-cache@v2.9.3' + '.github/workflows/house-rules.yml': [] + '.github/workflows/i18n.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/jpeg.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/rust-cache@v2.9.2' + '.github/workflows/keymap.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/rust-cache@v2.9.2' + '.github/workflows/lockfile.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/media.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/rust-cache@v2.9.2' + '.github/workflows/menu.yml': + - 'actions/checkout@v7.0.1' + - 'dtolnay/rust-toolchain@stable' + - 'swatinem/rust-cache@v2.9.3' + '.github/workflows/pcraft-format.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/pdf.yml': + - 'actions/checkout@v7.0.1' + - 'actions/setup-python@v7.1.0' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/rust-cache@v2.9.2' + - 'step-security/taiki-e-install-action@v2.87.32' + '.github/workflows/print.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/proof.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/rust-cache@v2.9.2' + '.github/workflows/sonarcloud.yml': [] + '.github/workflows/ui.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/rust-cache@v2.9.2' + '.github/workflows/update.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/vectorcraft-format.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/video.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/rust-cache@v2.9.2' + '.github/workflows/web-api.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/taiki-e-install-action@v2.87.32' + '.github/workflows/web-host.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + - 'step-security/rust-cache@v2.9.2' + - 'step-security/taiki-e-install-action@v2.87.32' + '.github/workflows/window.yml': + - 'actions/checkout@v7.0.1' + - 'step-security/dtolnay-rust-toolchain@v1.0.0' + - 'step-security/harden-runner@v2.22.4' + '.github/workflows/workspace.yml': [] +dependencies: + 'actions/cache@v6.1.0': + ref: 'v6.1.0' + commit: 'sha1-55cc8345863c7cc4c66a329aec7e433d2d1c52a9' + owner_id: 44036562 + repo_id: 215566462 + 'actions/checkout@v4.4.0': + ref: 'v4.4.0' + commit: 'sha1-11d5960a326750d5838078e36cf38b85af677262' + owner_id: 44036562 + repo_id: 197814629 + 'actions/checkout@v7.0.1': + ref: 'v7.0.1' + commit: 'sha1-3d3c42e5aac5ba805825da76410c181273ba90b1' + owner_id: 44036562 + repo_id: 197814629 + 'actions/setup-python@v7.1.0': + ref: 'v7.1.0' + commit: 'sha1-6d4fd22b20465ccc0f7354476588e8df7c9ca0e0' + owner_id: 44036562 + repo_id: 192625525 + 'dtolnay/rust-toolchain@stable': + ref: 'stable' + commit: 'sha1-39008827e6b4319a349e9a42dfc90579dd25f1ca' + owner_id: 1940490 + repo_id: 260749683 + 'step-security/dtolnay-rust-toolchain@v1.0.0': + ref: 'v1.0.0' + commit: 'sha1-8dd607977d06adf4ed49b190381e74c17a9dd25f' + owner_id: 88700172 + repo_id: 1249865941 + 'step-security/harden-runner@v2.22.4': + ref: 'v2.22.4' + commit: 'sha1-1d88384fd173af6ecb012dfd0264a306b711241a' + owner_id: 88700172 + repo_id: 422287306 + 'step-security/rust-cache@v2.9.2': + ref: 'v2.9.2' + commit: 'sha1-8871c8ba063416da7c32246932ea6ae08700453e' + owner_id: 88700172 + repo_id: 822636713 + 'step-security/taiki-e-install-action@v2.87.32': + ref: 'v2.87.32' + commit: 'sha1-76747d22a7567cc4438546a84f11bdf18e670ab8' + owner_id: 88700172 + repo_id: 1382753260 + 'swatinem/rust-cache@v2.9.3': + ref: 'v2.9.3' + commit: 'sha1-d1527c8f67ef9af357290475d4af4b0fab8f12e0' + owner_id: 580492 + repo_id: 298565987 diff --git a/.github/workflows/ai-core.yml b/.github/workflows/ai-core.yml index cc8d738ac..5471bf1ec 100644 --- a/.github/workflows/ai-core.yml +++ b/.github/workflows/ai-core.yml @@ -28,11 +28,11 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy diff --git a/.github/workflows/atomic.yml b/.github/workflows/atomic.yml index 1a20fa679..f72a64f1e 100644 --- a/.github/workflows/atomic.yml +++ b/.github/workflows/atomic.yml @@ -28,11 +28,11 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy @@ -50,7 +50,7 @@ jobs: name: Atomic WASM runs-on: ubuntu-latest steps: - - uses: actions/checkout@v4 + - uses: actions/checkout@v4.4.0 - uses: dtolnay/rust-toolchain@stable with: targets: wasm32-unknown-unknown diff --git a/.github/workflows/audio.yml b/.github/workflows/audio.yml index 0104f848e..d6a7ff5e2 100644 --- a/.github/workflows/audio.yml +++ b/.github/workflows/audio.yml @@ -25,18 +25,18 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy targets: wasm32-unknown-unknown - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-audio-lint save-if: ${{ github.ref == 'refs/heads/main' }} @@ -44,7 +44,7 @@ jobs: # Each feature on its own, so code that needs another feature fails here. - name: Clippy, each feature alone run: | - for f in "" decode decode-native symphonia encode lossless-encode pcm-encode flac-encode peaks aac-decode aac-encode "aac-encode,rayon" opus-decode ac3-decode mpa-decode serde; do + for f in "" decode decode-native symphonia encode lossless-encode pcm-encode flac-encode peaks aac-decode aac-encode "aac-encode,rayon" opus-decode opus-encode ac3-decode mpa-decode serde; do echo "::group::features: ${f:-none}" cargo clippy -p craft-audio --all-targets --locked --no-default-features ${f:+--features "$f"} -- -D warnings echo "::endgroup::" @@ -69,15 +69,15 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false # The workspace's rust-version. - run: rustup toolchain install 1.95.0 --profile minimal - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-audio-msrv save-if: ${{ github.ref == 'refs/heads/main' }} @@ -88,16 +88,16 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-audio-features save-if: ${{ github.ref == 'refs/heads/main' }} @@ -113,20 +113,20 @@ jobs: runs-on: ubuntu-26.04 steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-audio-test save-if: ${{ github.ref == 'refs/heads/main' }} - - uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0 + - uses: actions/cache@v6.1.0 with: path: ${{ runner.temp }}/media-fixtures/opus/vectors # The archive's SHA-256 (scripts/fetch-opus-vectors.sh checks it). diff --git a/.github/workflows/cms.yml b/.github/workflows/cms.yml index a15a48c9d..74ce0815f 100644 --- a/.github/workflows/cms.yml +++ b/.github/workflows/cms.yml @@ -16,7 +16,7 @@ jobs: timeout-minutes: 20 runs-on: ubuntu-latest steps: - - uses: actions/checkout@v4 + - uses: actions/checkout@v4.4.0 - uses: dtolnay/rust-toolchain@stable with: components: clippy diff --git a/.github/workflows/crw.yml b/.github/workflows/crw.yml index 8fec98661..54be7ba3b 100644 --- a/.github/workflows/crw.yml +++ b/.github/workflows/crw.yml @@ -29,17 +29,17 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-crw-${{ matrix.os }} save-if: ${{ github.ref == 'refs/heads/main' }} @@ -53,17 +53,17 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable targets: wasm32-unknown-unknown - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-crw-wasm save-if: ${{ github.ref == 'refs/heads/main' }} @@ -75,21 +75,21 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-crw-samples save-if: ${{ github.ref == 'refs/heads/main' }} - name: Cache the CC0 sample files - uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0 + uses: actions/cache@v6.1.0 with: path: ${{ runner.temp }}/crw-samples key: crw-samples-${{ hashFiles('crates/crw/tests/samples.json') }} diff --git a/.github/workflows/dialogs.yml b/.github/workflows/dialogs.yml index 97a931aec..c8d43d725 100644 --- a/.github/workflows/dialogs.yml +++ b/.github/workflows/dialogs.yml @@ -23,15 +23,15 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: # The token isn't needed after checkout; cargo runs build scripts and proc macros from # dependencies, so don't leave it in .git/config for that third-party code to read. persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: clippy diff --git a/.github/workflows/doc.yml b/.github/workflows/doc.yml index a0c619c1e..6caa4f0ee 100644 --- a/.github/workflows/doc.yml +++ b/.github/workflows/doc.yml @@ -28,13 +28,13 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy diff --git a/.github/workflows/docpath.yml b/.github/workflows/docpath.yml index 571b65eca..e0115f6f5 100644 --- a/.github/workflows/docpath.yml +++ b/.github/workflows/docpath.yml @@ -28,11 +28,11 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy diff --git a/.github/workflows/ecproj-format.yml b/.github/workflows/ecproj-format.yml index 16bab05c7..24a3350af 100644 --- a/.github/workflows/ecproj-format.yml +++ b/.github/workflows/ecproj-format.yml @@ -28,13 +28,13 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy diff --git a/.github/workflows/font-store.yml b/.github/workflows/font-store.yml index f246f621a..0578f18ec 100644 --- a/.github/workflows/font-store.yml +++ b/.github/workflows/font-store.yml @@ -28,11 +28,11 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy diff --git a/.github/workflows/formats.yml b/.github/workflows/formats.yml index e19902e8c..84d2c0d7e 100644 --- a/.github/workflows/formats.yml +++ b/.github/workflows/formats.yml @@ -72,17 +72,17 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: dtolnay/rust-toolchain@b725eda525362e7cf34af9a7243cac1e3f01a7ec # stable + - uses: dtolnay/rust-toolchain@stable with: components: rustfmt, clippy targets: wasm32-unknown-unknown - - uses: Swatinem/rust-cache@d1527c8f67ef9af357290475d4af4b0fab8f12e0 # v2.9.3 + - uses: Swatinem/rust-cache@v2.9.3 with: key: ${{ matrix.crate }} save-if: ${{ github.ref == 'refs/heads/main' }} diff --git a/.github/workflows/i18n.yml b/.github/workflows/i18n.yml index 5f717aeff..7e6903188 100644 --- a/.github/workflows/i18n.yml +++ b/.github/workflows/i18n.yml @@ -28,11 +28,11 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy diff --git a/.github/workflows/jpeg.yml b/.github/workflows/jpeg.yml index 900baf678..21db0ab79 100644 --- a/.github/workflows/jpeg.yml +++ b/.github/workflows/jpeg.yml @@ -25,18 +25,18 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy targets: wasm32-unknown-unknown - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-jpeg save-if: ${{ github.ref == 'refs/heads/main' }} diff --git a/.github/workflows/keymap.yml b/.github/workflows/keymap.yml index aba535265..41fcc0502 100644 --- a/.github/workflows/keymap.yml +++ b/.github/workflows/keymap.yml @@ -27,24 +27,24 @@ jobs: CRAFT_FONTS_DIR: ${{ github.workspace }}/.test-inputs/craft-fonts steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - name: Checkout test fonts - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + uses: actions/checkout@v7.0.1 with: repository: storytold/craft-fonts ref: b076417b5880bd4154bd22f00695b1d28c6bea83 path: .test-inputs/craft-fonts persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-keymap-linux save-if: ${{ github.ref == 'refs/heads/main' }} @@ -71,23 +71,23 @@ jobs: CRAFT_FONTS_DIR: ${{ github.workspace }}/.test-inputs/craft-fonts steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - name: Checkout test fonts - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + uses: actions/checkout@v7.0.1 with: repository: storytold/craft-fonts ref: b076417b5880bd4154bd22f00695b1d28c6bea83 path: .test-inputs/craft-fonts persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-keymap-${{ matrix.os }} save-if: ${{ github.ref == 'refs/heads/main' }} @@ -98,17 +98,17 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable targets: wasm32-unknown-unknown - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-keymap-wasm save-if: ${{ github.ref == 'refs/heads/main' }} diff --git a/.github/workflows/lockfile.yml b/.github/workflows/lockfile.yml new file mode 100644 index 000000000..e630c9a8f --- /dev/null +++ b/.github/workflows/lockfile.yml @@ -0,0 +1,34 @@ +name: Lockfile + +# Fails when .github/workflows/actions.lock is out of sync with the workflows, or when a locked pin +# no longer verifies upstream, so a moved tag cannot slip a different commit into a run. +# Docs: https://gh.io/actions-lockfile +on: + pull_request: + push: + branches: [main] + workflow_dispatch: + +permissions: + contents: read + +env: + GH_TOKEN: ${{ github.token }} + +jobs: + lockfile: + name: Lockfile is in sync + runs-on: ubuntu-24.04 + timeout-minutes: 10 + steps: + - name: Harden runner + uses: step-security/harden-runner@v2.22.4 + with: + egress-policy: audit + - uses: actions/checkout@v7.0.1 + with: + persist-credentials: false + - name: Install gh-actions-lock + run: gh extension install github/gh-actions-lock --pin v0.4.1 + - name: Verify the lockfile + run: gh actions-lock --verify diff --git a/.github/workflows/media.yml b/.github/workflows/media.yml index cb3912d54..4814e35ca 100644 --- a/.github/workflows/media.yml +++ b/.github/workflows/media.yml @@ -25,18 +25,18 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy targets: wasm32-unknown-unknown - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-media-lint save-if: ${{ github.ref == 'refs/heads/main' }} @@ -56,16 +56,16 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-media-test save-if: ${{ github.ref == 'refs/heads/main' }} diff --git a/.github/workflows/menu.yml b/.github/workflows/menu.yml index 6d0e082cb..8d1058731 100644 --- a/.github/workflows/menu.yml +++ b/.github/workflows/menu.yml @@ -21,14 +21,14 @@ jobs: name: Menu (Linux) runs-on: ubuntu-latest steps: - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: dtolnay/rust-toolchain@b725eda525362e7cf34af9a7243cac1e3f01a7ec # stable + - uses: dtolnay/rust-toolchain@stable with: components: rustfmt, clippy targets: aarch64-apple-darwin, wasm32-unknown-unknown - - uses: Swatinem/rust-cache@d1527c8f67ef9af357290475d4af4b0fab8f12e0 # v2.9.3 + - uses: Swatinem/rust-cache@v2.9.3 with: key: craft-menu-linux - run: cargo fmt -p craft-menu --check diff --git a/.github/workflows/pcraft-format.yml b/.github/workflows/pcraft-format.yml index 177bb4593..9c66b5a10 100644 --- a/.github/workflows/pcraft-format.yml +++ b/.github/workflows/pcraft-format.yml @@ -28,13 +28,13 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: "1.95" components: rustfmt, clippy diff --git a/.github/workflows/pdf.yml b/.github/workflows/pdf.yml index a494d085c..09ddfabca 100644 --- a/.github/workflows/pdf.yml +++ b/.github/workflows/pdf.yml @@ -31,19 +31,19 @@ jobs: RUSTFLAGS: --cfg craft_pdf_unicode_fixtures steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-pdf save-if: ${{ github.ref == 'refs/heads/main' }} - - uses: actions/setup-python@v7 + - uses: actions/setup-python@v7.1.0 with: python-version: "3.12" - name: Provision licensed PDF test fonts @@ -61,14 +61,14 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-pdf-backends save-if: ${{ github.ref == 'refs/heads/main' }} @@ -81,19 +81,19 @@ jobs: CARGO_TARGET_WASM32_UNKNOWN_UNKNOWN_RUNNER: wasm-bindgen-test-runner steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable targets: wasm32-unknown-unknown - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-pdf-wasm save-if: ${{ github.ref == 'refs/heads/main' }} - - uses: actions/setup-python@v7 + - uses: actions/setup-python@v7.1.0 with: python-version: "3.12" - name: Provision licensed PDF test fonts @@ -111,7 +111,7 @@ jobs: - name: Install the matching WASM test runner # A prebuilt, checksum-verified binary instead of building it from source on every run. # A version the pinned action does not list yet falls back to `cargo install --locked`. - uses: step-security/taiki-e-install-action@76747d22a7567cc4438546a84f11bdf18e670ab8 # v2.87.32 + uses: step-security/taiki-e-install-action@v2.87.32 with: tool: wasm-bindgen@${{ steps.wasm-bindgen.outputs.version }} fallback: cargo-install @@ -122,19 +122,19 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: clippy - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-plate-test save-if: ${{ github.ref == 'refs/heads/main' }} - - uses: actions/setup-python@v7 + - uses: actions/setup-python@v7.1.0 with: python-version: "3.12" - name: Provision licensed PDF test fonts diff --git a/.github/workflows/print.yml b/.github/workflows/print.yml index 3838e6c42..8e2108e09 100644 --- a/.github/workflows/print.yml +++ b/.github/workflows/print.yml @@ -24,11 +24,11 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy diff --git a/.github/workflows/proof.yml b/.github/workflows/proof.yml index 0035b1c04..10365a230 100644 --- a/.github/workflows/proof.yml +++ b/.github/workflows/proof.yml @@ -25,27 +25,27 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false # The Language data (craft-i18n `language-data/`), embedded by crates/proof/build.rs. # Pinned to storytold/craft-i18n#114's head (#109's SCOWL dictionaries + judgment/judgement, # CP2.1); TODO(CP2.1): move to #114's merge commit once it merges. - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: repository: storytold/craft-i18n ref: c239f934ace90b7939afaceb23b472ceeafd6b8a path: craft-i18n persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy targets: wasm32-unknown-unknown - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-proof save-if: ${{ github.ref == 'refs/heads/main' }} diff --git a/.github/workflows/sonarcloud.yml b/.github/workflows/sonarcloud.yml index 654fe4c02..e40c1b4bc 100644 --- a/.github/workflows/sonarcloud.yml +++ b/.github/workflows/sonarcloud.yml @@ -23,5 +23,5 @@ jobs: # The codec and format code behind features that are off by default. Without these the # coverage build never compiles it, and new code there reads as untested. cargo-features: >- - craft-video/dnx,craft-video/vp8,craft-video/prores,craft-video/h264,craft-video/hevc,craft-video/vp9,craft-video/av1,craft-video/threads,craft-audio/decode,craft-audio/encode,craft-audio/peaks,craft-audio/rayon,craft-audio/serde,craft-media/matroska,craft-media/ogg,craft-media/bitstream,craft-media/serde,craft-mesh/stl,craft-mesh/obj,craft-mesh/ply,craft-mesh/gltf,craft-metafile/write + craft-image-codecs/avif,craft-image-codecs/avif-decode,craft-video/dnx,craft-video/vp8,craft-video/prores,craft-video/h264,craft-video/hevc,craft-video/vp9,craft-video/av1,craft-video/threads,craft-audio/decode,craft-audio/encode,craft-audio/peaks,craft-audio/rayon,craft-audio/serde,craft-media/matroska,craft-media/ogg,craft-media/bitstream,craft-media/serde,craft-mesh/stl,craft-mesh/obj,craft-mesh/ply,craft-mesh/gltf,craft-metafile/write,craft-image-codecs/sgi,craft-image-codecs/sunraster,craft-jsonrpc/async-lines secrets: inherit diff --git a/.github/workflows/ui.yml b/.github/workflows/ui.yml index 0892f5a2c..f15ef6ddc 100644 --- a/.github/workflows/ui.yml +++ b/.github/workflows/ui.yml @@ -30,21 +30,21 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 - name: Install software Vulkan renderer if: runner.os == 'Linux' run: | sudo apt-get update sudo apt-get install -y mesa-vulkan-drivers libvulkan1 echo 'WGPU_BACKEND=vulkan' >> "$GITHUB_ENV" - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-ui save-if: ${{ github.ref == 'refs/heads/main' }} @@ -59,15 +59,15 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: actions/checkout@v7.0.1 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable targets: wasm32-unknown-unknown - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-ui-wasm save-if: ${{ github.ref == 'refs/heads/main' }} diff --git a/.github/workflows/update.yml b/.github/workflows/update.yml index f9c8e1498..9a7139026 100644 --- a/.github/workflows/update.yml +++ b/.github/workflows/update.yml @@ -28,15 +28,15 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: # The token isn't needed after checkout; cargo runs build scripts and proc macros from # dependencies, so don't leave it in .git/config for that third-party code to read. persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy diff --git a/.github/workflows/vectorcraft-format.yml b/.github/workflows/vectorcraft-format.yml index b2b20463a..74144f30f 100644 --- a/.github/workflows/vectorcraft-format.yml +++ b/.github/workflows/vectorcraft-format.yml @@ -28,13 +28,13 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: "1.95" components: rustfmt, clippy diff --git a/.github/workflows/video.yml b/.github/workflows/video.yml index ec1c16eda..fa667bdfc 100644 --- a/.github/workflows/video.yml +++ b/.github/workflows/video.yml @@ -22,18 +22,18 @@ jobs: runs-on: ubuntu-latest steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: "1.95" components: rustfmt, clippy targets: wasm32-unknown-unknown - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-video-lint - run: cargo fmt -p craft-video --check @@ -60,16 +60,16 @@ jobs: name: Video tests (${{ matrix.os }}) steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: "1.95" - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-video-${{ matrix.os }} - name: Install ffmpeg (Linux) diff --git a/.github/workflows/web-api.yml b/.github/workflows/web-api.yml index 2460fb569..4bf1a30fe 100644 --- a/.github/workflows/web-api.yml +++ b/.github/workflows/web-api.yml @@ -26,13 +26,13 @@ jobs: CARGO_TARGET_WASM32_UNKNOWN_UNKNOWN_RUNNER: wasm-bindgen-test-runner steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy @@ -46,7 +46,7 @@ jobs: - name: Install the matching WASM test runner # A prebuilt, checksum-verified binary instead of building it from source on every run. # A version the pinned action does not list yet falls back to `cargo install --locked`. - uses: step-security/taiki-e-install-action@76747d22a7567cc4438546a84f11bdf18e670ab8 # v2.87.32 + uses: step-security/taiki-e-install-action@v2.87.32 with: tool: wasm-bindgen@${{ steps.wasm-bindgen.outputs.version }} fallback: cargo-install diff --git a/.github/workflows/web-host.yml b/.github/workflows/web-host.yml index b780ee76b..76e6701e5 100644 --- a/.github/workflows/web-host.yml +++ b/.github/workflows/web-host.yml @@ -27,18 +27,18 @@ jobs: CARGO_TARGET_WASM32_UNKNOWN_UNKNOWN_RUNNER: wasm-bindgen-test-runner steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 + - uses: actions/checkout@v7.0.1 with: persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy targets: wasm32-unknown-unknown - - uses: step-security/rust-cache@8871c8ba063416da7c32246932ea6ae08700453e # v2.9.2 + - uses: step-security/rust-cache@v2.9.2 with: key: craft-web-host save-if: ${{ github.ref == 'refs/heads/main' }} @@ -55,7 +55,7 @@ jobs: - name: Install the matching WASM test runner # A prebuilt, checksum-verified binary instead of building it from source on every run. # A version the pinned action does not list yet falls back to `cargo install --locked`. - uses: step-security/taiki-e-install-action@76747d22a7567cc4438546a84f11bdf18e670ab8 # v2.87.32 + uses: step-security/taiki-e-install-action@v2.87.32 with: tool: wasm-bindgen@${{ steps.wasm-bindgen.outputs.version }} fallback: cargo-install diff --git a/.github/workflows/window.yml b/.github/workflows/window.yml index f570ef5ce..fdd28e203 100644 --- a/.github/workflows/window.yml +++ b/.github/workflows/window.yml @@ -28,15 +28,15 @@ jobs: runs-on: ${{ matrix.os }} steps: - name: Harden runner - uses: step-security/harden-runner@1d88384fd173af6ecb012dfd0264a306b711241a # v2.22.4 + uses: step-security/harden-runner@v2.22.4 with: egress-policy: audit - - uses: actions/checkout@v7 + - uses: actions/checkout@v7.0.1 with: # The token isn't needed after checkout; cargo runs build scripts and proc macros from # dependencies, so don't leave it in .git/config for that third-party code to read. persist-credentials: false - - uses: step-security/dtolnay-rust-toolchain@8dd607977d06adf4ed49b190381e74c17a9dd25f # v1.0.0 + - uses: step-security/dtolnay-rust-toolchain@v1.0.0 with: toolchain: stable components: rustfmt, clippy diff --git a/.github/workflows/workspace.yml b/.github/workflows/workspace.yml index 4002900d9..20e947480 100644 --- a/.github/workflows/workspace.yml +++ b/.github/workflows/workspace.yml @@ -10,66 +10,79 @@ on: permissions: contents: read +concurrency: + # Pull requests share one group per branch, so a newer push cancels the older run. Every other + # run (push to main, a release tag, a manual run) gets a group of its own: it is neither + # cancelled nor replaced while queued, so a main run always finishes and saves the dependency + # cache, and a release run is never cancelled. + group: workspace-${{ github.event_name == 'pull_request' && github.ref || github.run_id }} + cancel-in-progress: ${{ github.event_name == 'pull_request' }} + jobs: native: - name: Workspace (${{ matrix.os }}) - timeout-minutes: 30 - strategy: - fail-fast: false - matrix: - os: [ubuntu-latest, windows-latest, macos-latest] - runs-on: ${{ matrix.os }} - steps: - - uses: actions/checkout@v4 - - name: Install software Vulkan renderer - if: runner.os == 'Linux' - run: | - sudo apt-get update - sudo apt-get install -y mesa-vulkan-drivers libvulkan1 - echo 'WGPU_BACKEND=vulkan' >> "$GITHUB_ENV" - - uses: dtolnay/rust-toolchain@stable - with: - components: rustfmt, clippy - - uses: Swatinem/rust-cache@v2 - with: - key: workspace - - name: Prepare the Windows test driver - if: runner.os == 'Windows' - shell: powershell - run: | - $ErrorActionPreference = 'Stop' - Start-Service Spooler - Add-PrinterDriver -Name 'Generic / Text Only' - - run: cargo fmt --check - - run: cargo clippy --workspace --all-targets --locked -- -D warnings - - run: cargo clippy --workspace --all-targets --all-features --locked -- -D warnings - - run: cargo test --workspace --locked - env: - CRAFT_PRINT_REQUIRE_QUEUE: ${{ runner.os == 'Windows' && '1' || '0' }} - - run: cargo test --workspace --all-features --locked - env: - CRAFT_PRINT_REQUIRE_QUEUE: ${{ runner.os == 'Windows' && '1' || '0' }} + name: Workspace + uses: storytold/craft-common-actions/.github/workflows/gate.yml@main + with: + os: '["ubuntu-latest","windows-latest","macos-latest"]' + timeout-minutes: 120 + tools: "" + packages: mesa-vulkan-drivers libvulkan1 + craft-fonts-ref: b076417b5880bd4154bd22f00695b1d28c6bea83 + linux-env: WGPU_BACKEND=vulkan + windows-env: CRAFT_PRINT_REQUIRE_QUEUE=1 + env: CRAFT_PRINT_REQUIRE_QUEUE=0 + setup-command: | + set -euo pipefail + corpus_fetcher=.test-inputs/photocraft + mkdir -p "$corpus_fetcher" + git -C "$corpus_fetcher" init + git -C "$corpus_fetcher" remote add origin https://github.com/storytold/PhotoCraft.git + git -C "$corpus_fetcher" fetch --quiet --depth=1 origin 74330157cb601573ae4523a95bb9132822fc3918 + git -C "$corpus_fetcher" checkout --quiet --detach FETCH_HEAD + (cd "$corpus_fetcher" && cargo xtask corpus --psd --psd-tools --atn --exr --heif --pngsuite) + mkdir -p corpus + cp -a "$corpus_fetcher/corpus/." corpus/ + python scripts/test-fetch-pdf-test-fonts.py + python scripts/fetch-pdf-test-fonts.py "$RUNNER_TEMP/craft-pdf-test-fonts" + printf 'CRAFT_PSD_CORPUS=%s\n' "$GITHUB_WORKSPACE/$corpus_fetcher/corpus" >> "$GITHUB_ENV" + printf 'CRAFT_PDF_TEST_FONT_DIR=%s\n' "$RUNNER_TEMP/craft-pdf-test-fonts" >> "$GITHUB_ENV" + command: | + python3 scripts/test-workspace-concurrency.py + cargo fmt --check + cargo clippy --workspace --all-targets --locked -- -D warnings + cargo clippy --workspace --all-targets --all-features --locked -- -D warnings + cargo test --workspace --locked + cargo test --workspace --all-features --locked + windows-command: | + powershell.exe -NoProfile -Command '$ErrorActionPreference = "Stop"; Start-Service Spooler; Add-PrinterDriver -Name "Generic / Text Only"' + cargo fmt --check + cargo clippy --workspace --all-targets --locked -- -D warnings + cargo clippy --workspace --all-targets --all-features --locked -- -D warnings + cargo test --workspace --locked + cargo test --workspace --all-features --locked + msrv: - name: Rust 1.95 (Linux libraries) - runs-on: ubuntu-latest - timeout-minutes: 15 - steps: - - uses: actions/checkout@v4 - - uses: dtolnay/rust-toolchain@1.95.0 - - uses: Swatinem/rust-cache@v2 - with: - key: workspace-msrv - - run: cargo check --workspace --lib --all-features --locked + name: Rust 1.95 + uses: storytold/craft-common-actions/.github/workflows/gate.yml@main + with: + os: '["ubuntu-latest"]' + toolchain: 1.95.0 + targets: "" + tools: "" + craft-fonts-ref: "" + require-craft-fonts: false + timeout-minutes: 15 + dev-debug: "" + command: cargo check --workspace --lib --all-features --locked + wasm: name: UI and settings WASM - runs-on: ubuntu-latest - timeout-minutes: 15 - steps: - - uses: actions/checkout@v4 - - uses: dtolnay/rust-toolchain@stable - with: - targets: wasm32-unknown-unknown - - uses: Swatinem/rust-cache@v2 - with: - key: workspace-wasm - - run: cargo check -p craft-ui -p craft-settings --all-features --locked --target wasm32-unknown-unknown + uses: storytold/craft-common-actions/.github/workflows/gate.yml@main + with: + os: '["ubuntu-latest"]' + targets: wasm32-unknown-unknown + tools: "" + craft-fonts-ref: "" + require-craft-fonts: false + timeout-minutes: 15 + command: cargo check -p craft-ui -p craft-settings --all-features --locked --target wasm32-unknown-unknown diff --git a/ATTRIBUTION.md b/ATTRIBUTION.md index 965b348e4..ab167207c 100644 --- a/ATTRIBUTION.md +++ b/ATTRIBUTION.md @@ -6,6 +6,9 @@ author, source and license. Library code and original assets are under the repos | File | Author | Source | License | |---|---|---|---| +| `licenses/jpeg-encoder-0.6.1/`, `crates/image-codecs/NOTICE` | jpeg-encoder contributors and the IJG, libjpeg-turbo and SIMD authors named in the retained notices | [jpeg-encoder 0.6.1](https://crates.io/crates/jpeg-encoder/0.6.1), used unmodified; source checksums in `SOURCE.json` | (MIT OR Apache-2.0) AND IJG | +| `crates/smime/testdata/` | MailCraft contributors; RFC 9216 example certificates | [MailCraft at 240d1d42](https://github.com/UCS-Care-Group/mailcraft/tree/240d1d42fe394cf3274402b180663ba9b43ca180/crates/smime/testdata), restored unmodified; provenance in the directory README | MIT OR Apache-2.0 (source repository) | +| `licenses/dwg/acadrust-MPL-2.0.txt`, `licenses/dwg/acadrust-THIRD_PARTY_NOTICES.md` | `acadrust` contributors; ACadSharp authors for the noted ported routines | [acadrust 0.6.6](https://github.com/hakanaktt/acadrust), used unmodified as `craft-dwg`'s optional native dependency | MPL-2.0; embedded ACadSharp notice is MIT; see both retained files | | `crates/cms/profiles/photocraft-coated-cmyk.icc` | PhotoCraft contributors; generated by `craft_cms::synth::coated_cmyk` | [PhotoCraft `crates/cms`](https://github.com/storytold/photocraft/tree/b3222ee8e7e6bcb4799c37f904c595f20f9baf5e/crates/cms) | CC0-1.0 | | `crates/cms/profiles/craft-coated-fogra39.icc` | craft-libs contributors; generated by `crates/cms/tests/fogra39.rs` from the FOGRA39 data below | this repository | CC0-1.0 | | `crates/font-store/index/sources.json` (unmodified) | craft-font-index contributors | [aiongg/craft-font-index@5a9a7f8](https://github.com/aiongg/craft-font-index/tree/5a9a7f8038c8f3e2e1583a79c92a816e8e71b34f) | Apache-2.0 | @@ -82,6 +85,7 @@ and the exact transfer revision are recorded in each vendored crate's SOURCE.jso | `hayro-interpret 0.7.0 / src/font/generated/metrics.rs` | Adobe (Core 14 AFM metrics, PDF specification encodings); tables by The Hayro Authors | https://github.com/storytold/pdfcraft/blob/7437bde378f43856eada40af25b983812e615e8a/vendor/hayro-interpret/src/font/generated/metrics.rs | MIT OR Apache-2.0 | `2ec9e608573eda459de1e1d9bda3bb494315e36bee885acc88912a11fe0caa41` | Glyph widths and encodings for non-embedded standard fonts. Non-visual dependency data; existing embedding unchanged | | `crates/audio/src/ac3/golden_frames.bin` | FilmCraft contributors (original work: 12 AC-3 and E-AC-3 syncframes that ffmpeg's `ac3` and `eac3` encoders made from tones and seeded noise synthesised by ffmpeg's lavfi generators; no third-party or recorded audio) | [FilmCraft `crates/ac3`](https://github.com/storytold/filmcraft/tree/1c9c2e82/crates/ac3/src) | MIT OR Apache-2.0 | | `crates/audio/tests/fixtures/` | SoundCraft and craft-libs contributors (original work: 0.1 s test tones from ffmpeg's `sine` generator and hand-built silent MPEG frames; no third-party or recorded audio; commands in `crates/audio/tests/fixtures/README.md`) | [SoundCraft `crates/audio-io/tests/fixtures`](https://github.com/storytold/soundcraft/tree/e86d09a/crates/audio-io/tests/fixtures); `tone.mov`, `tone.mka`, `tone.opus`, `tone.ac3` and `tone.eac3` made in this repository | MIT OR Apache-2.0 | +| `crates/image-codecs/src/codecs/xpm/x11_colors.rs` | X.Org Foundation and X11 contributors (X11 `rgb.txt` colour names and values, normalised into a lookup table) | , via [PhotoCraft `crates/codecs`](https://github.com/storytold/photocraft/tree/5000240c6/crates/codecs/src/codecs/xpm) | MIT (X.Org app/rgb) | | `crates/raster-codecs/src/xpm/x11_colors.rs` | X.Org Foundation and X11 contributors (X11 `rgb.txt` colour names and values, normalised into a lookup table) | , via [PhotoCraft `crates/codecs`](https://github.com/storytold/photocraft/tree/63b641b55e54bf8f5c83efd977fdb16a2c255431/crates/codecs/src/codecs/xpm) | MIT (X.Org app/rgb) | | `crates/psd/tests/fixtures/atn/*.atn` (seven synthetic action sets, encoded field by field; `crates/psd/tests/fixtures/atn/README.md`) | PhotoCraft contributors | [PhotoCraft `crates/psd/tests/fixtures/atn`](https://github.com/storytold/photocraft/tree/63b641b55e54bf8f5c83efd977fdb16a2c255431/crates/psd/tests/fixtures/atn) | MIT OR Apache-2.0 | | `crates/lottie/tests/fixtures/` | EffectCraft contributors (original work: four hand-written Lottie JSON documents, written from the public Lottie format description; not exported from any Adobe product or third-party file) | [EffectCraft `crates/lottie/tests/fixtures`](https://github.com/storytold/effectcraft/tree/e18f48fb/crates/lottie/tests/fixtures) | MIT OR Apache-2.0 | @@ -92,3 +96,10 @@ and the exact transfer revision are recorded in each vendored crate's SOURCE.jso | `crates/raw/docs/raf-corpus.sha256` (hashes only; the files are not bundled) | The photographers who contributed each file to raw.pixls.us | [raw.pixls.us](https://raw.pixls.us/) | CC0-1.0 | | `crates/drawingml/src/preset_text.txt` | DeckCraft contributors (Nikita Rakitin); generated by `crates/drawingml/tests/fixtures/make_preset_guides.py` from the guide formulas of ECMA-376 Part 1 `presetShapeDefinitions.xml` | [storytold/deckcraft#650](https://github.com/storytold/deckcraft/pull/650) | MIT OR Apache-2.0 (the formulas are the format's definition in ECMA-376) | | `crates/drawingml/tests/fixtures/preset_table.txt` | DeckCraft contributors (Nikita Rakitin); generated by `crates/drawingml/tests/fixtures/make_preset_table.py` from ECMA-376 Part 1 `presetShapeDefinitions.xml` | [storytold/deckcraft#650](https://github.com/storytold/deckcraft/pull/650) | MIT OR Apache-2.0 (the formulas are the format's definition in ECMA-376) | + +## Binary Apple property lists + +`crates/plist` is extracted from `storytold/craft-anything` (`crates/formats/src/plist.rs` and checked byte reads), at source main `88d1d9cac852aec1c7fd97886db451e005e49d2a`. MIT OR Apache-2.0; source license texts are retained in the crate. Primary import Git authors preserve Brandon Thomas, BFlat, RRFZ and Afloat. Regression imports preserve BFlat. XML, JSON projection, calendar formatting and app policy were not imported into the public API. + +`crates/tga/src/metadata_tests.rs` and metadata-boundary helpers in `crates/tga/src/lib.rs` are adapted from nuclearspike's FilmCraft #1932 at `e7842f76f887005c460c64a8ab51d59d086e6f27` (MIT OR Apache-2.0). Synthetic fixtures are original contributor test data; primary Git author retained. +The XML reader in `crates/plist/src/xml.rs` and its base64 compatibility helper are ported from Craft Anything `8f3b775157d331dd31cfa4da911e26d8856b8c77` (same MIT OR Apache-2.0 source licenses). Brandon Thomas is the primary import author; retained hardening credits include Afloat, BFlat, Rahul, RRFZ and NetterSilver. The finite-real fix preserves gmrmad1 as author from Craft Anything #1905 (`d85e104f`); quoted-attribute/line-ending regressions preserve RRFZ. Bounds/adaptation and bounded entity scanning are subsequent Dylan M. Taylor changes. App-specific JSON projections and keyed-archive interpretation remain outside this crate. diff --git a/Cargo.lock b/Cargo.lock index cf0e0bd47..d1e5d3cec 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -2,6 +2,30 @@ # It is not intended for manual editing. version = 4 +[[package]] +name = "acadrust" +version = "0.6.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "8923c99ea1f151cb7fc17ce2fc14f8b1e330fa22611e535602d985c0cc3c89a2" +dependencies = [ + "anyhow", + "bitflags 2.13.2", + "byteorder", + "encoding_rs", + "flate2", + "foldhash 0.2.0", + "indexmap", + "itoa", + "memmap2", + "nom 7.1.3", + "oem_cp", + "once_cell", + "rayon", + "ryu", + "thiserror 2.0.21", + "web-time", +] + [[package]] name = "accesskit" version = "0.25.1" @@ -98,6 +122,24 @@ dependencies = [ "zerocopy", ] +[[package]] +name = "aligned" +version = "0.4.3" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "ee4508988c62edf04abd8d92897fca0c2995d907ce1dfeaf369dac3716a40685" +dependencies = [ + "as-slice", +] + +[[package]] +name = "aligned-vec" +version = "0.6.4" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "dc890384c8602f339876ded803c97ad529f3842aba97f6392b3dba0dd171769b" +dependencies = [ + "equator", +] + [[package]] name = "alloc-no-stdlib" version = "2.0.4" @@ -180,6 +222,23 @@ version = "1.0.104" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "330a5ed07fa54e4702c9d6c4174f74427fc0ef6e214bbd677ae50a5099946470" +[[package]] +name = "arbitrary" +version = "1.5.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "3bc62ac97cc33321f50863d514c3bc38a453947a8f9e781137e47c7401020aed" + +[[package]] +name = "arg_enum_proc_macro" +version = "0.3.4" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "0ae92a5119aa49cdbcf6b9f893fe4e1d98b04ccbf82ee0584ad948a44a734dea" +dependencies = [ + "proc-macro2", + "quote", + "syn 2.0.119", +] + [[package]] name = "argon2" version = "0.5.3" @@ -214,6 +273,15 @@ version = "1.0.1" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "175571dd1d178ced59193a6fc02dde1b972eb0bc56c892cde9beeceac5bf0f6b" +[[package]] +name = "as-slice" +version = "0.2.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "516b6b4f0e40d50dcda9365d53964ec74560ad4284da2e7fc97122cd83174516" +dependencies = [ + "stable_deref_trait", +] + [[package]] name = "ash" version = "0.38.0+1.3.281" @@ -223,6 +291,51 @@ dependencies = [ "libloading", ] +[[package]] +name = "asn1-rs" +version = "0.7.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "b7f43a50ac4fdca5df8e885c21b835997f0a1cdee65494a6847694a98652d9d8" +dependencies = [ + "asn1-rs-derive", + "asn1-rs-impl", + "displaydoc", + "nom 7.1.3", + "num-traits", + "rusticata-macros", + "thiserror 2.0.21", + "time", +] + +[[package]] +name = "asn1-rs-derive" +version = "0.6.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "3109e49b1e4909e9db6515a30c633684d68cdeaa252f215214cb4fa1a5bfee2c" +dependencies = [ + "proc-macro2", + "quote", + "syn 2.0.119", + "synstructure 0.13.2", +] + +[[package]] +name = "asn1-rs-impl" +version = "0.2.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "7b18050c2cd6fe86c3a76584ef5e0baf286d038cda203eb6223df2cc413565f7" +dependencies = [ + "proc-macro2", + "quote", + "syn 2.0.119", +] + +[[package]] +name = "astc-decode" +version = "0.3.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "2d6058e6202e20b7fa445692ec4b77ca8eb02508091fcd479ea753d9f60e376b" + [[package]] name = "async-broadcast" version = "0.7.2" @@ -366,6 +479,49 @@ version = "1.5.1" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "f2032f911046de80f0a198e0901378627c33f59ea0ac00e363d481118bd70a53" +[[package]] +name = "av-scenechange" +version = "0.14.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "0f321d77c20e19b92c39e7471cf986812cbb46659d2af674adc4331ef3f18394" +dependencies = [ + "aligned", + "anyhow", + "arg_enum_proc_macro", + "arrayvec", + "log", + "num-rational", + "num-traits", + "pastey", + "rayon", + "thiserror 2.0.21", + "v_frame", + "y4m", +] + +[[package]] +name = "av1-grain" +version = "0.2.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "8cfddb07216410377231960af4fcab838eaa12e013417781b78bd95ee22077f8" +dependencies = [ + "anyhow", + "arrayvec", + "log", + "nom 8.0.0", + "num-rational", + "v_frame", +] + +[[package]] +name = "avif-serialize" +version = "0.8.9" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "e7178fe5f7d460b13895ebb9dcb28a3a6216d2df2574a0806cb51b555d297f38" +dependencies = [ + "arrayvec", +] + [[package]] name = "aws-lc-rs" version = "1.18.1" @@ -443,6 +599,9 @@ name = "bit-vec" version = "0.9.1" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "b71798fca2c1fe1086445a7258a4bc81e6e49dcd24c8d0dd9a1e57395b603f51" +dependencies = [ + "serde", +] [[package]] name = "bit-vec" @@ -493,6 +652,15 @@ version = "2.13.2" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "3ded4057c258ba199e2d26386d3af3780957ecaee6c4ef4041c6b4b8b97c0b06" +[[package]] +name = "bitstream-io" +version = "4.10.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "7eff00be299a18769011411c9def0d827e8f2d7bf0c3dbf53633147a8867fd1f" +dependencies = [ + "no_std_io2", +] + [[package]] name = "bitvec" version = "1.1.1" @@ -703,6 +871,12 @@ dependencies = [ "memchr", ] +[[package]] +name = "built" +version = "0.8.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "5c0e531d93d39c34eef561e929e8a7f86d77a5af08aac4f6d6e39976c51858e9" + [[package]] name = "bumpalo" version = "3.20.3" @@ -957,7 +1131,7 @@ dependencies = [ "crmf", "der 0.7.10", "spki 0.7.3", - "x509-cert", + "x509-cert 0.2.5", ] [[package]] @@ -969,7 +1143,19 @@ dependencies = [ "const-oid 0.9.6", "der 0.7.10", "spki 0.7.3", - "x509-cert", + "x509-cert 0.2.5", +] + +[[package]] +name = "cms" +version = "0.3.0-rc.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d089eddc8a5c984d82fcf1427367a6fa0d019d5af04b6e3197e86c8989533b8d" +dependencies = [ + "const-oid 0.10.3", + "der 0.8.3", + "spki 0.8.1", + "x509-cert 0.3.0", ] [[package]] @@ -1187,6 +1373,14 @@ dependencies = [ "serde_json", ] +[[package]] +name = "craft-anicraft-format" +version = "0.1.0" +dependencies = [ + "serde", + "serde_json", +] + [[package]] name = "craft-assuan" version = "0.1.0" @@ -1208,6 +1402,7 @@ version = "0.1.0" dependencies = [ "craft-media", "craft-media-testkit", + "craft-xml", "log", "proptest", "rayon", @@ -1217,10 +1412,15 @@ dependencies = [ "thiserror 2.0.21", ] +[[package]] +name = "craft-ber" +version = "0.1.0" + [[package]] name = "craft-captions" version = "0.1.0" dependencies = [ + "craft-media-testkit", "craft-timeline", "craft-xml", "proptest", @@ -1251,6 +1451,14 @@ dependencies = [ "encoding_rs", ] +[[package]] +name = "craft-control-protocol" +version = "0.1.0" +dependencies = [ + "serde_json", + "subtle", +] + [[package]] name = "craft-crs" version = "0.1.0" @@ -1286,6 +1494,28 @@ dependencies = [ "proptest", ] +[[package]] +name = "craft-deckcraft-format" +version = "0.1.0" +dependencies = [ + "craft-zip", + "serde", + "serde_json", + "thiserror 2.0.21", + "zip", +] + +[[package]] +name = "craft-designcraft-format" +version = "0.1.0" +dependencies = [ + "craft-document-bundle", + "craft-eps", + "serde", + "serde_json", + "thiserror 2.0.21", +] + [[package]] name = "craft-dialogs" version = "0.1.0" @@ -1336,6 +1566,16 @@ dependencies = [ name = "craft-docpath" version = "0.1.0" +[[package]] +name = "craft-document-bundle" +version = "0.1.0" +dependencies = [ + "craft-zip", + "serde_json", + "thiserror 2.0.21", + "zip", +] + [[package]] name = "craft-docx" version = "0.1.0" @@ -1378,12 +1618,21 @@ dependencies = [ "serde_json", ] +[[package]] +name = "craft-dwg" +version = "0.1.0" +dependencies = [ + "acadrust", + "log", +] + [[package]] name = "craft-dxf" version = "0.1.0" dependencies = [ "craft-codepage", "encoding_rs", + "kurbo 0.13.1", "log", "thiserror 2.0.21", ] @@ -1400,6 +1649,12 @@ dependencies = [ [[package]] name = "craft-eps" version = "0.1.0" +dependencies = [ + "kurbo 0.13.1", + "thiserror 2.0.21", + "vectorcraft-doc", + "vectorcraft-eps", +] [[package]] name = "craft-epub" @@ -1430,6 +1685,15 @@ dependencies = [ "thiserror 2.0.21", ] +[[package]] +name = "craft-fcproj-format" +version = "0.1.0" +dependencies = [ + "serde", + "serde_json", + "thiserror 2.0.21", +] + [[package]] name = "craft-fig" version = "0.1.0" @@ -1445,6 +1709,17 @@ dependencies = [ "zip", ] +[[package]] +name = "craft-figcraft-format" +version = "0.1.0" +dependencies = [ + "craft-zip", + "serde", + "serde_json", + "thiserror 2.0.21", + "zip", +] + [[package]] name = "craft-figma-rest" version = "0.1.0" @@ -1454,6 +1729,38 @@ dependencies = [ "thiserror 2.0.21", ] +[[package]] +name = "craft-file-formats" +version = "0.1.0" +dependencies = [ + "craft-eps", + "craft-idml", + "craft-image-codecs", + "craft-kra", + "craft-ora", + "craft-pcraft-bundle", + "craft-raw", + "craft-svg", + "craft-zip", + "serde", + "serde_json", +] + +[[package]] +name = "craft-file-import" +version = "0.1.0" +dependencies = [ + "craft-dxf", + "craft-eps", + "craft-file-formats", + "craft-pdf 0.1.0", + "craft-raster-codecs", + "craft-svg", + "kurbo 0.13.1", + "thiserror 2.0.21", + "vectorcraft-geom", +] + [[package]] name = "craft-flags" version = "0.1.0" @@ -1478,8 +1785,12 @@ name = "craft-font-catalog" version = "0.1.0" dependencies = [ "fontique", + "harfrust 0.13.3", + "kurbo 0.13.1", "skrifa 0.47.0", "thiserror 2.0.21", + "unicode-bidi", + "unicode-segmentation", "windows-registry", "write-fonts 0.54.1", ] @@ -1538,6 +1849,16 @@ dependencies = [ "craft-fonts-build", ] +[[package]] +name = "craft-format" +version = "0.1.0" +dependencies = [ + "craft-zip", + "proptest", + "serde_json", + "sha2 0.10.9", +] + [[package]] name = "craft-gainmap" version = "0.1.0" @@ -1556,9 +1877,26 @@ dependencies = [ name = "craft-gif" version = "0.1.0" dependencies = [ + "gif", "image", ] +[[package]] +name = "craft-gridcraft-format" +version = "0.1.0" +dependencies = [ + "serde", + "serde_json", + "thiserror 2.0.21", +] + +[[package]] +name = "craft-gzip" +version = "0.1.0" +dependencies = [ + "flate2", +] + [[package]] name = "craft-hdr" version = "0.1.0" @@ -1571,6 +1909,7 @@ name = "craft-heif" version = "0.1.0" dependencies = [ "heic-rs", + "proptest", "rayon", ] @@ -1587,6 +1926,10 @@ dependencies = [ "proptest", ] +[[package]] +name = "craft-http-url" +version = "0.1.0" + [[package]] name = "craft-i18n" version = "0.1.0" @@ -1628,6 +1971,47 @@ dependencies = [ "zip", ] +[[package]] +name = "craft-image-codecs" +version = "0.1.0" +dependencies = [ + "avif-serialize", + "craft-heif", + "craft-jpeg", + "craft-media-testkit", + "craft-metadata", + "craft-tiff", + "craft-video", + "crc32fast", + "exr", + "flate2", + "half", + "hayro-jpeg2000", + "image", + "image-webp", + "jpeg-encoder 0.6.1", + "jxl-oxide", + "png 0.18.1", + "proptest", + "rav1e", + "ravif", + "rayon", + "serde_json", + "sha2 0.10.9", + "thiserror 2.0.21", + "tiff", + "zune-core", + "zune-jpeg", + "zune-jpegxl", +] + +[[package]] +name = "craft-indd" +version = "0.1.0" +dependencies = [ + "indd", +] + [[package]] name = "craft-jmap" version = "0.1.0" @@ -1640,11 +2024,31 @@ dependencies = [ name = "craft-jpeg" version = "0.1.0" dependencies = [ + "jpeg-decoder", "jpeg-encoder 0.6.1", + "rayon", "thiserror 2.0.21", "zune-jpeg", ] +[[package]] +name = "craft-jsonrpc" +version = "0.1.0" +dependencies = [ + "serde_json", + "tokio", +] + +[[package]] +name = "craft-jxl" +version = "0.1.0" +dependencies = [ + "craft-cms", + "jixel", + "jxl-oxide", + "thiserror 2.0.21", +] + [[package]] name = "craft-keymap" version = "0.1.0" @@ -1691,6 +2095,7 @@ dependencies = [ name = "craft-lottie" version = "0.1.0" dependencies = [ + "craft-zip", "serde_json", "thiserror 2.0.21", ] @@ -1704,6 +2109,14 @@ dependencies = [ "craft-wordmodel", ] +[[package]] +name = "craft-materialx" +version = "0.1.0" +dependencies = [ + "craft-xml", + "thiserror 2.0.21", +] + [[package]] name = "craft-math" version = "0.1.0" @@ -1714,6 +2127,24 @@ dependencies = [ "serde_json", ] +[[package]] +name = "craft-mcmat" +version = "0.1.0" +dependencies = [ + "craft-zip", + "serde", + "serde_json", + "thiserror 2.0.21", +] + +[[package]] +name = "craft-mcp" +version = "0.1.0" +dependencies = [ + "craft-jsonrpc", + "serde_json", +] + [[package]] name = "craft-media" version = "0.1.0" @@ -1748,6 +2179,7 @@ version = "0.1.0" dependencies = [ "craft-opc", "craft-xml", + "craft-zip", "miniz_oxide 0.9.1", "serde_json", ] @@ -1885,10 +2317,23 @@ dependencies = [ "zip", ] +[[package]] +name = "craft-pcraft-bundle" +version = "0.1.0" +dependencies = [ + "blake3", + "craft-atomic", + "craft-pcraft-format", + "craft-zip", + "ruzstd", + "thiserror 2.0.21", +] + [[package]] name = "craft-pcraft-format" version = "0.1.0" dependencies = [ + "flate2", "serde", "serde_json", "thiserror 2.0.21", @@ -1902,6 +2347,7 @@ dependencies = [ "bytemuck", "chrono", "craft-stream-codecs", + "craft-svg", "flate2", "hayro", "hayro-cmap", @@ -1909,19 +2355,33 @@ dependencies = [ "hayro-syntax", "js-sys", "krilla", + "krilla-svg", "kurbo 0.13.1", "md-5 0.11.0", "proptest", + "roxmltree 0.21.2", "serde_json", "sha2 0.11.1", "stringprep", "thiserror 2.0.21", "ttf-parser", + "usvg 0.47.0", "wasm-bindgen", "wasm-bindgen-test", "zeroize", ] +[[package]] +name = "craft-pdf" +version = "0.1.0" +source = "git+https://github.com/storytold/craft-libs?rev=92d392cff132eaf53e13afff77f38b5c11b73aa0#92d392cff132eaf53e13afff77f38b5c11b73aa0" +dependencies = [ + "hayro-interpret", + "hayro-syntax", + "krilla", + "thiserror 2.0.21", +] + [[package]] name = "craft-pdn" version = "0.1.0" @@ -1978,7 +2438,7 @@ name = "craft-plate-test" version = "0.1.0" dependencies = [ "craft-cms", - "craft-pdf", + "craft-pdf 0.1.0", "hayro-syntax", "serde", "serde_json", @@ -2003,6 +2463,13 @@ dependencies = [ "zbus", ] +[[package]] +name = "craft-plist" +version = "0.1.0" +dependencies = [ + "thiserror 2.0.21", +] + [[package]] name = "craft-plotter" version = "0.1.0" @@ -2039,7 +2506,7 @@ name = "craft-prepress" version = "0.1.0" dependencies = [ "craft-cms", - "craft-pdf", + "craft-pdf 0.1.0", "craft-plate-test", "kurbo 0.13.1", "linesweeper", @@ -2071,6 +2538,28 @@ dependencies = [ "unicode-normalization", ] +[[package]] +name = "craft-proto" +version = "0.1.0" +dependencies = [ + "base64 0.23.1", + "log", + "proptest", + "zeroize", +] + +[[package]] +name = "craft-ps" +version = "0.1.0" +dependencies = [ + "craft-eps", + "kurbo 0.13.1", + "miniz_oxide 0.8.9", + "pdf-writer", + "skrifa 0.47.0", + "weezl 0.1.12", +] + [[package]] name = "craft-psd" version = "0.1.0" @@ -2106,7 +2595,9 @@ dependencies = [ name = "craft-raster-codecs" version = "0.1.0" dependencies = [ - "rayon", + "base64 0.23.1", + "craft-image-codecs", + "image", "thiserror 2.0.21", ] @@ -2145,6 +2636,20 @@ dependencies = [ "proptest", ] +[[package]] +name = "craft-scraft-format" +version = "0.1.0" +dependencies = [ + "craft-audio", + "craft-midi", + "craft-time", + "craft-zip", + "proptest", + "serde", + "serde_json", + "thiserror 2.0.21", +] + [[package]] name = "craft-settings" version = "0.1.0" @@ -2168,6 +2673,10 @@ dependencies = [ name = "craft-sheet" version = "0.1.0" +[[package]] +name = "craft-shx" +version = "0.1.0" + [[package]] name = "craft-silhouette-studio" version = "0.1.0" @@ -2188,6 +2697,59 @@ dependencies = [ "zip", ] +[[package]] +name = "craft-smime" +version = "0.1.0" +dependencies = [ + "aes 0.9.4", + "base64 0.23.1", + "cbc", + "cms 0.3.0-rc.2", + "const-oid 0.10.3", + "craft-ber", + "der 0.8.3", + "des 0.9.0", + "hmac 0.13.0", + "p256 0.14.0", + "p384 0.14.0", + "pbkdf2", + "pkcs1 0.8.0", + "pkcs12", + "proptest", + "rcgen", + "ring", + "rsa 0.10.0-rc.18", + "serde", + "sha1 0.11.0", + "sha2 0.11.1", + "spki 0.8.1", + "thiserror 2.0.21", + "x509-cert 0.3.0", + "zeroize", +] + +[[package]] +name = "craft-stencilcraft-document" +version = "0.1.0" +dependencies = [ + "craft-atomic", + "craft-stencilcraft-format", + "craft-svg", + "kurbo 0.13.1", + "serde", + "serde_json", + "thiserror 2.0.21", +] + +[[package]] +name = "craft-stencilcraft-format" +version = "0.1.0" +dependencies = [ + "serde", + "serde_json", + "thiserror 2.0.21", +] + [[package]] name = "craft-stream-codecs" version = "0.1.0" @@ -2204,12 +2766,18 @@ dependencies = [ name = "craft-svg" version = "0.1.0" dependencies = [ + "craft-image-codecs", "craft-xml", + "data-url", "flate2", "kurbo 0.13.1", + "quick-xml 0.41.0", + "resvg 0.48.1", "roxmltree 0.21.2", "thiserror 2.0.21", "usvg 0.48.1", + "vectorcraft-geom", + "vectorcraft-pathops", ] [[package]] @@ -2220,12 +2788,23 @@ dependencies = [ "kurbo 0.13.1", ] +[[package]] +name = "craft-tcraft" +version = "0.1.0" +dependencies = [ + "craft-zip", + "serde", + "serde_json", + "thiserror 2.0.21", +] + [[package]] name = "craft-text" version = "0.1.0" dependencies = [ "hypher", "icu_segmenter", + "kurbo 0.13.1", "miniz_oxide 0.8.9", "unicode-bidi", "unicode-bidi-mirroring", @@ -2234,6 +2813,16 @@ dependencies = [ "unicode-segmentation", ] +[[package]] +name = "craft-texture" +version = "0.1.0" +dependencies = [ + "bytemuck", + "dds", + "rayon", + "thiserror 2.0.21", +] + [[package]] name = "craft-tga" version = "0.1.0" @@ -2250,10 +2839,21 @@ dependencies = [ "thiserror 2.0.21", ] +[[package]] +name = "craft-time" +version = "0.1.0" +dependencies = [ + "proptest", + "serde", + "serde_json", + "thiserror 2.0.21", +] + [[package]] name = "craft-timeline" version = "0.1.0" dependencies = [ + "craft-time", "craft-xml", "proptest", "serde", @@ -2316,6 +2916,14 @@ dependencies = [ "zip", ] +[[package]] +name = "craft-usd" +version = "0.1.0" +dependencies = [ + "lz4_flex", + "thiserror 2.0.21", +] + [[package]] name = "craft-vcard" version = "0.1.0" @@ -2330,6 +2938,7 @@ name = "craft-vectorcraft-format" version = "0.1.0" dependencies = [ "flate2", + "proptest", "serde", "serde_json", "thiserror 2.0.21", @@ -2349,6 +2958,7 @@ version = "0.1.0" dependencies = [ "craft-media", "craft-media-testkit", + "rav1e", "rayon", "thiserror 2.0.21", ] @@ -2427,6 +3037,15 @@ dependencies = [ "thiserror 2.0.21", ] +[[package]] +name = "craft-xfl" +version = "0.1.0" +dependencies = [ + "craft-xml", + "miniz_oxide 0.8.9", + "thiserror 2.0.21", +] + [[package]] name = "craft-xls" version = "0.1.0" @@ -2455,6 +3074,7 @@ name = "craft-xml" version = "0.1.0" dependencies = [ "quick-xml 0.41.0", + "serde", "thiserror 2.0.21", ] @@ -2496,10 +3116,10 @@ version = "0.2.0" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "36fe21b96d5b87f5de4b5b7202ec41c00110ac817ce6728fe75fb2fe5962ed92" dependencies = [ - "cms", + "cms 0.2.3", "der 0.7.10", "spki 0.7.3", - "x509-cert", + "x509-cert 0.2.5", ] [[package]] @@ -2723,6 +3343,12 @@ dependencies = [ "syn 2.0.119", ] +[[package]] +name = "data-encoding" +version = "2.11.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "4583a4551df46e2792f82ceeac45e850d2e2d5debba0b91f102385cda5b11f06" + [[package]] name = "data-url" version = "0.3.2" @@ -2738,6 +3364,19 @@ dependencies = [ "generic-array", ] +[[package]] +name = "dds" +version = "0.2.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "e917a58e17b27d2f15626c490754ccced7f46592dc5b5f39a3013b89cba5f068" +dependencies = [ + "astc-decode", + "bitflags 2.13.2", + "glam", + "resize", + "zerocopy", +] + [[package]] name = "der" version = "0.7.10" @@ -2745,7 +3384,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "e7c1832837b905bbfb5101e07cc24c8deddf52f93225eee6ead5f4d63d53ddcb" dependencies = [ "const-oid 0.9.6", - "der_derive", + "der_derive 0.7.3", "flagset", "pem-rfc7468 0.7.0", "zeroize", @@ -2758,10 +3397,26 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "6a996a108feb8b427a6e12fcf6af4aa83409b79d72c15337d7741368131ee004" dependencies = [ "const-oid 0.10.3", + "der_derive 0.8.0", + "flagset", "pem-rfc7468 1.0.0", "zeroize", ] +[[package]] +name = "der-parser" +version = "10.0.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "07da5016415d5a3c4dd39b11ed26f915f52fc4e0dc197d87908bc916e51bc1a6" +dependencies = [ + "asn1-rs", + "displaydoc", + "nom 7.1.3", + "num-bigint", + "num-traits", + "rusticata-macros", +] + [[package]] name = "der_derive" version = "0.7.3" @@ -2773,6 +3428,17 @@ dependencies = [ "syn 2.0.119", ] +[[package]] +name = "der_derive" +version = "0.8.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "59600e2c2d636fde9b65e99cc6445ac770c63d3628195ff39932b8d6d7409903" +dependencies = [ + "proc-macro2", + "quote", + "syn 2.0.119", +] + [[package]] name = "deranged" version = "0.5.9" @@ -3134,7 +3800,7 @@ dependencies = [ "bitflags 2.13.2", "emath", "epaint", - "itertools", + "itertools 0.15.0", "log", "nohash-hasher", "profiling", @@ -3191,10 +3857,10 @@ dependencies = [ "ahash", "egui", "enum-map", - "itertools", + "itertools 0.15.0", "log", "profiling", - "resvg", + "resvg 0.45.1", ] [[package]] @@ -3399,7 +4065,7 @@ dependencies = [ "emath", "epaint_default_fonts", "font-types 0.12.7", - "harfrust", + "harfrust 0.12.0", "log", "nohash-hasher", "parking_lot", @@ -3420,6 +4086,26 @@ name = "epaint_default_fonts" version = "0.37.0" source = "git+https://github.com/storytold/egui.git?rev=6d518fc799681571963842f95b55a6b68431b15a#6d518fc799681571963842f95b55a6b68431b15a" +[[package]] +name = "equator" +version = "0.4.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "4711b213838dfee0117e3be6ac926007d7f433d7bbe33595975d4190cb07e6fc" +dependencies = [ + "equator-macro", +] + +[[package]] +name = "equator-macro" +version = "0.4.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "44f23cf4b44bfce11a86ace86f8a73ffdec849c9fd00a386a53d278bd9e81fb3" +dependencies = [ + "proc-macro2", + "quote", + "syn 2.0.119", +] + [[package]] name = "equivalent" version = "1.0.2" @@ -3476,6 +4162,7 @@ dependencies = [ "lebe", "miniz_oxide 0.8.9", "multiversion", + "rayon-core", "smallvec", "zune-inflate", ] @@ -3627,6 +4314,12 @@ version = "1.0.7" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "3f9eec918d3f24069decb9af1554cad7c880e2da24a9afd88aca000531ab82c1" +[[package]] +name = "foldhash" +version = "0.1.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d9c4f5dac5e15c24eb999c26181a6ca40b39fe946cbe4c263c7209467bc83af2" + [[package]] name = "foldhash" version = "0.2.0" @@ -3661,6 +4354,20 @@ dependencies = [ "roxmltree 0.20.0", ] +[[package]] +name = "fontdb" +version = "0.23.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "457e789b3d1202543297a350643cf459f836cade38934e7a4cf6a39e7cde2905" +dependencies = [ + "fontconfig-parser", + "log", + "memmap2", + "slotmap", + "tinyvec", + "ttf-parser", +] + [[package]] name = "fontdb" version = "0.24.0" @@ -3918,6 +4625,12 @@ dependencies = [ "xml-rs", ] +[[package]] +name = "glam" +version = "0.25.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "151665d9be52f9bb40fc7966565d39666f2d1e69233571b71b87791c7e0528b3" + [[package]] name = "glifo" version = "0.1.1" @@ -3925,7 +4638,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "d99fc21d493812643aae86d53b7bbd02f376434a90317e8a790bc209fdd6605e" dependencies = [ "bytemuck", - "foldhash", + "foldhash 0.2.0", "hashbrown 0.17.1", "log", "peniko", @@ -3942,7 +4655,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "282a26c1e23de04bdab3e34a21b6f877a479b96737792516b9b8b8f69b6661be" dependencies = [ "bytemuck", - "foldhash", + "foldhash 0.2.0", "hashbrown 0.17.1", "log", "peniko", @@ -4117,6 +4830,28 @@ dependencies = [ "smallvec", ] +[[package]] +name = "harfrust" +version = "0.13.3" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "948d0741125ba89cd3e1c23e5642415b6ade7e1d29d67ba25fb925b533e989d6" +dependencies = [ + "bitflags 2.13.2", + "bytemuck", + "once_cell", + "read-fonts 0.43.3", + "smallvec", +] + +[[package]] +name = "hashbrown" +version = "0.15.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "9229cfe53dfd69f0609a49f65461bd93001ea1ef889cd5529dd176593f5338a1" +dependencies = [ + "foldhash 0.1.5", +] + [[package]] name = "hashbrown" version = "0.16.1" @@ -4125,7 +4860,7 @@ checksum = "841d1cc9bed7f9236f321df977030373f4a4163ae1a7dbfe1a51a2c1a51d9100" dependencies = [ "allocator-api2", "equivalent", - "foldhash", + "foldhash 0.2.0", ] [[package]] @@ -4134,7 +4869,7 @@ version = "0.17.1" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "ed5909b6e89a2db4456e54cd5f673791d7eca6732202bbf2a9cc504fe2f9b84a" dependencies = [ - "foldhash", + "foldhash 0.2.0", ] [[package]] @@ -4173,7 +4908,7 @@ dependencies = [ "hayro-syntax", "kurbo 0.13.1", "moxcms", - "phf", + "phf 0.13.1", "rustc-hash 2.1.3", "siphasher", "skrifa 0.47.0", @@ -4695,6 +5430,9 @@ dependencies = [ "moxcms", "num-traits", "png 0.18.1", + "qoi", + "ravif", + "rgb", "tiff", "zune-core", "zune-jpeg", @@ -4728,6 +5466,17 @@ version = "0.15.0" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "65b27460c2c92b037f3f94c538ed9a3342f3fdf923606781629ccb35f82d042a" +[[package]] +name = "imgref" +version = "1.12.3" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "6e44b0a4eaa4c82f441d50a963f2d5f05a787240aeee097597033e72accfd22f" + +[[package]] +name = "indd" +version = "0.1.0" +source = "git+https://github.com/storytold/indd-utils?rev=7f3060e5bce78d21ea70e0d5eacb96edce15dfbe#7f3060e5bce78d21ea70e0d5eacb96edce15dfbe" + [[package]] name = "indexmap" version = "2.14.2" @@ -4757,6 +5506,17 @@ dependencies = [ "hybrid-array", ] +[[package]] +name = "interpolate_name" +version = "0.2.4" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "c34819042dc3d3971c46c2190835914dfbe0c3c13f61449b2997f4e9722dfa60" +dependencies = [ + "proc-macro2", + "quote", + "syn 2.0.119", +] + [[package]] name = "io-extras" version = "0.19.0" @@ -4804,6 +5564,15 @@ dependencies = [ "once_cell", ] +[[package]] +name = "itertools" +version = "0.14.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "2b192c782037fadd9cfa75548310488aabdbf3d2da73885b31bd0abd03351285" +dependencies = [ + "either", +] + [[package]] name = "itertools" version = "0.15.0" @@ -4819,6 +5588,12 @@ version = "1.0.18" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "8f42a60cbdf9a97f5d2305f08a87dc4e09308d1276d28c869c684d7777685682" +[[package]] +name = "jixel" +version = "0.3.4" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "278a75e12caacec34be774d62ad05e45fc0b6a52d1f4e9f7fbc0209ecdf794f3" + [[package]] name = "jni" version = "0.22.4" @@ -4887,6 +5662,15 @@ dependencies = [ "libc", ] +[[package]] +name = "jpeg-decoder" +version = "0.3.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "00810f1d8b74be64b13dbf3db89ac67740615d6c891f0e7b6179326533011a07" +dependencies = [ + "rayon", +] + [[package]] name = "jpeg-encoder" version = "0.6.1" @@ -5068,6 +5852,8 @@ version = "1.0.0" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "25f15eb830aa77a7f21148d72e153562a26bfe570139bd4922eab1908dd499d3" dependencies = [ + "rayon", + "rayon-core", "tracing", ] @@ -5183,6 +5969,22 @@ dependencies = [ "zune-jpeg", ] +[[package]] +name = "krilla-svg" +version = "0.8.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "1237d7c37b16ca9fbc2e72dde13a10d321f9a44aceee35c062bc0a43bfc7ce16" +dependencies = [ + "flate2", + "fontdb 0.23.0", + "krilla", + "resvg 0.47.0", + "skrifa 0.42.1", + "smallvec", + "tiny-skia 0.12.0", + "usvg 0.47.0", +] + [[package]] name = "kurbo" version = "0.11.3" @@ -5228,6 +6030,16 @@ version = "0.2.190" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "ce5d3ddc6d3fa000eb1536d85e147bfe31aacaba692ed6a876f95cb7c855be78" +[[package]] +name = "libfuzzer-sys" +version = "0.4.13" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "a9fd2f41a1cba099f79a0b6b6c35656cf7c03351a7bae8ff0f28f25270f929d2" +dependencies = [ + "arbitrary", + "cc", +] + [[package]] name = "libloading" version = "0.8.9" @@ -5315,18 +6127,43 @@ version = "0.4.34" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "f9f8bd3e56ce4dfc153cf470fffbfa98c7620958b312ca5c3a4b8d5181fd13c6" +[[package]] +name = "loop9" +version = "0.1.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "0fae87c125b03c1d2c0150c90365d7d6bcc53fb73a9acaef207d2d065860f062" +dependencies = [ + "imgref", +] + [[package]] name = "lru-slab" version = "0.1.3" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "4050469837a6ff301cd14c1f8f24f88549e6d548f24f64e2148eb0f72cebc51f" +[[package]] +name = "lz4_flex" +version = "0.11.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "373f5eceeeab7925e0c1098212f2fbc4d416adec9d35051a6ab251e824c1854a" + [[package]] name = "maybe-owned" version = "0.3.4" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "4facc753ae494aeb6e3c22f839b158aebd4f9270f55cd3c79906c45476c47ab4" +[[package]] +name = "maybe-rayon" +version = "0.1.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "8ea1f30cedd69f0a2954655f7188c6a834246d2bcf1e315e2ac40c4b24dc9519" +dependencies = [ + "cfg-if", + "rayon", +] + [[package]] name = "md-5" version = "0.10.6" @@ -5387,6 +6224,12 @@ dependencies = [ "walkdir", ] +[[package]] +name = "minimal-lexical" +version = "0.2.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "68354c5c6bd36d73ff3feceb05efa59b6acb7626617f4962be322a825e61f79a" + [[package]] name = "miniz_oxide" version = "0.8.9" @@ -5542,6 +6385,12 @@ dependencies = [ "jni-sys 0.3.1", ] +[[package]] +name = "new_debug_unreachable" +version = "1.0.6" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "650eef8c711430f1a879fdd01d4745a7deea475becfb90269c06775983bbf086" + [[package]] name = "nix" version = "0.31.3" @@ -5554,12 +6403,31 @@ dependencies = [ "libc", ] +[[package]] +name = "no_std_io2" +version = "0.9.4" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "418abd1b6d34fbf6cae440dc874771b0525a604428704c76e48b29a5e67b8003" +dependencies = [ + "memchr", +] + [[package]] name = "nohash-hasher" version = "0.2.0" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "2bf50223579dc7cdcfb3bfcacf7069ff68243f8c363f62ffa99cf000a6b9c451" +[[package]] +name = "nom" +version = "7.1.3" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d273983c5a657a70a3e8f2a01329822f3b8c8172b73826411a55751e404a0a4a" +dependencies = [ + "memchr", + "minimal-lexical", +] + [[package]] name = "nom" version = "8.0.0" @@ -5569,6 +6437,12 @@ dependencies = [ "memchr", ] +[[package]] +name = "noop_proc_macro" +version = "0.3.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "0676bb32a98c1a483ce53e500a81ad9c3d5b3f7c920c28c24e9cb0980d0b5bc8" + [[package]] name = "nu-ansi-term" version = "0.50.3" @@ -5578,6 +6452,16 @@ dependencies = [ "windows-sys 0.61.2", ] +[[package]] +name = "num-bigint" +version = "0.4.8" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "c89e69e7e0f03bea5ef08013795c25018e101932225a656383bd384495ecc367" +dependencies = [ + "num-integer", + "num-traits", +] + [[package]] name = "num-bigint-dig" version = "0.8.6" @@ -5610,6 +6494,17 @@ version = "0.2.2" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "521739c6d2bac4aa25192232afe6841231376b2b26d4d9fae5ecf8ca5772e441" +[[package]] +name = "num-derive" +version = "0.4.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "ed3955f1a9c7c0c15e092f9c887db08b1fc683305fdf6eb6684f22555355e202" +dependencies = [ + "proc-macro2", + "quote", + "syn 2.0.119", +] + [[package]] name = "num-integer" version = "0.1.47" @@ -5629,6 +6524,17 @@ dependencies = [ "num-traits", ] +[[package]] +name = "num-rational" +version = "0.4.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "f83d14da390562dca69fc84082e73e548e1ad308d24accdedd2720017cb37824" +dependencies = [ + "num-bigint", + "num-integer", + "num-traits", +] + [[package]] name = "num-traits" version = "0.2.19" @@ -5992,6 +6898,24 @@ dependencies = [ "subtle", ] +[[package]] +name = "oem_cp" +version = "2.1.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "8c7959d3349f484aec36462f98226983578bf7d33f9b034c32967489c7000e1f" +dependencies = [ + "phf 0.11.3", +] + +[[package]] +name = "oid-registry" +version = "0.8.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "12f40cff3dde1b6087cc5d5f5d4d65712f34016a03ed60e9c08dcc392736b5b7" +dependencies = [ + "asn1-rs", +] + [[package]] name = "once_cell" version = "1.21.4" @@ -6042,6 +6966,15 @@ dependencies = [ "libredox", ] +[[package]] +name = "ordered-channel" +version = "1.2.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "95be4d57809897b5a7539fc15a7dfe0e84141bc3dfaa2e9b1b27caa90acf61ab" +dependencies = [ + "crossbeam-channel", +] + [[package]] name = "ordered-float" version = "5.5.0" @@ -6186,6 +7119,18 @@ dependencies = [ "subtle", ] +[[package]] +name = "paste" +version = "1.0.15" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "57c0d7b74b563b49d38dae00a0c37d4d6de9b432382b2892f0574ddcae73fd0a" + +[[package]] +name = "pastey" +version = "0.1.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "35fb2e5f958ec131621fdd531e9fc186ed768cbe395337403ae56c17a74c68ec" + [[package]] name = "pbkdf2" version = "0.13.0" @@ -6300,7 +7245,7 @@ dependencies = [ "log", "md-5 0.10.6", "memchr", - "nom", + "nom 8.0.0", "num-bigint-dig", "num-traits", "num_enum", @@ -6326,6 +7271,15 @@ dependencies = [ "zeroize", ] +[[package]] +name = "phf" +version = "0.11.3" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "1fd6780a80ae0c52cc120a26a1a42c1ae51b247a253e4e06113d23d2c2edd078" +dependencies = [ + "phf_shared 0.11.3", +] + [[package]] name = "phf" version = "0.13.1" @@ -6333,7 +7287,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "c1562dc717473dbaa4c1f85a36410e03c047b2e7df7f45ee938fbef64ae7fadf" dependencies = [ "phf_macros", - "phf_shared", + "phf_shared 0.13.1", ] [[package]] @@ -6343,7 +7297,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "135ace3a761e564ec88c03a77317a7c6b80bb7f7135ef2544dbe054243b89737" dependencies = [ "fastrand", - "phf_shared", + "phf_shared 0.13.1", ] [[package]] @@ -6353,12 +7307,21 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "812f032b54b1e759ccd5f8b6677695d5268c588701effba24601f6932f8269ef" dependencies = [ "phf_generator", - "phf_shared", + "phf_shared 0.13.1", "proc-macro2", "quote", "syn 2.0.119", ] +[[package]] +name = "phf_shared" +version = "0.11.3" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "67eabc2ef2a60eb7faa00097bd1ffdb5bd28e62bf39990626a582201b7a754e5" +dependencies = [ + "siphasher", +] + [[package]] name = "phf_shared" version = "0.13.1" @@ -6442,6 +7405,21 @@ dependencies = [ "spki 0.8.1", ] +[[package]] +name = "pkcs12" +version = "0.2.0-rc.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "9fd993500a1de8a428a2e6cc9a2ec942c99a801a88d58a24fb986c5ad3b767ca" +dependencies = [ + "cms 0.3.0-rc.2", + "const-oid 0.10.3", + "der 0.8.3", + "digest 0.11.3", + "spki 0.8.1", + "x509-cert 0.3.0", + "zeroize", +] + [[package]] name = "pkcs8" version = "0.10.2" @@ -6712,6 +7690,19 @@ name = "profiling" version = "1.0.18" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "3d595e54a326bc53c1c197b32d295e14b169e3cfeaa8dc82b529f947fba6bcf5" +dependencies = [ + "profiling-procmacros", +] + +[[package]] +name = "profiling-procmacros" +version = "1.0.18" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "4488a4a36b9a4ba6b9334a32a39971f77c1436ec82c38707bce707699cc3bbcb" +dependencies = [ + "quote", + "syn 2.0.119", +] [[package]] name = "proptest" @@ -6739,6 +7730,15 @@ version = "0.1.30" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "d55d956fa96f5ec02be2e13af0e20391a5aa83d6a074e3ad368959d0fab299ea" +[[package]] +name = "qoi" +version = "0.4.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "7f6d64c71eb498fe9eae14ce4ec935c555749aef511cca85b5568910d6e48001" +dependencies = [ + "bytemuck", +] + [[package]] name = "quick-error" version = "1.2.3" @@ -6860,10 +7860,20 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "e058c7de0b26af77780c769414d6257830bb240f3c38477dbc2c16e5f54d6d4c" dependencies = [ "libc", - "rand_chacha", + "rand_chacha 0.3.1", "rand_core 0.6.4", ] +[[package]] +name = "rand" +version = "0.9.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "b9ef1d0d795eb7d84685bca4f72f3649f064e6641543d3a8c415898726a57b41" +dependencies = [ + "rand_chacha 0.9.0", + "rand_core 0.9.5", +] + [[package]] name = "rand" version = "0.10.3" @@ -6885,6 +7895,16 @@ dependencies = [ "rand_core 0.6.4", ] +[[package]] +name = "rand_chacha" +version = "0.9.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d3022b5f1df60f26e1ffddd6c66e8aa15de382ae63b3a0c1bfc0e4d3e3f325cb" +dependencies = [ + "ppv-lite86", + "rand_core 0.9.5", +] + [[package]] name = "rand_core" version = "0.6.4" @@ -6894,6 +7914,15 @@ dependencies = [ "getrandom 0.2.17", ] +[[package]] +name = "rand_core" +version = "0.9.5" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "76afc826de14238e6e8c374ddcc1fa19e374fd8dd986b0d2af0d02377261d83c" +dependencies = [ + "getrandom 0.3.4", +] + [[package]] name = "rand_core" version = "0.10.1" @@ -6924,6 +7953,56 @@ version = "0.1.5" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "ca45419789ae5a7899559e9512e58ca889e41f04f1f2445e9f4b290ceccd1d08" +[[package]] +name = "rav1e" +version = "0.8.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "43b6dd56e85d9483277cde964fd1bdb0428de4fec5ebba7540995639a21cb32b" +dependencies = [ + "aligned-vec", + "arbitrary", + "arg_enum_proc_macro", + "arrayvec", + "av-scenechange", + "av1-grain", + "bitstream-io", + "built", + "cfg-if", + "interpolate_name", + "itertools 0.14.0", + "libc", + "libfuzzer-sys", + "log", + "maybe-rayon", + "new_debug_unreachable", + "noop_proc_macro", + "num-derive", + "num-traits", + "paste", + "profiling", + "rand 0.9.5", + "rand_chacha 0.9.0", + "simd_helpers", + "thiserror 2.0.21", + "v_frame", + "wasm-bindgen", +] + +[[package]] +name = "ravif" +version = "0.13.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "e52310197d971b0f5be7fe6b57530dcd27beb35c1b013f29d66c1ad73fbbcc45" +dependencies = [ + "avif-serialize", + "imgref", + "loop9", + "quick-error 2.0.1", + "rav1e", + "rayon", + "rgb", +] + [[package]] name = "raw-window-handle" version = "0.6.2" @@ -6971,6 +8050,19 @@ dependencies = [ "cipher 0.5.2", ] +[[package]] +name = "rcgen" +version = "0.14.10" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "8774e05a7d0de114588e6a28fe7e71694b82614ed569d86d8b389dfbc98b8ad8" +dependencies = [ + "ring", + "rustls-pki-types", + "time", + "x509-parser", + "yasna", +] + [[package]] name = "read-fonts" version = "0.39.2" @@ -6992,6 +8084,17 @@ dependencies = [ "once_cell", ] +[[package]] +name = "read-fonts" +version = "0.43.3" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "005c8acf251756c478b0bf402885bfd88a1476020c4c7e6060edc1aa68da38ea" +dependencies = [ + "bytemuck", + "font-types 0.12.7", + "once_cell", +] + [[package]] name = "read-fonts" version = "0.44.0" @@ -7010,7 +8113,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "4edc8db0d972de88e53a085b1f01fb3fbdae7bb0cb7a0ce2b2f8285a412fe3ef" dependencies = [ "bytemuck", - "foldhash", + "foldhash 0.2.0", "font-types 0.12.7", "once_cell", "yoke", @@ -7120,6 +8223,15 @@ dependencies = [ "web-sys", ] +[[package]] +name = "resize" +version = "0.8.9" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "71725ecd5e0197b54fe859055b108688472ab6a358f8fbe5cee4a556b1b5bfea" +dependencies = [ + "rgb", +] + [[package]] name = "resvg" version = "0.45.1" @@ -7130,10 +8242,45 @@ dependencies = [ "pico-args", "rgb", "svgtypes 0.15.3", - "tiny-skia", + "tiny-skia 0.11.4", "usvg 0.45.1", ] +[[package]] +name = "resvg" +version = "0.47.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "9be183ad6a216aa96f33e4c8033b0988b8b3ea6fd2359d19af5bac4643fd8e81" +dependencies = [ + "gif", + "image-webp", + "log", + "pico-args", + "rgb", + "svgtypes 0.16.2", + "tiny-skia 0.12.0", + "usvg 0.47.0", + "zune-jpeg", +] + +[[package]] +name = "resvg" +version = "0.48.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "67e3803f97b999e80cbf7c6ecdd07a8102204d92e1633cf48783720c521196bd" +dependencies = [ + "bytemuck", + "gif", + "image-webp", + "log", + "pico-args", + "rgb", + "svgtypes 0.16.2", + "tiny-skia 0.12.0", + "usvg 0.48.1", + "zune-jpeg", +] + [[package]] name = "rfc6979" version = "0.4.0" @@ -7292,6 +8439,15 @@ dependencies = [ "semver", ] +[[package]] +name = "rusticata-macros" +version = "4.1.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "faf0c4a6ece9950b9abdb62b1cfcf2a68b3b67a10ba445b3bb85be2a293d0632" +dependencies = [ + "nom 7.1.3", +] + [[package]] name = "rustix" version = "0.38.44" @@ -7395,9 +8551,9 @@ checksum = "eec689c0bc40ff2458a5977b6619cb718087084a18e02a131c599b62d05e1a5f" [[package]] name = "rustls-webpki" -version = "0.103.15" +version = "0.103.16" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "f3c3cf1d8b1e7d4927e2d154c3fcb02979afb9939629c62cd9048d4f07b60ac2" +checksum = "4443b1b6d39d7c663139942cc4558262562d2c84123bb6282ef177009ca8b281" dependencies = [ "aws-lc-rs", "ring", @@ -7822,7 +8978,7 @@ dependencies = [ "spki 0.7.3", "thiserror 2.0.21", "tracing", - "x509-cert", + "x509-cert 0.2.5", ] [[package]] @@ -7871,7 +9027,7 @@ dependencies = [ "sigstore-crypto", "sigstore-types", "thiserror 2.0.21", - "x509-cert", + "x509-cert 0.2.5", ] [[package]] @@ -7884,7 +9040,7 @@ dependencies = [ "base64 0.22.1", "chrono", "cmpv2", - "cms", + "cms 0.2.3", "const-oid 0.9.6", "der 0.7.10", "hex", @@ -7896,7 +9052,7 @@ dependencies = [ "sigstore-types", "thiserror 2.0.21", "tracing", - "x509-cert", + "x509-cert 0.2.5", "x509-tsp", ] @@ -7923,7 +9079,7 @@ checksum = "d30aff69036e4579b416956e9ba68cec45b793116f78f10842d04c7f8b924bab" dependencies = [ "base64 0.22.1", "chrono", - "cms", + "cms 0.2.3", "const-oid 0.9.6", "hex", "pem", @@ -7942,7 +9098,7 @@ dependencies = [ "thiserror 2.0.21", "tls_codec", "tracing", - "x509-cert", + "x509-cert 0.2.5", ] [[package]] @@ -7961,6 +9117,15 @@ dependencies = [ "simdutf8", ] +[[package]] +name = "simd_helpers" +version = "0.1.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "95890f873bec569a0362c235787f3aca6e1e887302ba4840839bcc6459c42da6" +dependencies = [ + "quote", +] + [[package]] name = "simdutf8" version = "0.1.5" @@ -8160,6 +9325,16 @@ dependencies = [ "float-cmp", ] +[[package]] +name = "string-interner" +version = "0.19.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "23de088478b31c349c9ba67816fa55d9355232d63c3afea8bf513e31f0f1d2c0" +dependencies = [ + "hashbrown 0.15.5", + "serde", +] + [[package]] name = "stringprep" version = "0.1.5" @@ -8474,6 +9649,17 @@ dependencies = [ "futures-core", ] +[[package]] +name = "synstructure" +version = "0.13.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "728a70f3dbaf5bab7f0c4b1ac8d7ae5ea60a4b5549c8a5914361c99147a709d2" +dependencies = [ + "proc-macro2", + "quote", + "syn 2.0.119", +] + [[package]] name = "synstructure" version = "0.14.0" @@ -8603,6 +9789,15 @@ dependencies = [ "syn 3.0.6", ] +[[package]] +name = "thread_local" +version = "1.1.10" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "1ad99c4c6d32803332c548b1af0540b357b3f5fc0be8f6c6bfe8b2e6ae784070" +dependencies = [ + "cfg-if", +] + [[package]] name = "tiff" version = "0.11.3" @@ -8662,6 +9857,21 @@ dependencies = [ "tiny-skia-path 0.11.4", ] +[[package]] +name = "tiny-skia" +version = "0.12.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "47ffee5eaaf5527f630fb0e356b90ebdec84d5d18d937c5e440350f88c5a91ea" +dependencies = [ + "arrayref", + "arrayvec", + "bytemuck", + "cfg-if", + "log", + "png 0.18.1", + "tiny-skia-path 0.12.0", +] + [[package]] name = "tiny-skia-path" version = "0.11.4" @@ -8733,9 +9943,21 @@ dependencies = [ "mio", "pin-project-lite", "socket2", + "tokio-macros", "windows-sys 0.61.2", ] +[[package]] +name = "tokio-macros" +version = "2.7.2" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "78773a2a397f451582ce068015985c33193cf6dea8b74d2a639fe457b2f07b0e" +dependencies = [ + "proc-macro2", + "quote", + "syn 3.0.6", +] + [[package]] name = "tokio-rustls" version = "0.26.6" @@ -9119,6 +10341,34 @@ dependencies = [ "xmlwriter", ] +[[package]] +name = "usvg" +version = "0.47.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d46cf96c5f498d36b7a9693bc6a7075c0bb9303189d61b2249b0dc3d309c07de" +dependencies = [ + "base64 0.22.1", + "data-url", + "flate2", + "fontdb 0.23.0", + "imagesize 0.14.0", + "kurbo 0.13.1", + "log", + "pico-args", + "roxmltree 0.21.2", + "rustybuzz", + "simplecss", + "siphasher", + "strict-num", + "svgtypes 0.16.2", + "tiny-skia-path 0.12.0", + "ttf-parser", + "unicode-bidi", + "unicode-script", + "unicode-vo", + "xmlwriter", +] + [[package]] name = "usvg" version = "0.48.1" @@ -9127,8 +10377,8 @@ checksum = "977d0a4abdef933f424a99fe09f95576e089b90aebc6f016a3bc813762493e91" dependencies = [ "base64 0.23.1", "data-url", - "fontdb", - "harfrust", + "fontdb 0.24.0", + "harfrust 0.12.0", "imagesize 0.15.0", "kurbo 0.13.1", "log", @@ -9175,6 +10425,232 @@ dependencies = [ "wasm-bindgen", ] +[[package]] +name = "v_frame" +version = "0.3.9" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "666b7727c8875d6ab5db9533418d7c764233ac9c0cff1d469aec8fa127597be2" +dependencies = [ + "aligned-vec", + "num-traits", + "wasm-bindgen", +] + +[[package]] +name = "vectorcraft-brush" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "kurbo 0.13.1", + "serde", + "serde_json", + "vectorcraft-color", + "vectorcraft-doc", + "vectorcraft-geom", +] + +[[package]] +name = "vectorcraft-color" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "kurbo 0.13.1", + "moxcms", + "serde", + "serde_json", + "thiserror 2.0.21", +] + +[[package]] +name = "vectorcraft-doc" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "chrono", + "hayro-ccitt", + "image", + "kurbo 0.13.1", + "serde", + "serde_json", + "thiserror 2.0.21", + "tiff", + "vectorcraft-color", + "vectorcraft-geom", + "zune-jpeg", +] + +[[package]] +name = "vectorcraft-effects" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "image", + "kurbo 0.13.1", + "serde_json", + "vectorcraft-doc", + "vectorcraft-geom", + "vectorcraft-pathops", + "vectorcraft-plugins", + "vectorcraft-text", + "vectorcraft-three-d", +] + +[[package]] +name = "vectorcraft-eps" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "flate2", + "hayro-ccitt", + "image", + "kurbo 0.13.1", + "memchr", + "ruzstd", + "serde_json", + "vectorcraft-brush", + "vectorcraft-color", + "vectorcraft-doc", + "vectorcraft-effects", + "vectorcraft-geom", + "vectorcraft-metafile", + "vectorcraft-pdf", + "vectorcraft-render", + "vectorcraft-text", +] + +[[package]] +name = "vectorcraft-geom" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "kurbo 0.13.1", + "serde", + "thiserror 2.0.21", +] + +[[package]] +name = "vectorcraft-metafile" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "image", + "kurbo 0.13.1", + "vectorcraft-brush", + "vectorcraft-color", + "vectorcraft-doc", + "vectorcraft-effects", + "vectorcraft-geom", + "vectorcraft-text", +] + +[[package]] +name = "vectorcraft-pathops" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "kurbo 0.13.1", + "linesweeper", + "thiserror 2.0.21", + "vectorcraft-geom", +] + +[[package]] +name = "vectorcraft-pdf" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "aes 0.8.4", + "chrono", + "craft-pdf 0.1.0 (git+https://github.com/storytold/craft-libs?rev=92d392cff132eaf53e13afff77f38b5c11b73aa0)", + "flate2", + "getrandom 0.4.3", + "image", + "jpeg-encoder 0.6.1", + "kurbo 0.13.1", + "log", + "md-5 0.10.6", + "roxmltree 0.21.2", + "serde", + "sha2 0.10.9", + "skrifa 0.47.0", + "smallvec", + "thiserror 2.0.21", + "vectorcraft-brush", + "vectorcraft-color", + "vectorcraft-doc", + "vectorcraft-effects", + "vectorcraft-geom", + "vectorcraft-text", + "weezl 0.2.1", +] + +[[package]] +name = "vectorcraft-plugins" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "serde", + "serde_json", + "thiserror 2.0.21", + "vectorcraft-color", + "vectorcraft-doc", + "vectorcraft-geom", + "wasmi", +] + +[[package]] +name = "vectorcraft-render" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "flate2", + "gif", + "image", + "jpeg-encoder 0.6.1", + "log", + "png 0.18.1", + "vectorcraft-brush", + "vectorcraft-color", + "vectorcraft-doc", + "vectorcraft-effects", + "vectorcraft-geom", + "vectorcraft-text", + "vello_common 0.2.0", + "vello_cpu 0.2.0", + "weezl 0.2.1", + "winsafe", +] + +[[package]] +name = "vectorcraft-text" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "harfrust 0.13.3", + "icu_segmenter", + "kurbo 0.13.1", + "log", + "skrifa 0.47.0", + "subsetter", + "unicode-bidi", + "unicode-general-category", + "unicode-script", + "unicode-segmentation", + "vectorcraft-doc", + "vectorcraft-geom", + "vello_cpu 0.2.0", + "windows-registry", +] + +[[package]] +name = "vectorcraft-three-d" +version = "0.18.0" +source = "git+https://github.com/storytold/vectorcraft.git?rev=6b3a6b8788ebcb420d69449199cbdf323888d27e#6b3a6b8788ebcb420d69449199cbdf323888d27e" +dependencies = [ + "vectorcraft-geom", + "vectorcraft-pathops", +] + [[package]] name = "vello_common" version = "0.0.8" @@ -9245,9 +10721,13 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "ace506ca414548966fcf3c283be94eb1cad8254de488791ee698bf438b0890a9" dependencies = [ "bytemuck", + "crossbeam-channel", "glifo 0.3.0", "hashbrown 0.17.1", + "ordered-channel", "png 0.18.1", + "rayon", + "thread_local", "vello_common 0.2.0", ] @@ -9405,6 +10885,56 @@ version = "0.2.129" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "4f692aa943ccd88363733b77063f32cfed5bc6cbea8e6e8b251b302f881606fe" +[[package]] +name = "wasmi" +version = "2.0.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "78693fcdd618e0fc34af59c6b8efa9ac5d58c68df940beff4bedddb6acfe7c27" +dependencies = [ + "spin", + "wasmi_collections", + "wasmi_core", + "wasmi_ir", + "wasmparser", +] + +[[package]] +name = "wasmi_collections" +version = "2.0.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "8a8be2aa467cf2d29e96ff759472c36eeb44a3c81c67fc9cb76c9a24c519c557" +dependencies = [ + "string-interner", +] + +[[package]] +name = "wasmi_core" +version = "2.0.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "69372d5fda3ea3d1e0aa6603c7888110e0187e88ea17cd8fc2e2df0a0e1f37fa" +dependencies = [ + "libm", +] + +[[package]] +name = "wasmi_ir" +version = "2.0.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "8f17b774caa13c618c7244f1ee51fe23c5e7b8538a471fa46d9949779758aed6" +dependencies = [ + "wasmi_core", +] + +[[package]] +name = "wasmparser" +version = "0.228.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "4abf1132c1fdf747d56bbc1bb52152400c70f336870f968b85e89ea422198ae3" +dependencies = [ + "bitflags 2.13.2", + "indexmap", +] + [[package]] name = "wayland-backend" version = "0.3.17" @@ -9962,7 +11492,7 @@ dependencies = [ [[package]] name = "winprint" version = "0.2.1" -source = "git+https://github.com/dotCooCoo/winprint.rs?rev=72951d7455ebcab569fbd9386c513d3d9f53fb02#72951d7455ebcab569fbd9386c513d3d9f53fb02" +source = "git+https://github.com/storytold/winprint.rs?rev=8ae097ef941b432b0e7671b1ec3a506800e65e57#8ae097ef941b432b0e7671b1ec3a506800e65e57" dependencies = [ "flate2", "fmt-derive", @@ -10133,6 +11663,35 @@ dependencies = [ "tls_codec", ] +[[package]] +name = "x509-cert" +version = "0.3.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "105ef4642d9cb137ef83d623d0e4bf08b8adf69e9918ca904a174adb6d3d038b" +dependencies = [ + "const-oid 0.10.3", + "der 0.8.3", + "spki 0.8.1", +] + +[[package]] +name = "x509-parser" +version = "0.18.1" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "d43b0f71ce057da06bc0851b23ee24f3f86190b07203dd8f567d0b706a185202" +dependencies = [ + "asn1-rs", + "data-encoding", + "der-parser", + "lazy_static", + "nom 7.1.3", + "oid-registry", + "ring", + "rusticata-macros", + "thiserror 2.0.21", + "time", +] + [[package]] name = "x509-tsp" version = "0.1.0" @@ -10140,7 +11699,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "f5ceece934a21607055b7ac5c25adb56a2ff559804b10705dc674d1d838c15e1" dependencies = [ "cmpv2", - "cms", + "cms 0.2.3", "der 0.7.10", ] @@ -10197,6 +11756,22 @@ version = "0.3.3" source = "registry+https://github.com/rust-lang/crates.io-index" checksum = "9440ea3e5aeabb0ac63af70daf835274065238cdd0cec83418f417eae38bacee" +[[package]] +name = "y4m" +version = "0.8.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "7a5a4b21e1a62b67a2970e6831bc091d7b87e119e7f9791aef9702e3bef04448" + +[[package]] +name = "yasna" +version = "0.6.0" +source = "registry+https://github.com/rust-lang/crates.io-index" +checksum = "b5f6765e852b9b4dc8e2a76843e4d64d1cea8e79bcde0b6901aea8e7c7f08282" +dependencies = [ + "bit-vec 0.9.1", + "time", +] + [[package]] name = "yeslogic-fontconfig-sys" version = "6.0.1" @@ -10228,7 +11803,7 @@ dependencies = [ "proc-macro2", "quote", "syn 3.0.6", - "synstructure", + "synstructure 0.14.0", ] [[package]] @@ -10339,7 +11914,7 @@ dependencies = [ "proc-macro2", "quote", "syn 3.0.6", - "synstructure", + "synstructure 0.14.0", ] [[package]] @@ -10401,9 +11976,9 @@ dependencies = [ [[package]] name = "zip" -version = "9.1.0" +version = "9.1.3" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "2886b00003a72c4ccee3f42cfc34908d33b2508a74d827dc2a4244706475c704" +checksum = "1156b39d142a34b999c2310a50455e0cb3760bf7efad0122f796d813bee26ccb" dependencies = [ "crc32fast", "flate2", diff --git a/Cargo.toml b/Cargo.toml index 629bab64c..bd08fa725 100644 --- a/Cargo.toml +++ b/Cargo.toml @@ -11,14 +11,29 @@ rust-version = "1.95" repository = "https://github.com/storytold/craft-libs" [workspace.dependencies] +craft-shx = { path = "crates/shx" } +craft-http-url = { path = "crates/http-url" } +craft-usd = { path = "crates/usd" } +craft-smime = { path = "crates/smime" } +craft-ber = { path = "crates/ber" } +craft-proto = { path = "crates/proto" } +acadrust = "=0.6.6" +log = "0.4" # internal: add each crate here as `craft- = { path = "crates/" }` +craft-plist = { path = "crates/plist" } craft-platform = { path = "crates/platform" } craft-aep = { path = "crates/aep" } craft-affinity = { path = "crates/affinity" } craft-ai-core = { path = "crates/ai-core" } +craft-anicraft-format = { path = "crates/anicraft-format" } craft-pdf = { path = "crates/pdf", version = "=0.1.0" } +craft-ps = { path = "crates/ps", version = "=0.1.0" } craft-pdn = { path = "crates/pdn" } craft-plate-test = { path = "crates/plate-test" } +pdf-writer = "0.15" +miniz_oxide = "0.8" +weezl = "0.1" +skrifa = "0.47" krilla = { version = "=0.8.2", default-features = false, features = ["raster-images"] } hayro = { version = "=0.7.1", default-features = false } hayro-interpret = { version = "=0.7.0", default-features = false, features = ["embed-fonts", "embed-cmaps"] } @@ -26,6 +41,8 @@ hayro-syntax = { version = "=0.7.2", default-features = false, features = ["std" bytemuck = "1" kurbo = "0.13" thiserror = "2" +craft-gzip = { path = "crates/gzip" } +craft-gridcraft-format = { path = "crates/gridcraft-format" } craft-aaf = { path = "crates/aaf" } craft-atomic = { path = "crates/atomic" } craft-captions = { path = "crates/captions" } @@ -39,12 +56,24 @@ craft-dav = { path = "crates/dav" } craft-dialogs = { path = "crates/dialogs" } craft-doc-binary = { path = "crates/doc-binary" } craft-docx = { path = "crates/docx" } +craft-document-bundle = { path = "crates/document-bundle" } +craft-mcmat = { path = "crates/mcmat" } +craft-deckcraft-format = { path = "crates/deckcraft-format" } +craft-designcraft-format = { path = "crates/designcraft-format" } craft-dxf = { path = "crates/dxf" } craft-drawingml = { path = "crates/drawingml" } craft-eps = { path = "crates/eps" } +craft-file-formats = { path = "crates/file-formats" } +craft-file-import = { path = "crates/file-import" } +vectorcraft-doc = { git = "https://github.com/storytold/vectorcraft.git", rev = "6b3a6b8788ebcb420d69449199cbdf323888d27e" } +vectorcraft-eps = { git = "https://github.com/storytold/vectorcraft.git", rev = "6b3a6b8788ebcb420d69449199cbdf323888d27e" } +vectorcraft-geom = { git = "https://github.com/storytold/vectorcraft.git", rev = "6b3a6b8788ebcb420d69449199cbdf323888d27e" } +vectorcraft-pathops = { git = "https://github.com/storytold/vectorcraft.git", rev = "6b3a6b8788ebcb420d69449199cbdf323888d27e" } craft-doc = { path = "crates/doc" } craft-docpath = { path = "crates/docpath" } craft-dpx = { path = "crates/dpx" } +craft-dwg = { path = "crates/dwg" } +craft-fcproj-format = { path = "crates/fcproj-format" } craft-ecproj-format = { path = "crates/ecproj-format" } craft-figma-rest = { path = "crates/figma-rest" } craft-exr = { path = "crates/exr" } @@ -64,19 +93,25 @@ craft-fig = { path = "crates/fig" } craft-fonts-build = { path = "crates/fonts-build" } craft-i18n = { path = "crates/i18n" } craft-idml = { path = "crates/idml" } +craft-indd = { path = "crates/indd" } craft-jmap = { path = "crates/jmap" } craft-ical = { path = "crates/ical" } craft-jpeg = { path = "crates/jpeg" } +craft-control-protocol = { path = "crates/control-protocol" } +craft-jsonrpc = { path = "crates/jsonrpc" } +craft-mcp = { path = "crates/mcp" } +craft-jxl = { path = "crates/jxl" } craft-kiwi = { path = "crates/kiwi" } craft-keymap = { path = "crates/keymap" } craft-assuan = { path = "crates/assuan" } craft-kra = { path = "crates/kra" } craft-audio = { path = "crates/audio" } -# No default features, so craft-audio's `encode` builds craft-media without `isobmff`. craft-latex = { path = "crates/latex" } craft-markdown = { path = "crates/markdown" } craft-math = { path = "crates/math" } +# No default features: craft-audio turns on the craft-media modules each of its features needs. craft-media = { path = "crates/media", default-features = false } +craft-materialx = { path = "crates/materialx" } craft-media-testkit = { path = "crates/media-testkit" } craft-midi = { path = "crates/midi" } craft-sf2 = { path = "crates/sf2" } @@ -94,12 +129,14 @@ craft-ora = { path = "crates/ora" } craft-pptx = { path = "crates/pptx" } craft-penpot = { path = "crates/penpot" } craft-pcraft-format = { path = "crates/pcraft-format" } +craft-pcraft-bundle = { path = "crates/pcraft-bundle" } craft-pkcs7 = { path = "crates/pkcs7" } craft-plotter = { path = "crates/plotter" } craft-prepress = { path = "crates/prepress" } craft-print = { path = "crates/print" } craft-proof = { path = "crates/proof" } craft-raster-codecs = { path = "crates/raster-codecs" } +craft-image-codecs = { path = "crates/image-codecs" } craft-psd = { path = "crates/psd" } craft-pst = { path = "crates/pst" } craft-quit = { path = "crates/quit" } @@ -108,11 +145,17 @@ craft-raw = { path = "crates/raw" } craft-sheet = { path = "crates/sheet" } craft-sketch = { path = "crates/sketch" } craft-settings = { path = "crates/settings" } +craft-scraft-format = { path = "crates/scraft-format" } +craft-time = { path = "crates/time" } craft-silhouette-studio = { path = "crates/silhouette-studio" } craft-svg = { path = "crates/svg" } +craft-stencilcraft-format = { path = "crates/stencilcraft-format" } +craft-stencilcraft-document = { path = "crates/stencilcraft-document" } craft-svg-paths = { path = "crates/svg-paths" } craft-stream-codecs = { path = "crates/stream-codecs" } craft-text = { path = "crates/text" } +craft-tcraft = { path = "crates/tcraft" } +craft-texture = { path = "crates/texture" } craft-tga = { path = "crates/tga" } craft-timeline = { path = "crates/timeline" } craft-tiff = { path = "crates/tiff" } @@ -123,8 +166,10 @@ craft-vcard = { path = "crates/vcard" } craft-vectorcraft-format = { path = "crates/vectorcraft-format" } craft-vectordrawable = { path = "crates/vectordrawable" } craft-xls = { path = "crates/xls" } +craft-xfl = { path = "crates/xfl" } craft-wordmodel = { path = "crates/wordmodel" } craft-xlsx = { path = "crates/xlsx" } +indd = { git = "https://github.com/storytold/indd-utils", rev = "7f3060e5bce78d21ea70e0d5eacb96edce15dfbe" } craft-xcf = { path = "crates/xcf" } craft-xml = { path = "crates/xml" } craft-zip = { path = "crates/zip" } @@ -138,6 +183,8 @@ craft-web-host = { path = "crates/web-host" } craft-webp = { path = "crates/webp" } egui = { version = "0.37", default-features = false } md5 = { package = "md-5", version = "0.10", default-features = false } +jxl-oxide = { version = "0.12", default-features = false } +jixel = "=0.3.4" rayon = "1" serde = { version = "1", features = ["derive"] } serde_json = "1" diff --git a/NOTICE b/NOTICE index c6583cc70..c3cdbabc8 100644 --- a/NOTICE +++ b/NOTICE @@ -5,6 +5,17 @@ Licensed under the MIT License or the Apache License, Version 2.0, at your optio Third-party material +Optional DWG bridge + +`craft-dwg` uses `acadrust` 0.6.6 unmodified as a separate dependency for DWG reading and writing. +It is licensed under the Mozilla Public License 2.0. The license text and upstream third-party +notices are retained in `licenses/dwg/`; the crate is available from crates.io and +https://github.com/hakanaktt/acadrust. + +The byte-level bridge source was moved from CADCraft `crates/dwg` at revision +`664837df8925f932e90a7ae37c58d2f54e7a1320`. The bridge itself is MIT OR Apache-2.0; its only +MPL-2.0 component is the unmodified `acadrust` dependency. + This product bundles third-party material under its own licenses, listed below. The per-file list (author, source, license) is in ATTRIBUTION.md. @@ -89,6 +100,14 @@ The optional PDF writer bundles four CC0 ICC profiles by Clinton Ingram. See ATT Source provenance +crates/image-codecs (craft-image-codecs) is moved from PhotoCraft's standalone `photocraft-codecs` +crate, including its bounded decoders, encoders, format registry, image model, metadata handling, +orientation, fidelity warnings, and tests, from revision 5000240c6: +https://github.com/storytold/photocraft +Copyright (c) 2026 ArtCraft Team and the PhotoCraft contributors. The format implementations are +clean-room work based on their public specifications; dependency licences and X.Org's `rgb.txt` +attribution are listed in ATTRIBUTION.md. + crates/raw (craft-raw) is moved from LightCraft, with behaviour ported from two other readers: - LightCraft, crates/raw (src/lib.rs, dng.rs, dngwrite.rs, tiffraw.rs, unpack.rs, ljpeg.rs, jxl.rs, opcodes.rs, profile.rs, gaintable.rs, semantic.rs, preview.rs, vendor/crw.rs, vendor/hasselblad.rs, @@ -235,9 +254,10 @@ applications retain their own document-font selection and licensing policy. linesweeper linesweeper 0.5.0 (https://crates.io/crates/linesweeper/0.5.0), MIT OR Apache-2.0, is stored in -vendor/linesweeper with its upstream license files and one local patch. Its manifest names Joe Neeman -as author, and its LICENSE-MIT reads "Copyright (c) 2018 Raph Levien". See -vendor/linesweeper/README.craft.md and vendor/linesweeper/SOURCE.json for provenance, the local patch +vendor/linesweeper with its upstream license files and local patches, to src/sweep/sweep_line.rs and +docs/sweep.typ. Its manifest names Joe Neeman as author, and its LICENSE-MIT reads "Copyright (c) 2018 +Raph Levien". See vendor/linesweeper/README.craft.md and vendor/linesweeper/SOURCE.json for +provenance, the local patches and the files left out. Platform lifecycle source provenance @@ -276,3 +296,126 @@ implementation, https://github.com/macosforge/alac at c38887c5c5e64a4b31108733bd (LICENSE-APACHE). The files are modified: translated to Rust, with bounds checks on hostile input, wrapping integer arithmetic, the unrolled predictor and per-width matrix routines folded into one, and the average bit rate filled in. Those files are Apache-2.0 only. + +JPEG encoding dependency: jpeg-encoder 0.6.1 + +Source: https://crates.io/crates/jpeg-encoder/0.6.1 +Declared license: (MIT OR Apache-2.0) AND IJG. +This software is based in part on the work of the Independent JPEG Group. + +The MIT license and unaltered upstream IJG and SIMD notices follow. The Apache-2.0 +alternative is also retained in licenses/jpeg-encoder-0.6.1 at the repository root. +The dependency is used unmodified; no encoder implementation is vendored here. + +LICENSE-MIT + +Copyright (c) 2021 Volker Ströbel + +Permission is hereby granted, free of charge, to any +person obtaining a copy of this software and associated +documentation files (the "Software"), to deal in the +Software without restriction, including without +limitation the rights to use, copy, modify, merge, +publish, distribute, sublicense, and/or sell copies of +the Software, and to permit persons to whom the Software +is furnished to do so, subject to the following +conditions: + +The above copyright notice and this permission notice +shall be included in all copies or substantial portions +of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF +ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED +TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A +PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT +SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY +CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION +OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR +IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER +DEALINGS IN THE SOFTWARE. +Original src/fdct.rs opening notice + +/* + * Ported from mozjpeg to rust + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2015, 2020, D. R. Commander. + * + * Conditions of distribution and use: + * In plain English: + * + * 1. We don't promise that this software works. (But if you find any bugs, + * please let us know!) + * 2. You can use this software for whatever you want. You don't have to pay us. + * 3. You may not pretend that you wrote this software. If you use it in a + * program, you must acknowledge somewhere in your documentation that + * you've used the IJG code. + * + * In legalese: + * + * The authors make NO WARRANTY or representation, either express or implied, + * with respect to this software, its quality, accuracy, merchantability, or + * fitness for a particular purpose. This software is provided "AS IS", and you, + * its user, assume the entire risk as to its quality and accuracy. + * + * This software is copyright (C) 1991-2020, Thomas G. Lane, Guido Vollbeding. + * All Rights Reserved except as specified below. + * + * Permission is hereby granted to use, copy, modify, and distribute this + * software (or portions thereof) for any purpose, without fee, subject to these + * conditions: + * (1) If any part of the source code for this software is distributed, then this + * README file must be included, with this copyright and no-warranty notice + * unaltered; and any additions, deletions, or changes to the original files + * must be clearly indicated in accompanying documentation. + * (2) If only executable code is distributed, then the accompanying + * documentation must state that "this software is based in part on the work of + * the Independent JPEG Group". + * (3) Permission for use of this software is granted only if the user accepts + * full responsibility for any undesirable consequences; the authors accept + * NO LIABILITY for damages of any kind. + * + * These conditions apply to any software derived from or based on the IJG code, + * not just to the unmodified library. If you use our work, you ought to + * acknowledge us. + * + * Permission is NOT granted for the use of any IJG author's name or company name + * in advertising or publicity relating to this software or products derived from + * it. This software may be referred to only as "the Independent JPEG Group's + * software". + * + * We specifically permit and encourage the use of this software as the basis of + * commercial products, provided that all warranty or liability claims are + * assumed by the product vendor. + * + * This file contains a slower but more accurate integer implementation of the + * forward DCT (Discrete Cosine Transform). + * + * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT + * on each column. Direct algorithms are also available, but they are + * much more complex and seem not to be any faster when reduced to code. + * + * This implementation is based on an algorithm described in + * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT + * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics, + * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991. + * The primary algorithm described there uses 11 multiplies and 29 adds. + * We use their alternate method with 12 multiplies and 32 adds. + * The advantage of this method is that no data path contains more than one + * multiplication; this allows a very simple and accurate implementation in + * scaled fixed-point arithmetic, with a minimal number of shifts. + */ + +Original src/avx2/fdct.rs opening notice + +/* + * Ported from mozjpeg / jfdctint-avx2.asm to rust + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2009, 2016, 2018, 2020, D. R. Commander. + * + * Based on the x86 SIMD extension for IJG JPEG library + * Copyright (C) 1999-2006, MIYASAKA Masaru. + */ diff --git a/README.md b/README.md index a45fd5870..8150dfea9 100644 --- a/README.md +++ b/README.md @@ -120,6 +120,13 @@ See the [release notes](CHANGELOG.md) for included functionality and known limit pages and version detection, and written back; and `model`, the drawing above the tags (header variables, tables, blocks, layouts, entities) and its writer (R12 to 2018), merged from CADCraft's and VectorCraft's readers. Bounded on hostile input. Mapping the model onto a document stays in each app. +- [`craft-deckcraft-format`](crates/deckcraft-format/README.md): the DeckCraft `.deckcraft` package: + identity and sniffing, the `presentation.json` envelope with the presentation as a type + parameter, NaN-safe JSON and the reader for older files with `null` numbers, member names and + compression, and bounded, budgeted member reads. The presentation model and its mapping stay in DeckCraft. +- [`craft-dwg`](crates/dwg/README.md): bounded byte-level DWG↔DXF conversion with diagnostics, + capped output and an optional child-process worker; the `acadrust` dependency remains isolated + behind bytes and is unavailable on wasm32 and FreeBSD/aarch64. - [`craft-ecproj-format`](crates/ecproj-format/README.md): the envelope of EffectCraft `.ecproj` project files: `savedBy`, the schema version and its check, and the file object with the project and its saved view as type parameters. The project model stays in EffectCraft. @@ -169,6 +176,9 @@ See the [release notes](CHANGELOG.md) for included functionality and known limit every frame's position, delay, disposal and transparent colour, the loop count and the first frame's colour table, without decompressing pixels. Tells an animation from a still and whether a file can hold the picture it declares before a decoder runs. +- [`craft-gridcraft-format`](crates/gridcraft-format/README.md): GridCraft's native JSON workbook + (`.json`, `.scjson`): the extensions, the reader, the pretty writer, and newer entries kept + unread and written back into a list. The workbook model and what follows a read stay in GridCraft. - [`craft-hdr`](crates/hdr/src/lib.rs): Radiance HDR (RGBE) picture reader, from FilmCraft: flat, old and new run-length scan lines and all eight axis orders, to linear float RGB. Scan lines are validated before the picture is allocated. @@ -203,6 +213,8 @@ See the [release notes](CHANGELOG.md) for included functionality and known limit decoder is bounded and returns its partial output on damage. From PdfCraft; a PDF filter chain is built on it in `craft-pdf`. - [`craft-timeline`](crates/timeline/README.md): editorial timeline interchange: exact media time and SMPTE timecode, import/export reports, media path and `file://` URL handling, Avid Log Exchange (ALE) and CMX 3600 EDL documents, format detection. From FilmCraft. +- [`craft-time`](crates/time/README.md): exact rational ticks and SMPTE time across media, with a `soundcraft` compatibility module retaining SoundCraft's sample-index timebase, tempo maps, grids and session serialization semantics. +- [`craft-scraft-format`](crates/scraft-format/README.md): SoundCraft's typed `.scraft` session schema, versioned reader/writer, hostile-input validation, and `.scraftz` bundle reader/writer. Engine and UI mapping stays in SoundCraft. - [`craft-metadata`](crates/metadata/README.md): photo metadata, from LightCraft: EXIF (read and write, maker-note lens and focus point, every tag by name), IPTC-IIM, XMP packets (read, write, and a merge that keeps what another program wrote), GPX track logs, and the blocks that carry them in JPEG, PNG, @@ -226,9 +238,10 @@ See the [release notes](CHANGELOG.md) for included functionality and known limit sockets: session, requests of method calls with back-references, responses, ids, address templates, credentials and the judgement of HTTP answers, over any HTTP implementation. Bounded on hostile answers; what a program asks and makes of the answers stays in it. -- [`craft-jpeg`](crates/jpeg/README.md): clean-room decoder for 12-bit and arithmetic-coded - JPEG (T.81 SOF0/SOF1 at 12 bits, SOF9, SOF10), from PhotoCraft, and a frame probe for routing. - Metadata stays in each app. +- [`craft-jpeg`](crates/jpeg/README.md): clean-room JPEG decoder for 12-bit and arithmetic-coded + frames (T.81 SOF0/SOF1 at 12 bits, SOF9, SOF10), band decoding, and an optimized 8-bit baseline + encoder, from PhotoCraft; includes a frame probe for routing. Metadata interpretation stays in + each app. - [`craft-raster-codecs`](crates/raster-codecs/README.md): small raster formats, one module each, from PhotoCraft: DDS (BC1 to BC3, read and write), DICOM, FITS (read and write), IFF/ILBM (read and write), PCX, PNM/PAM/PFM (read and write) and XPM, into one flat image type. Bounded and @@ -325,6 +338,8 @@ See the [release notes](CHANGELOG.md) for included functionality and known limit - [`craft-sheet`](crates/sheet/README.md): spreadsheet vocabulary shared by `craft-xls`, `craft-xlsx` and `craft-ods`: cell positions and A1 text, cell values and error values, sheet-name quoting, Excel's built-in number formats and indexed colour palette. Not a file format; no dependencies. - [`craft-sf2`](crates/sf2/README.md): SoundFont 2 (`.sf2`) reader: the sample pool and, per preset, the key and velocity regions a note sounds, with tuning, loop points, pan, level and volume envelope resolved from the preset and instrument zones. Bounded on hostile input. From LoopCraft. - [`craft-silhouette-studio`](crates/silhouette-studio/src/lib.rs): bounded, fail-closed reader for Silhouette Studio `.studio` and `.studio3` documents: outlines, line colours and page size in millimetres. Protected designs are refused. +- [`craft-stencilcraft-format`](crates/stencilcraft-format/README.md): the versioned `.stencilcraft` JSON wire schema and envelope: required members, version 1 defaults, optional fields, and header checks. +- `craft-stencilcraft-document`: StencilCraft's typed document model, geometry helpers, validation, and native `.stencilcraft` codec. - [`craft-sketch`](crates/sketch/README.md): clean-room reader for the `.sketch` container (Sketch 48 and newer): the ZIP archive, the version cut, `document.json`, the pages in document order and images, as the JSON the file holds. Bounded on hostile input. Carved from FigCraft. - [`craft-svg`](crates/svg/src/lib.rs): SVG. `sniff` recognises SVG and SVGZ data; `svgz` unpacks SVGZ within a byte limit, decodes UTF-8, UTF-16 and ISO 8859-1, and compresses; `guard` bounds @@ -340,6 +355,9 @@ See the [release notes](CHANGELOG.md) for included functionality and known limit - [`craft-text`](crates/text/README.md): text processing below the composer. Its `hyphen` module finds hyphenation points, with a bundled US English dictionary and patterns and optional hypher patterns for 47 more languages (one cargo feature each). +- [`craft-texture`](crates/texture/README.md): linear float image planes, gray or RGBA, with + nearest and bilinear sampling that repeats or clamps, resizing and display conversion. Sizes are + bounded by `Limits`. From MaterialCraft. - [`craft-ui`](crates/ui/README.md): SVG catalogs, icon buttons, tabs, pane controls, transactional docking, print settings and theme helpers for egui frontends; assets, themes, localization and command handling stay in each app. Requires Rust 1.95 or later. - [`craft-vcard`](crates/vcard/README.md): vCard contacts, from MailCraft: vCard 2.1, 3.0 and 4.0 read diff --git a/crates/aaf/src/aaf.rs b/crates/aaf/src/aaf.rs index 05894f94e..efa618fb3 100644 --- a/crates/aaf/src/aaf.rs +++ b/crates/aaf/src/aaf.rs @@ -57,13 +57,56 @@ pub struct WriteOptions<'a> { /// reported. pub fn write(doc: &Document, opts: &WriteOptions<'_>, report: &mut Report) -> Result> { let version = if opts.small_sectors { craft_cfb::Version::V3 } else { craft_cfb::Version::V4 }; - write::write(doc, version, &opts.identity, report) + write::write(doc, version, &opts.identity, &ReferenceOptions::default(), report) +} + +/// Optional interpretation choices for applications retaining older interchange behavior. +#[derive(Clone, Copy, Debug)] +pub struct ReadOptions { + /// Interpret a source clip's start using its target slot's edit rate. + pub source_slot_rate: bool, + /// Use the referenced mob name when the clip has no explicit name. + pub infer_clip_name: bool, + /// Decode percent escapes in relative network locators. + pub decode_relative_locator: bool, + /// Report selector and nested-scope flattening. + pub report_flattening: bool, + /// Accept truncated PCM WAVE essence summaries. + pub permissive_pcm: bool, +} +impl Default for ReadOptions { + fn default() -> Self { + Self { source_slot_rate: true, infer_clip_name: true, decode_relative_locator: true, report_flattening: true, permissive_pcm: false } + } +} +/// Read using explicitly selected interpretation choices. +pub fn read_with_options(bytes: &[u8], options: &ReadOptions, report: &mut Report) -> Result { + read::read(bytes, options, report) +} + +/// Optional compatibility choices for writing source references and locators. +#[derive(Clone, Copy, Debug)] +pub struct ReferenceOptions { + /// Write source offsets in target-slot units. + pub source_slot_rate: bool, + /// Percent-encode relative locators. + pub encode_relative_locator: bool, +} +impl Default for ReferenceOptions { + fn default() -> Self { + Self { source_slot_rate: true, encode_relative_locator: true } + } +} +/// Write with explicitly selected reference and locator conventions. +pub fn write_with_references(doc: &Document, opts: &WriteOptions<'_>, references: &ReferenceOptions, report: &mut Report) -> Result> { + let version = if opts.small_sectors { craft_cfb::Version::V3 } else { craft_cfb::Version::V4 }; + write::write(doc, version, &opts.identity, references, report) } /// Read an AAF file: its top-level compositions, the compositions nested in them and their /// sources. What cannot be represented is skipped and named in `report`. pub fn read(bytes: &[u8], report: &mut Report) -> Result { - read::read(bytes, report) + read::read(bytes, &ReadOptions::default(), report) } /// Whether `bytes` look like an AAF file (a compound file whose root has the AAF root class or a diff --git a/crates/aaf/src/aaf/hostile_tests.rs b/crates/aaf/src/aaf/hostile_tests.rs index 853b0000c..527e50da8 100644 --- a/crates/aaf/src/aaf/hostile_tests.rs +++ b/crates/aaf/src/aaf/hostile_tests.rs @@ -108,3 +108,19 @@ fn objects_nested_too_deeply_are_an_error() { let e = store::read(&chain(250)).unwrap_err(); assert!(e.contains("nested too deeply"), "{e}"); } + +/// Many distinct property IDs still reject a duplicate without quadratic scanning. +#[test] +fn duplicate_after_many_distinct_properties_is_rejected() { + let mut props = vec![0x4C, 0x20]; + props.extend_from_slice(&8193u16.to_le_bytes()); + for id in (0..8192u16).chain(std::iter::once(7)) { + props.extend_from_slice(&id.to_le_bytes()); + props.extend_from_slice(&0x82u16.to_le_bytes()); + props.extend_from_slice(&0u16.to_le_bytes()); + } + let mut writer = Writer::new(Version::V3); + writer.stream(Writer::ROOT, "properties", props).unwrap(); + let e = store::read(&writer.finish()).unwrap_err(); + assert!(e.contains("duplicate property id"), "{e}"); +} diff --git a/crates/aaf/src/aaf/read.rs b/crates/aaf/src/aaf/read.rs index 31c841c5d..9739f87ba 100644 --- a/crates/aaf/src/aaf/read.rs +++ b/crates/aaf/src/aaf/read.rs @@ -67,6 +67,7 @@ fn indirect_f64(d: &[u8]) -> Option { struct R<'a> { report: &'a mut Report, + options: super::ReadOptions, mobs: HashMap, essence: HashMap, doc: Document, @@ -90,7 +91,7 @@ struct NestClip<'a> { /// Compositions nested deeper than this are read as gaps. const MAX_NESTING: usize = 16; -pub(crate) fn read(bytes: &[u8], report: &mut Report) -> Result { +pub(crate) fn read(bytes: &[u8], options: &super::ReadOptions, report: &mut Report) -> Result { let root = store::read(bytes).map_err(|e| Error::Parse { format: "AAF", message: e })?; let header = root.strong(pid::ROOT_HEADER).ok_or_else(|| Error::Parse { format: "AAF", message: "no header object".into() })?; let content = header.strong(pid::CONTENT).ok_or_else(|| Error::Parse { format: "AAF", message: "no content storage".into() })?; @@ -106,7 +107,8 @@ pub(crate) fn read(bytes: &[u8], report: &mut Report) -> Result { essence.insert(id, d); } } - let mut r = R { report, mobs, essence, doc: Document::default(), sources: HashMap::new(), nests: HashMap::new(), open: Vec::new() }; + let mut r = + R { report, options: *options, mobs, essence, doc: Document::default(), sources: HashMap::new(), nests: HashMap::new(), open: Vec::new() }; // Top-level compositions: those no other composition uses, among the ones tagged top-level // when any are. (Premiere Pro tags a nested sequence's composition top-level too; it is still // only a nested sequence.) Compositions that only use each other in a circle all count. The @@ -331,7 +333,9 @@ impl R<'_> { items } x if x == cls::SELECTOR => { - self.report.info("a selector was flattened to its selected input"); + if self.options.report_flattening { + self.report.info("a selector was flattened to its selected input"); + } match c.strong(pid::SELECTED) { Some(s) => { let mut items = self.segment(s, rate, kind, depth); @@ -342,7 +346,9 @@ impl R<'_> { } } x if x == cls::NESTED_SCOPE => { - self.report.info("a nested scope was flattened to its final input"); + if self.options.report_flattening { + self.report.info("a nested scope was flattened to its final input"); + } match c.objs(pid::NESTED_SLOTS).last() { Some(s) => { let mut items = self.segment(s, rate, kind, depth); @@ -388,7 +394,12 @@ impl R<'_> { let len = units_to_ticks(len_of(c), rate.0, rate.1); let tagged_name = tagged_values(c, pid::COMPONENT_USER_COMMENTS).into_iter().find(|(k, _)| k == "Clip Name").map(|(_, v)| v); let (Some(id), Some(slot)) = (c.mob_id(pid::SOURCE_ID), c.u32(pid::SOURCE_MOB_SLOT_ID)) else { return CItem::Filler(len) }; - let name = tagged_name.or_else(|| self.mobs.get(&id).and_then(|mob| mob.string(pid::MOB_NAME))).unwrap_or_default(); + let name = if self.options.infer_clip_name { + tagged_name.or_else(|| self.mobs.get(&id).and_then(|mob| mob.string(pid::MOB_NAME))) + } else { + tagged_name + } + .unwrap_or_default(); if id == [0; 32] { return CItem::Filler(len); } @@ -401,6 +412,7 @@ impl R<'_> { .and_then(|mob| mob.objs(pid::SLOTS).iter().find(|s| s.u32(pid::SLOT_ID) == Some(slot))) .and_then(|s| rate_of(s, pid::EDIT_RATE)) .unwrap_or(rate); + let source_rate = if self.options.source_slot_rate { source_rate } else { rate }; let start = units_to_ticks(c.i64(pid::START_TIME).unwrap_or(0), source_rate.0, source_rate.1); match self.resolve(id, slot, start, kind, master_key, 0) { Some((source, t)) => CItem::Clip(CClip { len, source, start: t, gain: None, name }), @@ -506,6 +518,7 @@ impl R<'_> { .and_then(|m| m.objs(pid::SLOTS).iter().find(|s| s.u32(pid::SLOT_ID) == Some(next_slot))) .and_then(|s| rate_of(s, pid::EDIT_RATE)) .unwrap_or(rate); + let source_rate = if self.options.source_slot_rate { source_rate } else { rate }; let inner = units_to_ticks(clip.i64(pid::START_TIME).unwrap_or(0), source_rate.0, source_rate.1); if next == [0; 32] { return None; @@ -545,7 +558,11 @@ impl R<'_> { DataKind::Sound => CKind::Sound, _ => kind, }; - let path = d.objs(pid::LOCATOR).iter().find_map(|l| l.string(pid::URL_STRING)).map(|u| locator_path(&u)); + let path = d + .objs(pid::LOCATOR) + .iter() + .find_map(|l| l.string(pid::URL_STRING)) + .map(|u| if self.options.decode_relative_locator { locator_path(&u) } else { craft_timeline::path::file_url_to_path(&u) }); let file_rate = rate_of(d, pid::SAMPLE_RATE).unwrap_or(rate); let length = d .i64(pid::FILE_LENGTH) @@ -604,7 +621,8 @@ impl R<'_> { if let Some(data) = self.essence.get(&id) { s.embedded = Some(data.to_vec()); // embedded essence may be a whole WAVE / AIFF file - if let Some((pcm, ch, sr, bits)) = crate::pcm::parse_wav(data) { + let pcm = if self.options.permissive_pcm { craft_audio::essence_pcm::parse_wav(data) } else { crate::pcm::parse_wav(data) }; + if let Some((pcm, ch, sr, bits)) = pcm { s.embedded = Some(pcm); s.file_channels = ch as u32; s.channels = if s.channel.is_some() { 1 } else { ch as u32 }; diff --git a/crates/aaf/src/aaf/store.rs b/crates/aaf/src/aaf/store.rs index 69b14fcf2..859c0ea77 100644 --- a/crates/aaf/src/aaf/store.rs +++ b/crates/aaf/src/aaf/store.rs @@ -461,11 +461,13 @@ impl R<'_> { if at > props.len() { return Err(bad("truncated property index")); } + // A properties stream can name tens of thousands of ids: do not rescan all earlier ids. + let mut seen = std::collections::HashSet::with_capacity(n); for i in 0..n { let (Some(pid), Some(sf), Some(len)) = (rd_u16(&props, 4 + i * 6), rd_u16(&props, 6 + i * 6), rd_u16(&props, 8 + i * 6)) else { return Err(bad("truncated property index")); }; - if o.props.iter().any(|(previous, _)| *previous == pid) { + if !seen.insert(pid) { return Err(bad("duplicate property id")); } let v = props.get(at..at + len as usize).ok_or_else(|| bad("truncated property value"))?; diff --git a/crates/aaf/src/aaf/tests.rs b/crates/aaf/src/aaf/tests.rs index 3e7794471..34eb2bbb3 100644 --- a/crates/aaf/src/aaf/tests.rs +++ b/crates/aaf/src/aaf/tests.rs @@ -557,3 +557,32 @@ fn relative_media_locator_round_trips_escaped_characters() { let got = read_clean(&bytes); assert!(got.sources.iter().any(|s| s.path.as_deref() == Some("Audio Files/100%20 #1.wav"))); } + +#[test] +fn compatibility_options_keep_relative_paths_and_parent_rate() { + let mut doc = crate::testdoc::sound_only(None); + doc.compositions[0].sample_rate = 44_100; + doc.sources[0].path = Some("Audio Files/100%20 #1.wav".into()); + let mut report = Report::default(); + let references = ReferenceOptions { source_slot_rate: false, encode_relative_locator: false }; + let bytes = write_with_references(&doc, &WriteOptions { identity: IDENTITY, small_sectors: false }, &references, &mut report).unwrap(); + let root = store::read(&bytes).unwrap(); + let mut all = Vec::new(); + walk(&root, &mut all); + assert!( + all.iter().filter(|o| o.class == cls::NETWORK_LOCATOR).any(|o| o.string(pid::URL_STRING).as_deref() == Some("Audio Files/100%20 #1.wav")) + ); + let starts: Vec = all.iter().filter(|o| o.class == cls::SOURCE_CLIP).filter_map(|o| o.i64(pid::START_TIME)).collect(); + assert!(starts.contains(&17_640), "parent rate start: {starts:?}"); + let opts = ReadOptions { + source_slot_rate: false, + infer_clip_name: false, + decode_relative_locator: false, + report_flattening: false, + permissive_pcm: true, + }; + let back = read_with_options(&bytes, &opts, &mut report).unwrap(); + assert!(back.sources.iter().any(|s| s.path.as_deref() == Some("Audio Files/100%20 #1.wav"))); + let clip = back.compositions[0].tracks[0].items.iter().find_map(|i| if let CItem::Clip(c) = i { Some(c) } else { None }).unwrap(); + assert_eq!(clip.start, FrameRate::FPS_25.tick_of(10)); +} diff --git a/crates/aaf/src/aaf/write.rs b/crates/aaf/src/aaf/write.rs index 8c9aae88e..1fcc6a332 100644 --- a/crates/aaf/src/aaf/write.rs +++ b/crates/aaf/src/aaf/write.rs @@ -75,6 +75,7 @@ fn definition(class: Auid, id: Auid, name: &str) -> Obj { struct W<'a> { report: &'a mut Report, inexact: bool, + references: super::ReferenceOptions, } struct IdGen { @@ -350,9 +351,15 @@ fn add_tape_mob(first: &Source, tape_id: [u8; 32], timecode: Option<(i64, FrameR } /// Serialise a composition document. -pub(crate) fn write(doc: &Document, version: craft_cfb::Version, identity: &Identity<'_>, report: &mut Report) -> Result> { +pub(crate) fn write( + doc: &Document, + version: craft_cfb::Version, + identity: &Identity<'_>, + references: &super::ReferenceOptions, + report: &mut Report, +) -> Result> { let comp = doc.compositions.first().ok_or(Error::Empty)?; - let mut w = W { report, inexact: false }; + let mut w = W { report, inexact: false, references: *references }; let mut ids = IdGen::for_document(doc, comp); let MediaOutput { mut mobs, essence, master_ref } = write_media(doc, &mut w, &mut ids); let mut used_ops: Vec = Vec::new(); @@ -360,7 +367,8 @@ pub(crate) fn write(doc: &Document, version: craft_cfb::Version, identity: &Iden // ---- nested compositions: a clip refers to one as it does to a master mob, through the // composition's first track of the clip's kind (a broken-out channel: that channel's track) - let mut refs = Refs { mobs: master_ref, rates: doc.sources.iter().enumerate().map(|(i, s)| (i, source_rate(s))).collect() }; + let rates = if references.source_slot_rate { doc.sources.iter().enumerate().map(|(i, s)| (i, source_rate(s))).collect() } else { HashMap::new() }; + let mut refs = Refs { mobs: master_ref, rates }; let nested_ids: Vec<[u8; 32]> = doc.nested.iter().map(|_| ids.next()).collect(); for (si, s) in doc.sources.iter().enumerate() { let Some((comp, id)) = s.nested.and_then(|n| Some((doc.nested.get(n)?, nested_ids.get(n)?))) else { continue }; @@ -795,8 +803,10 @@ fn descriptor(w: &mut W, s: &Source) -> Obj { if let Some(p) = &s.path { let url = if craft_timeline::path::is_absolute(p) { craft_timeline::path::path_to_file_url(p, false) - } else { + } else if w.references.encode_relative_locator { craft_timeline::path::encode_uri_path(p) + } else { + p.clone() }; d.set(pid::LOCATOR, Value::StrongVec(vec![Obj::new(cls::NETWORK_LOCATOR).data_prop(pid::URL_STRING, utf16z(&url))])); } diff --git a/crates/aaf/src/lib.rs b/crates/aaf/src/lib.rs index ee7448bc0..4968c5b39 100644 --- a/crates/aaf/src/lib.rs +++ b/crates/aaf/src/lib.rs @@ -29,7 +29,11 @@ mod roundtrip_tests; #[cfg(test)] mod testdoc; +/// Diagnostic types used by readers and writers. +pub use craft_timeline::report; pub use craft_timeline::report::Report; +/// Exact time types used by the composition model. +pub use craft_timeline::time; /// Error from a reader or writer. Unsupported *features* never error; they go in the [`Report`]. #[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)] diff --git a/crates/aaf/src/omf.rs b/crates/aaf/src/omf.rs index 57e0f756c..bbd2e8571 100644 --- a/crates/aaf/src/omf.rs +++ b/crates/aaf/src/omf.rs @@ -21,7 +21,7 @@ //! transition's name) are not in the specification; they were first written by FilmCraft and keep //! their names so that existing files read back. -pub(crate) mod bento; +pub mod bento; mod read; mod write; diff --git a/crates/aaf/src/omf/bento.rs b/crates/aaf/src/omf/bento.rs index 00a7e2a26..171196fe1 100644 --- a/crates/aaf/src/omf/bento.rs +++ b/crates/aaf/src/omf/bento.rs @@ -35,8 +35,8 @@ use std::collections::{BTreeMap, HashMap}; -pub(crate) const MAGIC: [u8; 8] = [0xA4, b'C', b'M', 0xA5, b'H', b'd', b'r', 0xD7]; -pub(crate) const LABEL_SIZE: usize = 24; +pub const MAGIC: [u8; 8] = [0xA4, b'C', b'M', 0xA5, b'H', b'd', b'r', 0xD7]; +pub const LABEL_SIZE: usize = 24; // standard object ids const TOC_OBJECT: u32 = 1; @@ -82,7 +82,7 @@ const MAX_VALUE_BYTES: usize = 16 << 20; const MAX_NAME_BYTES: usize = 4 << 20; /// Builds a Bento container. -pub(crate) struct Writer { +pub struct Writer { data: Vec, names: HashMap<(u32, String), u32>, next: u32, @@ -90,6 +90,12 @@ pub(crate) struct Writer { entries: Vec<(u32, u32, u32, Vec)>, } +impl Default for Writer { + fn default() -> Self { + Self::new() + } +} + impl Writer { pub fn new() -> Writer { Writer { data: Vec::new(), names: HashMap::new(), next: MIN_USER_ID, entries: Vec::new() } @@ -217,9 +223,9 @@ pub(crate) struct Prop { } /// A parsed Bento container. -pub(crate) struct Container<'a> { +pub struct Container<'a> { data: &'a [u8], - pub objects: BTreeMap>, + pub(crate) objects: BTreeMap>, names: HashMap, by_name: HashMap, } @@ -229,11 +235,16 @@ fn be32(b: &[u8], at: usize) -> Option { } /// Whether `b` ends with a Bento label. -pub(crate) fn sniff(b: &[u8]) -> bool { +pub fn sniff(b: &[u8]) -> bool { b.len() >= LABEL_SIZE && b[b.len() - LABEL_SIZE..b.len() - LABEL_SIZE + 8] == MAGIC } impl<'a> Container<'a> { + /// Object identifiers in ascending order. + pub fn object_ids(&self) -> impl Iterator + '_ { + self.objects.keys().copied() + } + pub fn open(data: &'a [u8]) -> Result, String> { if !sniff(data) { return Err("no Bento label at the end of the file".into()); @@ -324,7 +335,7 @@ impl<'a> Container<'a> { } /// The bytes of a value. - pub fn value(&self, p: &Prop) -> Result, String> { + pub(crate) fn value(&self, p: &Prop) -> Result, String> { let total = p.segs.iter().try_fold(0usize, |total, s| { let len = match s { Seg::Imm(b) => b.len(), diff --git a/crates/aaf/src/omf/read.rs b/crates/aaf/src/omf/read.rs index d45b6ff63..9e892a0ce 100644 --- a/crates/aaf/src/omf/read.rs +++ b/crates/aaf/src/omf/read.rs @@ -12,6 +12,10 @@ type Uid = [u8; 12]; const MAX_REFERENCE_VISITS: usize = 65_536; +/// A source-clip search stopped at its work limit. +#[derive(Clone, Copy)] +struct SearchLimit; + struct R<'a, 'b> { c: &'b Container<'a>, report: &'b mut Report, @@ -21,15 +25,49 @@ struct R<'a, 'b> { media: HashMap>, doc: Document, sources: HashMap<(Uid, u32, String), usize>, + /// Component visits left for [`R::segment`] over the whole document. + budget: usize, + /// Outcome of the source-clip search under each segment: deterministic, so it is made once. + clip_search: HashMap, SearchLimit>>, + /// Visits and queued references left for source-clip searches over the whole document: + /// distinct roots over one shared graph would otherwise each search it in full. + search_budget: usize, } +/// Components one document may expand into timeline items; a small file can reference the same +/// sequence from many places and would otherwise multiply without bound. +const MAX_COMPONENT_VISITS: usize = 1_000_000; + +/// Reported when the shared component budget runs out. +const BUDGET_WARNING: &str = "OMF composition is too large to import completely (the remaining components are dropped)"; +/// Most visited components, and most queued references, one source-clip search may spend. +const MAX_CLIP_SEARCH_VISITS: usize = 100_000; +const MAX_CLIP_SEARCH_PENDING: usize = 1_000_000; +/// Work all source-clip searches in one document may spend together. +const MAX_DOCUMENT_SEARCH_WORK: usize = 2_000_000; + pub(crate) fn read(bytes: &[u8], report: &mut Report) -> Result { + read_with_budget(bytes, report, MAX_COMPONENT_VISITS) +} + +fn read_with_budget(bytes: &[u8], report: &mut Report, budget: usize) -> Result { let perr = |m: String| Error::Parse { format: "OMF", message: m }; let c = Container::open(bytes).map_err(perr)?; let class_of = |o: u32| c.get(o, "OMFI:OOBJ:ObjClass").and_then(|v| v.get(..4).map(|b| [b[0], b[1], b[2], b[3]])); let head = c.objects.keys().copied().find(|&o| class_of(o) == Some(*b"HEAD")).ok_or_else(|| perr("no OMF header object".into()))?; let be = c.get(head, "OMFI:HEAD:ByteOrder").is_none_or(|v| v.first() != Some(&0x49)); - let mut r = R { c: &c, report, be, mobs: HashMap::new(), media: HashMap::new(), doc: Document::default(), sources: HashMap::new() }; + let mut r = R { + c: &c, + report, + be, + mobs: HashMap::new(), + media: HashMap::new(), + doc: Document::default(), + sources: HashMap::new(), + budget, + clip_search: HashMap::new(), + search_budget: MAX_DOCUMENT_SEARCH_WORK, + }; let mob_list = r.refs(head, "OMFI:HEAD:Mobs"); let mob_list = if mob_list.is_empty() { c.objects.keys().copied().filter(|&o| matches!(class_of(o).as_ref(), Some(b"CMOB" | b"MMOB" | b"SMOB"))).collect() @@ -66,8 +104,14 @@ pub(crate) fn read(bytes: &[u8], report: &mut Report) -> Result { cmobs.iter().copied().filter(|m| r.uid(*m, "OMFI:MOBJ:MobID").is_none_or(|id| !referenced.contains(&id))).collect() }; for m in top { + // every composition spends from the budget too: a list of up to 65,535 repeated entries must not yield that many + if r.budget == 0 && !r.doc.compositions.is_empty() { + r.report.warn(BUDGET_WARNING); + break; + } let comp = r.composition(m); r.doc.compositions.push(comp); + r.budget = r.budget.saturating_sub(1); } if r.doc.compositions.is_empty() { return Err(Error::Empty); @@ -166,6 +210,13 @@ impl R<'_, '_> { let slots = self.refs(m, "OMFI:MOBJ:Slots"); let mut na = 0; for s in slots { + if self.budget == 0 { + self.report.warn(BUDGET_WARNING); + break; + } + // Every slot spends a unit, skipped ones too: a list of up to 65,535 repeated slots + // must not cost that much work in each of 65,535 repeated compositions. + self.budget -= 1; let Some(seg) = self.r#ref(s, "OMFI:MSLT:Segment") else { continue }; let rate = self.rational(s, "OMFI:MSLT:EditRate").filter(|r| r.0 > 0 && r.1 > 0).unwrap_or((25, 1)); let kind = self.data_kind(seg); @@ -213,13 +264,32 @@ impl R<'_, '_> { fn segment(&mut self, seg: u32, rate: (i64, i64), depth: usize) -> Vec { let len = units_to_ticks(self.len(seg), rate.0, rate.1); + // Depth alone does not bound a graph that references the same component many times at + // every level, so each visit also spends from a per-document budget. + // A visit stopped by depth spends from it too. + if self.budget == 0 { + self.report.warn(BUDGET_WARNING); + return vec![CItem::Filler(len)]; + } + self.budget -= 1; if depth > 32 { return vec![CItem::Filler(len)]; } match &self.class(seg) { b"SEQU" => { let mut out = Vec::new(); + let mut expanded = 0i64; for c in self.refs(seg, "OMFI:SEQU:Components") { + if self.budget == 0 { + // the rest of the sequence keeps its duration as one gap, not a gap per reference + self.report.warn(BUDGET_WARNING); + let rest = self.len(seg).saturating_sub(expanded); + if rest > 0 { + out.push(CItem::Filler(units_to_ticks(rest, rate.0, rate.1))); + } + break; + } + expanded = expanded.saturating_add(self.len(c)); out.extend(self.segment(c, rate, depth + 1)); } out @@ -306,12 +376,66 @@ impl R<'_, '_> { .find(|&s| self.r#ref(s, "OMFI:MSLT:TrackDesc").and_then(|t| self.u32(t, "OMFI:TRKD:TrackID")) == Some(track)) } - fn first_clip(&self, seg: u32) -> Option { - match &self.class(seg) { - b"SCLP" => Some(seg), - b"SEQU" => self.refs(seg, "OMFI:SEQU:Components").into_iter().find_map(|c| self.first_clip(c)), - _ => None, + /// First source clip under `seg`, depth-first; a sequence referencing itself (directly or + /// through a cycle) ends the search rather than the stack. `Err` when the search spent its + /// budget of visited components or queued references before finishing (the answer is unknown). + fn first_clip(&mut self, seg: u32) -> std::result::Result, SearchLimit> { + if let Some(found) = self.clip_search.get(&seg) { + return *found; } + let mut spent = 0; + let found = if self.search_budget == 0 { + Err(SearchLimit) + } else { + let visits = MAX_CLIP_SEARCH_VISITS.min(self.search_budget); + let pending = MAX_CLIP_SEARCH_PENDING.min(self.search_budget); + self.first_clip_within(seg, visits, pending, &mut spent) + }; + self.search_budget = self.search_budget.saturating_sub(spent); + self.clip_search.insert(seg, found); + found + } + + /// `spent` receives the visits plus queued references the search used. + fn first_clip_within(&self, seg: u32, max_visits: usize, max_pending: usize, spent: &mut usize) -> std::result::Result, SearchLimit> { + let mut visited = HashSet::new(); + let mut todo = vec![seg]; + let mut queued = 1usize; + let found = Self::search_from(self, &mut visited, &mut todo, &mut queued, max_visits, max_pending); + *spent = visited.len().saturating_add(queued); + found + } + + fn search_from( + &self, + visited: &mut HashSet, + todo: &mut Vec, + queued: &mut usize, + max_visits: usize, + max_pending: usize, + ) -> std::result::Result, SearchLimit> { + while let Some(seg) = todo.pop() { + if !visited.insert(seg) { + continue; + } + if visited.len() > max_visits { + return Err(SearchLimit); + } + match &self.class(seg) { + b"SCLP" => return Ok(Some(seg)), + b"SEQU" => { + let comps = self.refs(seg, "OMFI:SEQU:Components"); + // sequences may share one component list, so queued references grow with expansions, not visits + *queued = queued.saturating_add(comps.len()); + if *queued > max_pending { + return Err(SearchLimit); + } + todo.extend(comps.into_iter().rev()); + } + _ => {} + } + } + Ok(None) } fn resolve(&mut self, id: Uid, track: u32, start: Tick, master: Option, depth: usize) -> Option<(usize, Tick)> { @@ -328,7 +452,13 @@ impl R<'_, '_> { match &self.class(mob) { b"MMOB" => { let key = master.unwrap_or_else(|| id.iter().map(|b| format!("{b:02x}")).collect()); - let clip = self.first_clip(seg)?; + let clip = match self.first_clip(seg) { + Ok(c) => c?, + Err(SearchLimit) => { + self.report.warn("a master mob is too large to search for its source clip (imported as a gap)"); + return None; + } + }; let inner = units_to_ticks(self.i32(clip, "OMFI:SCLP:StartTime").unwrap_or(0), rate.0, rate.1); let next = self.uid(clip, "OMFI:SCLP:SourceID")?; let next_track = self.u32(clip, "OMFI:SCLP:SourceTrackID")?; @@ -353,7 +483,9 @@ impl R<'_, '_> { let file_rate = sr.filter(|r| r.0 > 0 && r.1 > 0).unwrap_or(rate); let length = units_to_ticks(self.i32(desc, "OMFI:MDFL:Length").unwrap_or(0), file_rate.0, file_rate.1); let summary = self.c.get(desc, "OMFI:WAVD:Summary").or_else(|| self.c.get(desc, "OMFI:AIFD:Summary")).unwrap_or_default(); - let fmt = crate::pcm::parse_wav(&summary).or_else(|| crate::pcm::parse_aiff(&summary)).map(|(_, ch, sr, bits)| (ch as u32, sr, bits)); + let fmt = craft_audio::essence_pcm::parse_wav(&summary) + .or_else(|| craft_audio::essence_pcm::parse_aiff(&summary)) + .map(|(_, ch, sr, bits)| (ch as u32, sr, bits)); let mut path = None; for l in self.refs(desc, "OMFI:MDES:Locator") { let p = match &self.class(l) { @@ -373,7 +505,8 @@ impl R<'_, '_> { let (file_channels, sample_rate, bits) = fmt.unwrap_or((1, file_rate.0 as u32, 16)); let channel = (file_channels > 1 && sound_slots > 1).then(|| track.saturating_sub(1).min(file_channels - 1)); // timecode of the physical source - let tc = self.first_clip(seg).and_then(|clip| { + let first = self.first_clip(seg).ok().flatten(); + let tc = first.and_then(|clip| { let tid = self.uid(clip, "OMFI:SCLP:SourceID")?; let tape = *self.mobs.get(&tid)?; let (start_tc, fps, trate) = self.refs(tape, "OMFI:MOBJ:Slots").into_iter().find_map(|s| { @@ -415,7 +548,7 @@ impl R<'_, '_> { nested: None, }; if let Some(d) = self.media.get(&id) { - if let Some((pcm, ch, sr, b)) = crate::pcm::parse_wav(d).or_else(|| crate::pcm::parse_aiff(d)) { + if let Some((pcm, ch, sr, b)) = craft_audio::essence_pcm::parse_wav(d).or_else(|| craft_audio::essence_pcm::parse_aiff(d)) { s.embedded = Some(pcm); s.file_channels = ch as u32; s.channels = if s.channel.is_some() { 1 } else { ch as u32 }; @@ -437,3 +570,200 @@ impl R<'_, '_> { } } } + +#[cfg(test)] +mod tests { + use super::super::bento::Writer; + use super::*; + + fn obj(w: &mut Writer, class: &[u8; 4]) -> u32 { + let id = w.new_object(); + w.set(id, "OMFI:OOBJ:ObjClass", "omfi:ClassID", class); + id + } + fn list(w: &mut Writer, o: u32, p: &str, targets: &[u32]) { + let mut v = (targets.len() as u16).to_be_bytes().to_vec(); + for t in targets { + v.extend_from_slice(&t.to_be_bytes()); + } + w.set(o, p, "omfi:ObjRefArray", &v); + } + + fn search(max_visits: usize, max_pending: usize, build: impl FnOnce(&mut Writer) -> u32) -> std::result::Result, SearchLimit> { + let mut w = Writer::new(); + let root = build(&mut w); + let bytes = w.finish(); + let c = Container::open(&bytes).unwrap(); + let mut report = Report::default(); + let r = R { + c: &c, + report: &mut report, + be: true, + mobs: HashMap::new(), + media: HashMap::new(), + doc: Document::default(), + sources: HashMap::new(), + budget: 0, + clip_search: HashMap::new(), + search_budget: MAX_DOCUMENT_SEARCH_WORK, + }; + r.first_clip_within(root, max_visits, max_pending, &mut 0) + } + + #[test] + fn clip_search_reports_its_limits() { + // found within the limits + assert!(matches!( + search(10, 10, |w| { + let (s, c) = (obj(w, b"SEQU"), obj(w, b"SCLP")); + list(w, s, "OMFI:SEQU:Components", &[c]); + s + }), + Ok(Some(_)) + )); + // too many distinct components before any clip + assert!( + search(5, 1000, |w| { + let (s, f) = (obj(w, b"SEQU"), (0..20).map(|_| obj(w, b"FILL")).collect::>()); + list(w, s, "OMFI:SEQU:Components", &f); + s + }) + .is_err() + ); + // few visits, but sequences that each queue a long list of references + assert!( + search(1000, 500, |w| { + let f = obj(w, b"FILL"); + let inner: Vec = (0..20).map(|_| obj(w, b"SEQU")).collect(); + for &i in &inner { + list(w, i, "OMFI:SEQU:Components", &[f; 100]); + } + let s = obj(w, b"SEQU"); + list(w, s, "OMFI:SEQU:Components", &inner); + s + }) + .is_err() + ); + // a search that ends without a clip is not a limit + assert!(matches!(search(10, 10, |w| obj(w, b"FILL")), Ok(None))); + } + + #[test] + fn skipped_slots_spend_the_budget() { + let mut w = Writer::new(); + let head = obj(&mut w, b"HEAD"); + let cm = obj(&mut w, b"CMOB"); + let slot = obj(&mut w, b"MSLT"); + list(&mut w, cm, "OMFI:MOBJ:Slots", &[slot; 200]); + list(&mut w, head, "OMFI:HEAD:Mobs", &[cm]); + list(&mut w, head, "OMFI:HEAD:PrimaryMobs", &[cm]); + let bytes = w.finish(); + let mut report = Report::default(); + read_with_budget(&bytes, &mut report, 50).unwrap(); + assert!(report.mentions(BUDGET_WARNING), "slots with no segment cost nothing before"); + } + + #[test] + fn a_source_clip_search_is_made_once_per_segment() { + let mut w = Writer::new(); + let (s, c) = (obj(&mut w, b"SEQU"), obj(&mut w, b"SCLP")); + list(&mut w, s, "OMFI:SEQU:Components", &[c]); + let bytes = w.finish(); + let c = Container::open(&bytes).unwrap(); + let mut report = Report::default(); + let mut r = R { + c: &c, + report: &mut report, + be: true, + mobs: HashMap::new(), + media: HashMap::new(), + doc: Document::default(), + sources: HashMap::new(), + budget: 0, + clip_search: HashMap::new(), + search_budget: MAX_DOCUMENT_SEARCH_WORK, + }; + let first = r.first_clip(s).ok(); + assert!(matches!(first, Some(Some(_)))); + assert_eq!(r.clip_search.len(), 1); + assert_eq!(r.first_clip(s).ok(), first); + assert_eq!(r.clip_search.len(), 1); + } + + #[test] + fn repeated_slots_and_compositions_spend_the_budget() { + let mut w = Writer::new(); + let head = obj(&mut w, b"HEAD"); + let cm = obj(&mut w, b"CMOB"); + let dd = obj(&mut w, b"DDEF"); + w.set(dd, "OMFI:DDEF:DataKindID", "omfi:UniqueName", b"omfi:data:Sound\0"); + let fill = obj(&mut w, b"FILL"); + w.set(fill, "OMFI:CPNT:DataKind", "omfi:ObjRef", &dd.to_be_bytes()); + let slot = obj(&mut w, b"MSLT"); + w.set(slot, "OMFI:MSLT:Segment", "omfi:ObjRef", &fill.to_be_bytes()); + list(&mut w, cm, "OMFI:MOBJ:Slots", &[slot; 200]); + list(&mut w, head, "OMFI:HEAD:Mobs", &[cm]); + list(&mut w, head, "OMFI:HEAD:PrimaryMobs", &[cm; 200]); + let bytes = w.finish(); + let mut report = Report::default(); + let doc = read_with_budget(&bytes, &mut report, 50).unwrap(); + let tracks: usize = doc.compositions.iter().map(|c| c.tracks.len()).sum(); + assert!(tracks <= 50, "{tracks} tracks from a budget of 50"); + assert!(doc.compositions.len() <= 2, "{} compositions", doc.compositions.len()); + assert!(report.has_warnings()); + } + + fn with_reader(bytes: &[u8], budget: usize, search_budget: usize, f: impl FnOnce(&mut R) -> T) -> (T, Report) { + let c = Container::open(bytes).unwrap(); + let mut report = Report::default(); + let mut r = R { + c: &c, + report: &mut report, + be: true, + mobs: HashMap::new(), + media: HashMap::new(), + doc: Document::default(), + sources: HashMap::new(), + budget, + clip_search: HashMap::new(), + search_budget, + }; + let out = f(&mut r); + (out, report) + } + + #[test] + fn depth_limited_components_spend_the_budget() { + // a chain 33 sequences deep whose last one repeats a depth-limited sequence many times + let mut w = Writer::new(); + let deep = obj(&mut w, b"SEQU"); + let chain: Vec = (0..33).map(|_| obj(&mut w, b"SEQU")).collect(); + for pair in chain.windows(2) { + list(&mut w, pair[0], "OMFI:SEQU:Components", &[pair[1]]); + } + list(&mut w, chain[32], "OMFI:SEQU:Components", &[deep; 1000]); + let bytes = w.finish(); + let (items, report) = with_reader(&bytes, 100, MAX_DOCUMENT_SEARCH_WORK, |r| r.segment(chain[0], (48_000, 1), 0)); + assert!(items.len() <= 100, "{} items from a budget of 100", items.len()); + assert!(report.mentions(BUDGET_WARNING)); + } + + #[test] + fn source_clip_searches_share_a_document_budget() { + // distinct roots over one shared clip-free graph + let mut w = Writer::new(); + let fills: Vec = (0..50).map(|_| obj(&mut w, b"FILL")).collect(); + let shared = obj(&mut w, b"SEQU"); + list(&mut w, shared, "OMFI:SEQU:Components", &fills); + let roots: Vec = (0..30).map(|_| obj(&mut w, b"SEQU")).collect(); + for &root in &roots { + list(&mut w, root, "OMFI:SEQU:Components", &[shared]); + } + let bytes = w.finish(); + let (results, _) = with_reader(&bytes, 0, 500, |r| roots.iter().map(|&root| r.first_clip(root)).collect::>()); + assert!(matches!(results[0], Ok(None)), "the first search fits"); + assert!(results.iter().any(|r| r.is_err()), "later searches run out of the shared budget"); + let (results, _) = with_reader(&bytes, 0, MAX_DOCUMENT_SEARCH_WORK, |r| roots.iter().map(|&root| r.first_clip(root)).collect::>()); + assert!(results.iter().all(|r| matches!(r, Ok(None)))); + } +} diff --git a/crates/aaf/src/pcm.rs b/crates/aaf/src/pcm.rs index 789c976a8..9992b2d3b 100644 --- a/crates/aaf/src/pcm.rs +++ b/crates/aaf/src/pcm.rs @@ -4,8 +4,3 @@ pub(crate) fn parse_wav(b: &[u8]) -> Option<(Vec, u16, u32, u16)> { craft_audio::wav_pcm(b).map(|p| (p.data, p.channels, p.sample_rate, p.bits)) } - -/// The same for an AIFF / AIFF-C (`NONE`, `twos`, `sowt`) file. -pub(crate) fn parse_aiff(b: &[u8]) -> Option<(Vec, u16, u32, u16)> { - craft_audio::aiff_pcm(b).map(|p| (p.data, p.channels, p.sample_rate, p.bits)) -} diff --git a/crates/aaf/tests/bento_api.rs b/crates/aaf/tests/bento_api.rs new file mode 100644 index 000000000..93adc33c1 --- /dev/null +++ b/crates/aaf/tests/bento_api.rs @@ -0,0 +1,27 @@ +use craft_aaf::omf::bento::{Container, Writer, sniff}; + +#[test] +fn external_consumer_reads_named_values_and_references() { + let mut writer = Writer::new(); + let first = writer.new_object(); + let second = writer.new_object(); + writer.set(first, "test:value", "test:type", b"long value"); + writer.set_references(first, &[second]); + let bytes = writer.finish(); + assert!(sniff(&bytes)); + let container = Container::open(&bytes).unwrap(); + assert!(container.object_ids().any(|id| id == first)); + assert_eq!(container.get(first, "test:value").as_deref(), Some(b"long value".as_slice())); + assert_eq!(container.resolve(first, second), second); + assert!(Container::open(&bytes[..bytes.len() - 1]).is_err()); +} + +#[test] +fn default_writer_builds_a_readable_empty_container() -> Result<(), String> { + let bytes = Writer::default().finish(); + let container = Container::open(&bytes)?; + let ids: Vec<_> = container.object_ids().collect(); + assert!(!ids.is_empty()); + assert!(ids.windows(2).all(|pair| pair[0] < pair[1])); + Ok(()) +} diff --git a/crates/affinity/src/model/read.rs b/crates/affinity/src/model/read.rs index c0cf58e77..14b17ad99 100644 --- a/crates/affinity/src/model/read.rs +++ b/crates/affinity/src/model/read.rs @@ -679,7 +679,11 @@ impl<'s, 'a, 'b> Reader<'s, 'a, 'b> { } let class = s.class(id).unwrap_or(Tag::of(b"ShpN")); let (path, even_odd) = crate::geometry::outline(self, id, class, world).unwrap_or_else(|| { - self.warn("an artboard whose outline could not be read (clipped to its rectangle)"); + if class == Tag::of(b"PCrv") { + self.warn("a converted artboard outline could not be read (clipped to its rectangle)"); + } else { + self.warn("an artboard whose outline could not be read (clipped to its rectangle)"); + } (Path { subpaths: vec![crate::shapes::rectangle(b.x0, b.y0, b.x1, b.y1)] }, false) }); let (background, strokes) = paint::node_paint(self, id, world); diff --git a/crates/anicraft-format/Cargo.toml b/crates/anicraft-format/Cargo.toml new file mode 100644 index 000000000..8dad7af33 --- /dev/null +++ b/crates/anicraft-format/Cargo.toml @@ -0,0 +1,16 @@ +[package] +name = "craft-anicraft-format" +description = "Versioned JSON envelope codec for AniCraft native documents" +version.workspace = true +edition.workspace = true +license.workspace = true +rust-version.workspace = true +repository.workspace = true + +[lints] +workspace = true + +[dependencies] +serde = { workspace = true, features = ["derive"] } +serde_json = { workspace = true } + diff --git a/crates/anicraft-format/src/lib.rs b/crates/anicraft-format/src/lib.rs new file mode 100644 index 000000000..7cc5dc261 --- /dev/null +++ b/crates/anicraft-format/src/lib.rs @@ -0,0 +1,260 @@ +//! The versioned JSON envelope used by AniCraft's native project file. +//! +//! The app owns its document model and the path-table scope used while serializing it. This +//! crate owns the format envelope: field order, duplicate-field handling, and the point at which +//! the caller decodes the document relative to the `paths` field. +//! +//! [`read`] passes the actual nested Serde deserializer to the caller's [`PayloadDecoder`]. That +//! keeps payload parsing in the app's model crate while preserving the original stream order, +//! global JSON error locations, and path-table guard lifetime. + +use serde::{ + Serialize, + de::{Deserialize, DeserializeSeed, MapAccess, Visitor}, +}; +use std::{fmt, marker::PhantomData}; + +/// The native project envelope's identifying format string. +pub const FORMAT_NAME: &str = "anicraft"; +/// Current envelope version (v1 inline paths; v2 shared path table; v3 named scene timelines). +pub const FORMAT_VERSION: u32 = 3; +/// The largest native project file AniCraft accepts. +pub const MAX_FILE_BYTES: u64 = 1024 * 1024 * 1024; +/// Files below this size are pretty-printed for diffs and manual inspection. +pub const PRETTY_LIMIT: usize = 1024 * 1024; + +#[derive(Serialize)] +struct EnvelopeOut<'a, P, T> { + format: &'a str, + version: u32, + generator: &'a str, + paths: &'a [P], + document: &'a T, +} + +/// Serializes the envelope in its stable wire order and uses AniCraft's compact/pretty threshold. +pub fn write(generator: &str, paths: &[P], document: &T) -> Result { + let envelope = EnvelopeOut { format: FORMAT_NAME, version: FORMAT_VERSION, generator, paths, document }; + let compact = serde_json::to_string(&envelope)?; + if compact.len() < PRETTY_LIMIT { serde_json::to_string_pretty(&envelope) } else { Ok(compact) } +} + +/// Decodes one native-document payload under the path table that preceded it in the envelope. +pub trait PayloadDecoder

{ + type Document; + + fn decode<'de, D>(&mut self, deserializer: D, paths: Option>) -> Result + where + D: serde::Deserializer<'de>; +} + +/// The parsed top-level format, version and document payload. +#[derive(Debug, PartialEq)] +pub struct Envelope { + pub format: String, + pub version: u32, + /// Missing is distinct from JSON `null`, which the payload decoder decides whether to accept. + pub document: Option, +} + +struct DocumentSeed<'a, P, D> { + decoder: &'a mut D, + paths: Option>, +} + +impl<'de, P, D> DeserializeSeed<'de> for DocumentSeed<'_, P, D> +where + D: PayloadDecoder

, +{ + type Value = D::Document; + + fn deserialize(self, deserializer: De) -> Result + where + De: serde::Deserializer<'de>, + { + self.decoder.decode(deserializer, self.paths) + } +} + +struct EnvelopeSeed<'a, P, D> { + decoder: &'a mut D, + marker: PhantomData P>, +} + +impl<'de, P, D> DeserializeSeed<'de> for EnvelopeSeed<'_, P, D> +where + P: Deserialize<'de> + Clone, + D: PayloadDecoder

, +{ + type Value = Envelope; + + fn deserialize(self, deserializer: De) -> Result + where + De: serde::Deserializer<'de>, + { + struct EnvelopeVisitor<'a, P, D> { + decoder: &'a mut D, + marker: PhantomData P>, + } + + impl<'de, P, D> Visitor<'de> for EnvelopeVisitor<'_, P, D> + where + P: Deserialize<'de> + Clone, + D: PayloadDecoder

, + { + type Value = Envelope; + + fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result { + formatter.write_str("an .anicraft envelope object") + } + + fn visit_map(self, mut map: A) -> Result + where + A: MapAccess<'de>, + { + let (mut format, mut version, mut paths, mut document) = (None, None, None, None); + while let Some(key) = map.next_key::()? { + match key.as_str() { + "format" => format = Some(map.next_value::()?), + "version" => version = Some(map.next_value::()?), + "paths" => paths = Some(map.next_value::>()?), + "document" => { + document = Some(map.next_value_seed(DocumentSeed { decoder: self.decoder, paths: paths.clone() })?); + } + _ => { + map.next_value::()?; + } + } + } + Ok(Envelope { + format: format.ok_or_else(|| serde::de::Error::missing_field("format"))?, + version: version.ok_or_else(|| serde::de::Error::missing_field("version"))?, + document, + }) + } + } + + deserializer.deserialize_map(EnvelopeVisitor { decoder: self.decoder, marker: PhantomData }) + } +} + +/// Reads the envelope while delegating its typed document payload to `decoder`. +/// +/// Unknown top-level fields are ignored. Repeated recognized keys use the last successful value, +/// and each `document` is decoded immediately with the latest preceding `paths` field. A +/// malformed earlier `document` therefore errors before a later duplicate is considered, just as +/// it does with a streaming Serde visitor. +pub fn read<'de, P, D>(json: &'de str, decoder: &mut D) -> Result, serde_json::Error> +where + P: Deserialize<'de> + Clone, + D: PayloadDecoder

, +{ + let mut deserializer = serde_json::Deserializer::from_str(json); + let envelope = EnvelopeSeed { decoder, marker: PhantomData }.deserialize(&mut deserializer)?; + deserializer.end()?; + Ok(envelope) +} + +#[cfg(test)] +mod tests { + use super::*; + use serde::Deserialize; + use serde_json::Value; + + #[derive(Debug, Deserialize, Serialize, PartialEq)] + struct Document { + value: u32, + } + + #[derive(Default)] + struct CaptureDocument { + calls: usize, + } + + impl PayloadDecoder for CaptureDocument { + type Document = (Document, Option>); + + fn decode<'de, D>(&mut self, deserializer: D, paths: Option>) -> Result + where + D: serde::Deserializer<'de>, + { + self.calls += 1; + Ok((Document::deserialize(deserializer)?, paths)) + } + } + + struct ValueDocument; + + impl PayloadDecoder for ValueDocument { + type Document = (Value, Option>); + + fn decode<'de, D>(&mut self, deserializer: D, paths: Option>) -> Result + where + D: serde::Deserializer<'de>, + { + Ok((Value::deserialize(deserializer)?, paths)) + } + } + + #[test] + fn writer_keeps_the_versioned_field_order_and_pretty_threshold() -> Result<(), Box> { + let paths = ["path-a", "path-b"]; + let document = Document { value: 3 }; + let json = write("AniCraft 0.1", &paths, &document)?; + let keys = ["\"format\"", "\"version\"", "\"generator\"", "\"paths\"", "\"document\""]; + let offsets: Vec = keys.iter().map(|key| json.find(key).unwrap_or(usize::MAX)).collect(); + assert!(offsets.windows(2).all(|pair| pair[0] < pair[1])); + assert!(json.contains("\n \"format\""), "small envelopes stay human-readable"); + assert!(json.contains("\"format\": \"anicraft\"")); + assert!(json.contains("\"version\": 3")); + Ok(()) + } + + #[test] + fn repeated_fields_and_paths_before_document_keep_streaming_semantics() -> Result<(), Box> { + let json = r#"{"format":"old","format":"anicraft","version":1,"paths":["first"],"paths":["chosen"],"document":{"value":7},"ignored":true}"#; + let mut decoder = CaptureDocument::default(); + let envelope = read::(json, &mut decoder)?; + assert_eq!(envelope.format, "anicraft"); + assert_eq!(envelope.version, 1); + assert_eq!(envelope.document, Some((Document { value: 7 }, Some(vec!["chosen".to_string()])))); + assert_eq!(decoder.calls, 1); + Ok(()) + } + + #[test] + fn paths_after_document_do_not_retroactively_change_its_scope() -> Result<(), Box> { + let json = r#"{"format":"anicraft","version":1,"document":{"value":1},"paths":["later"]}"#; + let mut decoder = ValueDocument; + let envelope = read::(json, &mut decoder)?; + assert_eq!(envelope.document, Some((serde_json::json!({"value":1}), None))); + Ok(()) + } + + #[test] + fn malformed_earlier_duplicate_document_stops_before_later_document() { + let json = r#"{"format":"anicraft","version":1,"document":null,"document":{"value":2}}"#; + let mut decoder = CaptureDocument::default(); + let result = read::(json, &mut decoder); + assert!(result.is_err()); + assert_eq!(decoder.calls, 1); + } + + #[test] + fn malformed_top_level_keeps_the_native_envelope_diagnostic() { + let mut decoder = CaptureDocument::default(); + let error = read::("[]", &mut decoder).unwrap_err(); + assert!(error.to_string().contains("expected an .anicraft envelope object")); + assert_eq!(decoder.calls, 0); + } + + #[test] + fn duplicate_documents_use_the_last_successful_value() -> Result<(), Box> { + let json = r#"{"format":"anicraft","version":1,"document":{"value":1},"document":{"value":2}}"#; + let mut decoder = CaptureDocument::default(); + let envelope = read::(json, &mut decoder)?; + assert_eq!(envelope.document, Some((Document { value: 2 }, None))); + assert_eq!(decoder.calls, 2); + Ok(()) + } +} diff --git a/crates/audio/Cargo.toml b/crates/audio/Cargo.toml index 822c8a8ad..fe28cec3f 100644 --- a/crates/audio/Cargo.toml +++ b/crates/audio/Cargo.toml @@ -1,6 +1,6 @@ [package] name = "craft-audio" -description = "Audio file I/O and codecs: WAV/AIFF read and write, FLAC encoder, AAC and ALAC decoders and encoders, AC-3/E-AC-3 decoder and encoder, MP3 decoder and encoder, Opus and MP1/MP2 decoders." +description = "Audio file I/O and codecs: WAV/AIFF read and write, FLAC encoder, AAC and ALAC decoders and encoders, AC-3/E-AC-3 decoder and encoder, MP3 decoder and encoder, Opus decoder and encoder, Vorbis encoder, MP1/MP2 decoders." version.workspace = true edition.workspace = true license.workspace = true @@ -19,28 +19,41 @@ workspace = true # µ-law and free-format MP3 through symphonia. decode = ["decode-native", "symphonia"] # `decode` without symphonia: only the formats with a native decoder (others are Unsupported). -decode-native = ["aac-decode", "alac-decode", "opus-decode", "ac3-decode", "mpa-decode", "mp3-decode", "craft-media/isobmff", "craft-media/matroska", "craft-media/ogg", "dep:log"] +decode-native = ["aac-decode", "alac-decode", "opus-decode", "ac3-decode", "mpa-decode", "mp3-decode", "craft-media/isobmff", "craft-media/matroska", "craft-media/ogg", "dep:log", "dep:craft-xml"] # The symphonia fallback for the formats with no native decoder yet. symphonia = ["decode-native", "dep:symphonia"] -# Every encoder: the lossless file writers and the AAC, ALAC, AC-3/E-AC-3 and MP3 encoders. -encode = ["lossless-encode", "aac-encode", "alac-encode", "ac3-encode", "eac3-encode", "mp3-encode"] +# Every encoder: the lossless file writers and the AAC, ALAC, AC-3/E-AC-3, MP3 and Opus encoders. +# All but Opus are also file formats of `encode` and `FileEncoder`. +encode = ["lossless-encode", "aac-encode", "alac-encode", "ac3-encode", "eac3-encode", "mp3-encode", "opus-encode", "vorbis-encode"] # Encode WAV/BWF/RF64, AIFF/AIFF-C and FLAC files. lossless-encode = ["pcm-encode", "flac-encode"] # Encode WAV/BWF/RF64 and AIFF/AIFF-C files. pcm-encode = [] +# Decode FLAC files and frames natively (`flac::decode`, `flac::Decoder`). +flac-decode = [] # Encode FLAC files. flac-encode = ["craft-media/bitstream"] +# The pure-Rust Vorbis encoder (`vorbis::encoder`) and Ogg Vorbis files through `encode` and +# `encode_ogg_vorbis`. +vorbis-encode = ["craft-media/ogg"] # Codec modules on their own, for packet-level use. aac-decode = ["craft-media/bitstream"] -aac-encode = ["craft-media/bitstream"] +# The AAC encoder (`aac::Encoder`), and AAC in M4A through `encode` and `FileEncoder`. +aac-encode = ["craft-media/bitstream", "craft-media/isobmff"] opus-decode = [] +# QOA (Quite OK Audio) files: decoder and encoder in the `qoa` module, each on its own. +qoa = ["qoa-decode", "qoa-encode"] +qoa-decode = [] +qoa-encode = [] +opus-encode = [] ac3-decode = ["craft-media/bitstream"] mpa-decode = ["craft-media/bitstream"] # MPEG-1/2 Layer III (MP3) decoder (`mp3::Decoder`). mp3-decode = ["mpa-decode"] -# ALAC packet decoder and encoder (`alac`), ported from Apple's reference implementation. +# ALAC packet decoder and encoder (`alac`), ported from Apple's reference implementation. The +# encoder also writes ALAC in M4A through `encode` and `FileEncoder`. alac-decode = [] -alac-encode = [] +alac-encode = ["craft-media/isobmff"] ac3-encode = ["craft-media/bitstream"] eac3-encode = ["craft-media/bitstream"] # MPEG-1/2 Layer III (MP3) encoder (`mp3`). @@ -54,6 +67,7 @@ serde = ["dep:serde", "craft-media/serde"] [dependencies] craft-media = { workspace = true } +craft-xml = { workspace = true, optional = true } thiserror = { workspace = true } serde = { workspace = true, optional = true } log = { version = "0.4", optional = true } @@ -74,6 +88,14 @@ required-features = ["decode", "lossless-encode", "peaks"] name = "codecs" required-features = ["decode", "encode"] +[[test]] +name = "flac_decode" +required-features = ["flac-decode", "flac-encode"] + +[[test]] +name = "qoa_oracle" +required-features = ["qoa"] + [[example]] name = "info" required-features = ["decode"] diff --git a/crates/audio/README.md b/crates/audio/README.md index 30f09ea57..7469a9401 100644 --- a/crates/audio/README.md +++ b/crates/audio/README.md @@ -1,7 +1,8 @@ # craft-audio Audio file I/O and codecs for the Craft apps. Native implementations: WAV and AIFF read/write, -FLAC encoder, AAC and ALAC decoders/encoders, AC-3/E-AC-3 decoder/encoder, MP3 decoder/encoder, Opus and MP1/MP2 decoders. +FLAC encoder, AAC and ALAC decoders/encoders, AC-3/E-AC-3 decoder/encoder, MP3 decoder/encoder, Opus +decoder/encoder, MP1/MP2 decoders. Formats without a native decoder are decoded by [symphonia](#symphonia), which is to be replaced by native decoders. @@ -28,19 +29,21 @@ No default features. The shared types (`AudioBuffer`, `AudioInfo`, `ChannelLayou | `decode` | `decode`, `probe`, `Decoder`, `tracks`, `TrackDecoder`; every format in [Format support](#format-support) | `decode-native`, `symphonia` | | `decode-native` | the same, native formats only (others are `Unsupported`); `wav_pcm`, `aiff_pcm` (a file's integer PCM bit for bit, as `RawPcm`) | the six decoders below, craft-media `isobmff`/`matroska`/`ogg`, `log` | | `symphonia` | the fallback for formats with no native decoder; turning it on is the same as `decode` | `decode-native`, symphonia | -| `encode` | every encoder: the WAV, AIFF and FLAC file writers, and the AAC, ALAC, AC-3/E-AC-3 and MP3 encoders | `lossless-encode`, `aac-encode`, `alac-encode`, `ac3-encode`, `eac3-encode`, `mp3-encode` | +| `encode` | every encoder, and every format of `encode` and `FileEncoder` | `lossless-encode`, `aac-encode`, `alac-encode`, `ac3-encode`, `eac3-encode`, `mp3-encode`, `opus-encode`, `vorbis-encode` | | `lossless-encode` | WAV, AIFF and FLAC writing | `pcm-encode`, `flac-encode` | | `pcm-encode` | `encode`, `FileEncoder` for WAV and AIFF; `wav_float_header`; `pcm_wav_file`, `pcm_wav_header`, `pcm_aiff_file`, `pcm_aiff_header` (minimal files around integer PCM bytes, for essence embedded in AAF / OMF) | none | +| `flac-decode` | `flac::decode` (whole `.flac` files) and `flac::Decoder` (single frames, for callers that demux FLAC themselves); native, CRC-checked, 4 to 32 bits. `decode` still uses symphonia for FLAC files | none | | `flac-encode` | `encode`, `FileEncoder` for FLAC; `flac::Encoder`, the packet encoder for callers that mux FLAC themselves | craft-media `bitstream` | -| `aac-decode`, `aac-encode` | `aac::Decoder`, `aac::Encoder`; with `aac-decode`, `aac::latm_config` / `latm_split` for LOAS / LATM frames (AAC in MPEG transport streams) | craft-media `bitstream` | +| `aac-decode`, `aac-encode` | `aac::Decoder`, `aac::Encoder`; with `aac-decode`, `aac::latm_config` / `latm_split` for LOAS / LATM frames (AAC in MPEG transport streams); with `aac-encode`, `encode`, `FileEncoder` for AAC in M4A | craft-media `bitstream`; `aac-encode` also `isobmff` | | `alac-decode` | `alac::Decoder` | none | -| `alac-encode` | `alac::Encoder` | none | -| `opus-decode` | `opus::Decoder`/`StreamDecoder` | none | +| `alac-encode` | `alac::Encoder`; `encode`, `FileEncoder` for ALAC in M4A | craft-media `isobmff` | +| `opus-decode`, `opus-encode` | `opus::Decoder`/`StreamDecoder`, `opus::Encoder` | none | +| `vorbis-encode` | `vorbis::encoder::Encoder` (pure Rust); `encode_ogg_vorbis`, `encode_ogg` and `encode` for Ogg Vorbis files (whole buffer; not `FileEncoder`) | craft-media `ogg` | | `ac3-decode` | `ac3::Decoder`, `ac3::ProgrammeDecoder` | craft-media `bitstream` | -| `ac3-encode`, `eac3-encode` | `ac3::encoder::Encoder` | craft-media `bitstream` | +| `ac3-encode`, `eac3-encode` | `ac3::encoder::Encoder`; `encode`, `FileEncoder` for AC-3 and E-AC-3 streams | craft-media `bitstream` | | `mpa-decode` | `mpa::Decoder` | craft-media `bitstream` | | `mp3-decode` | `mp3::Decoder`, `mp3::InfoTag` | `mpa-decode` | -| `mp3-encode` | `mp3::Encoder` | craft-media `bitstream` | +| `mp3-encode` | `mp3::Encoder`; `encode`, `FileEncoder` for MP3 | craft-media `bitstream` | | `peaks` | `peaks::Peaks` | none | | `rayon` | parallel AAC encoding (with `aac-encode`) | `rayon` | | `serde` | `Serialize`/`Deserialize` on the shared types | `serde` | @@ -229,17 +232,86 @@ let wav = encode(&surround, &EncodeOptions::default().with_layout(ChannelLayout: let mut enc = FileEncoder::new(std::fs::File::create("take.wav")?, &EncodeOptions::default(), 48_000, 2)?; enc.write(&buf.channels, buf.frames())?; // one slice (or Vec) per channel enc.finish()?; + +// The coded formats go the same way: AAC in M4A at 192 kbit/s, 5.1 as E-AC-3, 16-bit ALAC. +let m4a = encode(&buf, &EncodeOptions::new(EncodeFormat::Aac).with_bitrate(192_000))?; +let eac3 = encode(&surround, &EncodeOptions::new(EncodeFormat::Eac3).with_layout(ChannelLayout::SURROUND_5_1_SIDE))?; +let alac = encode(&buf, &EncodeOptions::new(EncodeFormat::Alac).with_bit_depth(BitDepth::Int16))?; ``` `EncodeOptions::default()` is 24-bit WAV. `with_bwf` adds a BWF `bext` chunk and `with_force_rf64` writes RF64 below 4 GiB; both apply to WAV only. `encode_format_for("aiff")` -maps an extension to an `EncodeFormat`, and `ENCODE_EXTENSIONS` lists the extensions the enabled -features can write. `EncodeFormat::ALL` lists the formats; `extension()` is the extension a new -file gets and `mime_type()` its media type. `write` refuses a block without one slice per channel, or with a slice -shorter than `frames`, and writes nothing. After a write fails for another reason, every later +maps an extension to an `EncodeFormat` (`"m4a"` to AAC; `"alac"` names ALAC), and `ENCODE_EXTENSIONS` lists the extensions the enabled +features can write. `EncodeFormat::ALL` lists the formats and `is_built()` says whether this build +writes one; `extension()` is the extension a new file gets and `mime_type()` its media type. +`write` refuses a block without one slice per channel, or with a slice shorter than `frames`, and +writes nothing. After a write fails for another reason, every later `write` and `finish` returns an error. -### Encode AAC +| `EncodeFormat` | File | Audio | Options used | +|---|---|---|---| +| `Wav`, `Aiff`, `Flac` | WAV / BWF / RF64, AIFF / AIFF-C, FLAC | any rate; FLAC up to 8 channels | `bit_depth`, `dither`, `layout` (WAV, FLAC), `bwf`, `force_rf64` (WAV) | +| `Alac` | M4A | 1–8 channels, any rate | `bit_depth` (16, 24, 32), `dither`, `layout` | +| `Aac` | M4A | AAC-LC, 1–8 channels, the 13 AAC rates (7350–96000 Hz) | `rate_control`, `layout` | +| `Mp3` | `.mp3` stream | mono or stereo; 32, 44.1, 48 kHz or 16, 22.05, 24 kHz | `rate_control` (bit rate), `layout` | +| `Ac3`, `Eac3` | `.ac3` / `.eac3` stream | up to 5.1 (E-AC-3: 7.1); 32, 44.1, 48 kHz | `rate_control` (bit rate), `layout` | + +`with_bitrate(bits_per_second)` sets a constant bit rate for all channels together. A rate the +stream cannot carry is moved to the nearest one it can: MP3 and AC-3 use the highest rate of their +tables not above it, AAC and E-AC-3 keep it inside their range (AAC from 6 kbit/s or a quarter of +the sample rate per channel, whichever is more: from about half of that down the encoder codes +silence). `with_quality(0.0..=1.0)` is AAC's +constant quality (VBR); the other lossy formats return `Unsupported` for it. Without either, AAC +is written at 64 kbit/s per channel, MP3 at 96 kbit/s per channel, and AC-3 and E-AC-3 at +96 kbit/s per full-bandwidth channel (the LFE not counted). + +The audio is in WAV channel order for every format; `layout` (the usual layout for the channel +count when unset) names the speakers, and the channels are reordered into the codec's own order. +Each coded format has a fixed set of layouts, and audio with other speakers is `Unsupported` +rather than written to the wrong ones: + +- AAC: C; L R; L R C; L R C Cb; L R C Ls Rs; 5.1; 6.1; and for 8 channels L R C LFE Lb Rb Lc Rc + (a front pair left and right of centre, mask `0xFF`). ALAC: the same, with 6.1's back centre last. + WAV 7.1 (backs and sides) is not that 7.1, and quad (the usual 4-channel layout) has no centre: + both need another format. +- AC-3 and E-AC-3: the layout picks the coding mode (1/0 to 3/2, with or without LFE; E-AC-3 also + 7.1 with backs and sides). +- A surround pair is a surround pair: side and back surrounds are interchangeable where the format + has one pair. + +Before a file is created, the options and the format say what will happen: + +```rust +let opts = EncodeOptions::new(EncodeFormat::Ac3).with_layout(info.layout).with_bitrate(200_000); +opts.check(info.sample_rate, info.channels)?; // the error FileEncoder::new would return, nothing written +let bps = opts.bitrate(info.sample_rate, info.channels)?; // Some(192_000): the rate actually used; None when lossless or by quality + +EncodeFormat::Ac3.sample_rates(); // Some(&[32_000, 44_100, 48_000]); None when any rate will do +EncodeFormat::Ac3.max_channels(); // 6 +EncodeFormat::Ac3.fixed_layouts(3); // L R C, L R Cb, L R LFE; None for WAV, AIFF and FLAC (any layout) +EncodeFormat::Ac3.supports_layout(&info.layout); // one of those, sides and backs standing for each other +EncodeFormat::Ac3.supports_quality(); // false: AAC only +``` + +`FileEncoder::bitrate()` gives the rate of a file being written. The `EncodeFormat` queries answer +for a format whose feature is off too; `check` and `EncodeOptions::bitrate` (there with any +encoder feature) return `Unsupported` for it. + +The layout `decode` reports (`AudioInfo::layout`) can be passed to `with_layout` as it is. When a +coded format refuses it (2.1 or quad into AAC, a single channel tagged as left, channels with no +speaker assignment), the choices are another format, or audio rearranged into one of the layouts +above with that layout set; leaving `layout` unset declares the usual layout for the count. This +crate's decoder returns 7- and 8-channel AAC without a speaker assignment, in AAC order, so such a +file needs that step before it is written again. + +An M4A's edit list skips the AAC encoder's priming (1024 frames) and ends at the last frame +written, so the file decodes to the length that went in. An MP3, AC-3 or E-AC-3 stream has no +place for that (see [Limits](#limits)). For a container of your own, use the codec modules below. + +### Encode AAC packets + +For an `.m4a` file, use `EncodeFormat::Aac` above. `aac::Encoder` is the packet encoder behind it, +for muxing into a container of your own: ```rust use craft_audio::AudioError; @@ -264,6 +336,65 @@ Ls Rs) has no equivalent in that layout; passed unchanged, its channels go to th are a playable `.aac` file. To write MP4 or Matroska, pass the access units to craft-media's muxers ([MP4](../media/README.md#muxing), [Matroska](../media/README.md#muxing-1)). +### Encode Opus + +```rust +use craft_audio::opus::Encoder; + +let mut enc = Encoder::new(2, 128_000)?; // channels, bits per second +let frame_len = Encoder::FRAME_SIZE * 2; // 20 ms, interleaved stereo +let mut samples = buf.interleaved(); // 48 kHz audio +// Silence after the audio flushes the encoder's delay (the pre-skip) and fills the last frame. +let frames = samples.len() / 2 + enc.pre_skip() as usize; +samples.resize(frames.div_ceil(Encoder::FRAME_SIZE) * frame_len, 0.0); +let mut packets = Vec::new(); +for frame in samples.chunks_exact(frame_len) { + packets.push(enc.encode_float(frame)?); +} +let head = enc.opus_head(); // Matroska's CodecPrivate +let pre_skip = enc.pre_skip(); // samples the player drops at the start +``` + +The encoder takes 48 kHz only, and craft-audio has no resampler. + +Opus goes into Ogg, Matroska/WebM or MP4: + +- **Ogg (`.opus`):** `craft_media::ogg::PageWriter` with `head` and `opus_tags` as the two header + packets, each packet's granule position the samples decoded so far (pre-skip included, as + the input above starts with it), and `finish(Some(pre_skip + length))` to trim the end + (RFC 7845 §4). + +- **Matroska/WebM:** `head` is the track's `CodecPrivate`, and the pre-skip goes in + `TrackSpec::codec_delay_ns` (in nanoseconds at 48 kHz). The Matroska muxer cannot write + `DiscardPadding`, so the padding at the end stays in the file. +- **MP4:** the sample entry takes an `isobmff::OpusConfig` (`dOps`) with the same fields as the + `OpusHead`: channels, pre-skip, input rate, gain and mapping family. `TrackConfig::media_start` + set to the pre-skip skips it. To drop the padding as well, give `TrackConfig::edits` one + `Edit` instead, which replaces the edit `media_start` makes: `media_time` the pre-skip (track + timescale), `segment_duration` the audio's length (movie timescale), `media_rate` + `0x0001_0000`. + +### Encode Ogg Vorbis + +```rust +use craft_audio::{OggVorbisOptions, encode_ogg_vorbis}; + +// 1–8 channels in Vorbis order (5.1: L C R Ls Rs LFE), any sample rate. +let opts = OggVorbisOptions::new(6.0).with_tag("TITLE", "Take 3").with_tag("ARTIST", "Me"); +let ogg: Vec = encode_ogg_vorbis(&buf, &opts)?; +``` + +The encoder is pure Rust (`vorbis::encoder`, from FilmCraft #4481) and builds for wasm32. Quality +runs from -1 to 10 at a variable bit rate: about 18 kbit/s per channel at -1, 100 at 6 and 140 +at 10 on music. `encode` with `EncodeFormat::Ogg` takes a WAV-order layout and writes it in +Vorbis order; its `RateControl::Bitrate` (64–500 kbit/s) picks the quality whose nominal rate +that is (`quality_for_bitrate`), and `RateControl::Quality` 0.0–1.0 is quality -1 to 10. For +streaming, feed `vorbis::encoder::Encoder` and write its packets with `craft_media::ogg::PageWriter` +(identification, `vorbis_comment`, setup, then the audio with each packet's granule position). + +Not yet: short blocks (transients may pre-echo), channel coupling (stereo costs +about twice mono) and bit-rate management. + ### Waveform overviews ```rust @@ -308,13 +439,14 @@ match craft_audio::decode(&bytes, None) { | WAV, BWF, RF64/BW64 | native | native | | AIFF, AIFF-C | native | native | | FLAC | symphonia | native | -| AAC-LC | native | native | -| HE-AAC v1/v2 | native, core only (see [Limits](#limits)) | — | -| ALAC | native | native, packets (`alac::Encoder`; no file writer) | +| AAC-LC | native | native; M4A through `encode` / `FileEncoder`, packets through `aac::Encoder` | +| HE-AAC v1 | native (SBR) | — | +| HE-AAC v2 | native, without parametric stereo (see [Limits](#limits)) | — | +| ALAC | native | native; M4A through `encode` / `FileEncoder`, packets through `alac::Encoder` | | MP1, MP2 (MPEG audio Layer I/II) | native | — | -| MP3 (Layer III) | native | native, frames (`mp3::Encoder`; the frames in a row are an `.mp3` file) | -| Opus | native | — | -| AC-3, E-AC-3 | native | native packet encoder | +| MP3 (Layer III) | native | native; `.mp3` through `encode` / `FileEncoder`, frames through `mp3::Encoder` | +| Opus | native | native, packets (`opus::Encoder`; no file writer) | +| AC-3, E-AC-3 | native | native; `.ac3` / `.eac3` through `encode` / `FileEncoder`, packets through `ac3::encoder::Encoder` | | Vorbis, ADPCM, A-law, µ-law | symphonia | — | | Container | Demuxer | Timing | @@ -329,10 +461,10 @@ match craft_audio::decode(&bytes, None) { ## Limits -- HE-AAC: the AAC-LC core is decoded. `decode`, `Decoder` and `TrackDecoder` upsample it 2×; - `aac::Decoder` alone outputs the core rate. SBR is not reconstructed, so there is no content - above the core's Nyquist frequency. -- HE-AAC v2: parametric stereo is not reconstructed either, so the mono core decodes as mono. +- HE-AAC: `decode`, `Decoder` and `TrackDecoder` decode SBR at twice the core rate. + `aac::Decoder` outputs the core rate unless `set_sbr(true)` is called. Downsampled SBR (an + output rate equal to the core rate) plays the core alone. +- HE-AAC v2: parametric stereo is not reconstructed, so the mono core decodes as mono, with SBR. - AAC: Main, SSR and LTP profiles, coupling channels and 960-sample frames are unsupported. - Raw ADTS: frames with several raw data blocks are unsupported: a stream of them is refused, and one among other frames decodes as silence of its length and counts as damaged, as does a frame @@ -374,10 +506,26 @@ match craft_audio::decode(&bytes, None) { - `TrackDecoder` returns `Unsupported` for FLAC and Vorbis; `decode` and `Decoder` read those through symphonia. ALAC in CAF still goes through symphonia (craft-media has no CAF demuxer). -- ALAC: `EncodeFormat` has no ALAC entry; `alac::Encoder` writes packets and the configuration, - and the caller muxes them (MP4 `alac` box, Matroska `A_ALAC`). craft-media's MP4 muxer writes no - `chan` box, so an MP4 of more than two channels carries no explicit layout; decoders then use the - standard ALAC layout for the count. +- ALAC: craft-media's MP4 muxer writes no `chan` box, so an M4A of more than two channels (from + `EncodeFormat::Alac` or muxed from `alac::Encoder` packets) carries no explicit layout; decoders + then use the standard ALAC layout for the count, which is the one written. +- `EncodeFormat::Mp3`, `Ac3` and `Eac3` write raw streams, which cannot signal the codec's delay or + the length: the file decodes with 1057 (MP3) or 256 (AC-3, E-AC-3) frames of delay in front and + padded to whole frames at the end. MP3 gets no Xing / Info frame. `EncodeFormat::Aac` and `Alac` + (M4A) decode to the length written. +- An AAC M4A's length is in the edit list and the last packet's duration. This crate, Apple's + decoders and ffmpeg 9 end the audio there; ffmpeg 8 skips the priming but plays the last + packet's padding (under 1024 frames of near silence). +- `EncodeFormat::Aac` and `Alac` write the `moov` box after the audio (no fast start), and no + metadata. Raw ADTS AAC is not a file format of `encode`; `aac::EncoderConfig::with_adts` gives + the frames. +- This crate's file reader plays a raw E-AC-3 7.1 stream as its 5.1 independent substream. +- `EncodeFormat::Aac` with 8 channels is AAC channel configuration 7 (C, Lc Rc, L R, Ls Rs, LFE), + which Apple's decoders read as written. ffmpeg assumes such a stream is a mislabelled 7.1 with + backs and sides and moves the channels accordingly, unless run with `-strict 1`. +- A file with no audio: MP3, AC-3 and E-AC-3 are then zero bytes, which nothing can probe. An + empty AAC M4A (a track with no samples) is read by ffmpeg; this crate's native reader passes it + on, so `decode-native` alone refuses it and symphonia reads only some as zero frames. - ALAC 32-bit streams: ffmpeg does not decode their uncompressed (escaped) stereo frames, which Apple's encoder and this one write for audio that does not compress; Apple's decoder and this one do. @@ -413,8 +561,8 @@ Each codec module's rustdoc lists the coding tools it implements. |---|---|---|---| | WAV, AIFF | Microsoft RIFF, Apple AIFF/AIFF-C | read: integer 8–32-bit, float 32/64-bit, `WAVE_FORMAT_EXTENSIBLE` masks, RF64/BW64, BWF `bext`, truncated files; write: 16/24/32-bit integer and 32-bit float, RF64 above 4 GiB, any layout, TPDF dither | round trips; hostile-input tests | | FLAC encoder | RFC 9639 | 16, 24, 32 bits; 1–8 channels, layouts decoders would not assume for the count as a `WAVEFORMATEXTENSIBLE_CHANNEL_MASK` comment; up to 655,350 Hz; LPC, stereo decorrelation, wasted bits, STREAMINFO MD5 | bit-exact through ffmpeg, STREAMINFO MD5 checked against ffmpeg | -| AAC (`aac`) | ISO/IEC 14496-3 | AAC-LC decode and encode (CBR, VBR, ADTS), 1–8 channels; HE-AAC: see [Limits](#limits) | ffmpeg: ~135 dB SNR; PNS noise differs per sample at equal band energy | -| Opus (`opus`) | RFC 6716, 8251, 7845 | decoder: SILK, hybrid, CELT, all frame sizes, PLC and FEC, multistream (`Decoder`) and single stream (`StreamDecoder`), 8–48 kHz output | all 36 RFC 8251 conformance cases | +| AAC (`aac`) | ISO/IEC 14496-3; SBR tables from ETSI ES 201 980 Annex I | AAC-LC decode and encode (CBR, VBR, ADTS), 1–8 channels; HE-AAC (SBR) decode, see [Limits](#limits) | ffmpeg: ~135 dB SNR; PNS noise differs per sample at equal band energy. HE-AAC from AudioToolbox, 1–6 channels: same 962-sample delay as ffmpeg, ~54 dB SNR, high band level within 0.01 dB | +| Opus (`opus`) | RFC 6716, 8251, 7845 | decoder: SILK, hybrid, CELT, all frame sizes, PLC and FEC, multistream (`Decoder`) and single stream (`StreamDecoder`), 8–48 kHz output; encoder (`Encoder`): SILK, hybrid, CELT, 48 kHz, 20 ms frames, mono or stereo | all 36 RFC 8251 conformance cases; encoder output through ffmpeg's two Opus decoders | | AC-3/E-AC-3 (`ac3`) | decoder: ATSC A/52:2012; encoder/programme API: ETSI TS 102 366 V1.4.1 | single-syncframe and complete-programme decoding; packet encoding through discrete 7.1 (see below) | decoder comparisons and encoder output through ffmpeg, CRC/channel/timing/bit-budget regressions; synthetic quality limitations below | | MP1/MP2 (`mpa`) | ISO/IEC 11172-3, 13818-3 | Layer I and II, MPEG-1/2/2.5, 8–48 kHz; also the Layer II rate/mode pairs the spec disallows (ffmpeg's default mono 384 kbit/s); no free format | ffmpeg-encoded streams, 73–84 dB SNR | | MP3 decoder (`mp3`) | ISO/IEC 11172-3, 13818-3 | Layer III, MPEG-1/2/2.5, 8–48 kHz, every bit rate, CBR and VBR; mono, stereo, dual channel, joint stereo (mid/side and intensity); long, short and mixed blocks; bit reservoir; CRC read, not checked; Xing / Info / VBRI and LAME tags; no free format (read through symphonia) | ffmpeg's float decoder: 127–131 dB SNR on LAME streams of every sampling frequency (CBR, VBR, CRC, no reservoir, transients with short blocks) and on this crate's encoder's; intensity stereo and mixed blocks through LAME frames edited to use them (112–131 dB) | @@ -445,7 +593,7 @@ finite count as silence and the rest are clamped to ±4, so no input makes an in decoder's 7680-bit input buffer. - Not yet: block switching (short blocks; transients pre-echo), `scfsi` and `preflag`, intensity stereo, a tonality-aware model, VBR/ABR, the Xing/Info and LAME gapless tags (the delay is not - signalled), CRC, free format, and an `EncodeFormat` entry for `encode`/`FileEncoder`. + signalled), CRC and free format. ### AC-3 and E-AC-3 encoding @@ -582,6 +730,8 @@ the full tests with ffmpeg and the vectors required. | MP3 decoder | written here, from ISO/IEC 11172-3 and 13818-3; shares the encoder's tables and the MP1/MP2 synthesis filter | | MP3 encoder | FilmCraft `filmcraft-mp3enc` [4bec22b](https://github.com/storytold/filmcraft/commit/4bec22bd03bd1d995c9cf72e02a284f96d3d72cd) | | ALAC | Apple's reference implementation [macosforge/alac@c38887c](https://github.com/macosforge/alac/tree/c38887c5c5e64a4b31108733bd79ca9b2496d987/codec), Copyright © 2011 Apple Inc., Apache-2.0; ported to Rust and modified (see the `alac` module docs and [NOTICE](../../NOTICE)) | +| Opus encoder | EffectCraft `effectcraft-opusenc` [c618429](https://github.com/storytold/effectcraft/commit/c618429252587525a2a6d4b5bccbe4ec5261e1b9) | +| Vorbis encoder | FilmCraft `filmcraft-vorbis` by @JordyUSA ([storytold/filmcraft#4481](https://github.com/storytold/filmcraft/pull/4481)), with partitions of whole sample frames so libvorbis reads surround | Where apps had overlapping implementations, the one that measured better was kept, with the tests of both: @@ -598,7 +748,8 @@ For the apps switching from their own code; to be removed once all three have sw The codec modules' public items keep their names, so apps can re-export them (`pub use craft_audio::aac::*;`). Names kept only for that are `#[doc(hidden)]`: -`ac3::Decoded` (now `ac3::Frame`), `aac::Upsample2x`, `aac::ProgramConfig::parse`, +`ac3::Decoded` (now `ac3::Frame`), `opus::encoder::OpusEncoder` (now `opus::Encoder`), +`aac::Upsample2x`, `aac::ProgramConfig::parse`, `aac::config_layout` and `aac::sample_rate_index`. What else an app adjusts: - `AudioInfo` has `container`, `codec`, `layout` and `damaged` instead of SoundCraft's `format`. @@ -624,8 +775,10 @@ The codec modules' public items keep their names, so apps can re-export them - `TrackInfo::sample_rate` and `channels` are `Option`s. - Features are per codec. SoundCraft's `encode` (WAV, AIFF, FLAC) is now `lossless-encode`, made of `pcm-encode` (WAV, AIFF) and `flac-encode`. `encode` turns on every encoder, including AAC, - ALAC and AC-3/E-AC-3. Writing a format whose feature is off returns `AudioError::Unsupported`. + ALAC, AC-3/E-AC-3 and Opus. Writing a format whose feature is off returns `AudioError::Unsupported`. - AAC encoder settings it cannot use are `aac::Error::InvalidSettings` (`AudioError::Encode`). +- The Opus encoder's constructors and `encode_float` return `Result` instead of panicking on an + invalid channel count, mode/bandwidth or frame length. - `TrackDecoder::decode` returns an error for a corrupt Opus packet, as for the other codecs; call `conceal`. - `FileEncoder::write` takes any per-channel slices, but refuses a block with missing channels or @@ -636,3 +789,8 @@ The codec modules' public items keep their names, so apps can re-export them CAF. MP4 and Matroska files decode to presentation length. A packet that fails is concealed, not dropped. A failed source read is `AudioError::Io`. + +WAV timestamps use `bext` when present, including a zero timestamp. Otherwise a bounded iXML +`BWFXML/BEXT` time reference supplies the timestamp (two unsigned 32-bit words; an omitted high +word is zero). Only the first iXML chunk is considered. Invalid or oversized optional XML is +ignored without rejecting the audio. This works for byte-backed and streaming sources. diff --git a/crates/audio/examples/encode.rs b/crates/audio/examples/encode.rs index 28de0f5ad..51cdecc84 100644 --- a/crates/audio/examples/encode.rs +++ b/crates/audio/examples/encode.rs @@ -1,4 +1,4 @@ -//! Write a tone as WAV and FLAC, whole and streamed, and as AAC packets. +//! Write a tone as WAV, FLAC and AAC in M4A, whole and streamed, and as AAC packets. //! //! `cargo run -p craft-audio --example encode --features lossless-encode,aac-encode -- out` @@ -28,7 +28,11 @@ fn main() -> Result<(), Box> { } enc.finish()?; - // AAC packets for an MP4 or Matroska muxer, with the configuration its sample entry needs. + // AAC in M4A at 160 kbit/s: the same call, with a bit rate in place of a bit depth. + let m4a = encode(&stereo, &EncodeOptions::new(EncodeFormat::Aac).with_bitrate(160_000))?; + std::fs::write(dir.join("tone.m4a"), m4a)?; + + // AAC packets for a muxer of your own, with the configuration its sample entry needs. let mut aac = craft_audio::aac::Encoder::new(craft_audio::aac::EncoderConfig::cbr(rate as u32, 2, 128_000))?; let mut units = aac.encode(&[&tone, &tone]); units.extend(aac.flush()); diff --git a/crates/audio/src/aac/config.rs b/crates/audio/src/aac/config.rs index 88be1e214..3d12b0343 100644 --- a/crates/audio/src/aac/config.rs +++ b/crates/audio/src/aac/config.rs @@ -343,6 +343,51 @@ impl AudioSpecificConfig { } } +/// Why an AudioSpecificConfig prefix cannot be written as an ADTS header. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum AdtsConfigError { + /// Fewer than two bytes were supplied. + Short, + /// The prefix names a profile, indexed rate or channel configuration ADTS cannot carry. + Unsupported { + /// Audio object type from the prefix. + object_type: u8, + /// Sampling frequency index from the prefix. + sf_index: u8, + /// Channel configuration from the prefix. + channel_config: u8, + }, +} + +/// ADTS fixed fields taken from the prefix of an AudioSpecificConfig. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub struct AdtsConfig { + /// Audio object type, 1 through 4. + pub object_type: u8, + /// Indexed sample rate, 0 through 12. + pub sf_index: u8, + /// Channel configuration, 1 through 7. + pub channel_config: u8, +} +impl AdtsConfig { + /// Parse only the first two ASC bytes. Unlike a decoder's full ASC parser, this accepts all + /// four object types representable by ADTS and rejects configurations needing a PCE. + pub fn from_asc_prefix(asc: &[u8]) -> std::result::Result { + let (Some(&a), Some(&b)) = (asc.first(), asc.get(1)) else { return Err(AdtsConfigError::Short) }; + let object_type = a >> 3; + let sf_index = ((a & 7) << 1) | (b >> 7); + let channel_config = (b >> 3) & 15; + if !(1..=4).contains(&object_type) || sf_index > 12 || !(1..=7).contains(&channel_config) { + return Err(AdtsConfigError::Unsupported { object_type, sf_index, channel_config }); + } + Ok(Self { object_type, sf_index, channel_config }) + } + /// Write a no-CRC ADTS header with variable-rate buffer fullness for one payload. + pub fn header(self, payload_len: usize) -> Result<[u8; 7]> { + AdtsHeader::write(self.object_type, self.sf_index, self.channel_config, payload_len, 0x7ff) + } +} + /// A parsed ADTS header. #[non_exhaustive] #[derive(Debug, Clone, Copy, PartialEq, Eq)] @@ -519,6 +564,15 @@ mod tests { assert_eq!(AudioSpecificConfig::parse(&asc.to_bytes().unwrap()).unwrap(), asc); } + #[test] + fn adts_config_preserves_prefix_and_frame_limits() { + let config = AdtsConfig::from_asc_prefix(&[0x12, 0x10]).unwrap(); + assert_eq!(config.header(100).unwrap(), AdtsHeader::write(2, 4, 2, 100, 0x7ff).unwrap()); + assert!(config.header(8185).is_err()); + assert_eq!(AdtsConfig::from_asc_prefix(&[]), Err(AdtsConfigError::Short)); + assert!(matches!(AdtsConfig::from_asc_prefix(&[0x2a, 0x10]), Err(AdtsConfigError::Unsupported { object_type: 5, .. }))); + } + #[test] fn adts_headers_refuse_fields_they_cannot_hold() { for (object_type, sf_index, channel_config) in [(0, 4, 2), (5, 4, 2), (29, 4, 2), (2, 13, 2), (2, 15, 2), (2, 4, 8), (2, 4, 11), (2, 4, 12)] { diff --git a/crates/audio/src/aac/decoder.rs b/crates/audio/src/aac/decoder.rs index 8c4d41300..825ebb23a 100644 --- a/crates/audio/src/aac/decoder.rs +++ b/crates/audio/src/aac/decoder.rs @@ -6,6 +6,7 @@ use crate::aac::config::{AudioSpecificConfig, ElementType, ProgramConfig}; use crate::aac::huffman::{self, INTENSITY_HCB, INTENSITY_HCB2, NOISE_HCB, ZERO_HCB}; use crate::aac::ics::{IcsInfo, WindowSequence}; use crate::aac::mdct::{MdctScratch, WindowShape, mdct_long, mdct_short, window}; +use crate::aac::sbr; use crate::aac::tables::{pow43, pow43_table, sf_gain}; use crate::aac::tns::{self, TnsData}; use crate::aac::{Error, Result}; @@ -63,7 +64,14 @@ impl Scratch { } } -/// AAC-LC decoder. +/// Where the SBR data of a channel element sits in its access unit: the bit range of the fill +/// element's payload after its extension type, and whether a CRC leads it. +type SbrRef = (usize, usize, bool); + +/// An access unit's core output, and the SBR data of each layout entry. +type Decoded = (Vec>, Vec>); + +/// AAC-LC decoder, with the SBR tool of HE-AAC when [`Decoder::set_sbr`] turns it on. #[derive(Clone)] pub struct Decoder { asc: AudioSpecificConfig, @@ -75,6 +83,11 @@ pub struct Decoder { scratch: Scratch, /// A fill element carried SBR data (implicitly signalled HE-AAC). sbr_seen: bool, + /// The SBR decoder of each layout entry, when output is at twice the core rate. + sbr_out: Option>, + /// The layout `sbr_out` was built for. Kept apart from `layout`, so that an access unit whose + /// new program config fails (and whose layout is put back) leaves the SBR history alone. + sbr_layout: Vec<(ElementType, u8)>, } impl Decoder { @@ -110,6 +123,8 @@ impl Decoder { noise_state: NOISE_SEED, scratch: Scratch::new(), sbr_seen: false, + sbr_out: None, + sbr_layout: Vec::new(), }; d.set_layout(layout); Ok(d) @@ -131,16 +146,27 @@ impl Decoder { pub fn config(&self) -> &AudioSpecificConfig { &self.asc } - /// Output sampling rate in Hz: the AAC-LC core rate. + /// Output sampling rate in Hz: the AAC-LC core rate, or twice it with [`Self::set_sbr`] on. pub fn sample_rate(&self) -> u32 { - self.asc.sample_rate + if self.sbr_out.is_some() { self.asc.sample_rate.saturating_mul(2) } else { self.asc.sample_rate } } /// Whether the stream is HE-AAC: SBR signalled in the `AudioSpecificConfig` (explicit) or SBR - /// data seen in a fill element of a decoded access unit (implicit). Only the AAC-LC core is - /// decoded, at [`Self::sample_rate`]; the SBR high band is not reconstructed. + /// data seen in a fill element of a decoded access unit (implicit). Without [`Self::set_sbr`], + /// only the AAC-LC core is decoded. pub fn sbr(&self) -> bool { self.asc.sbr || self.sbr_seen } + /// Decode HE-AAC at twice the core rate (`true`) or the AAC-LC core alone (`false`, the + /// default). With SBR on, [`Self::decode`] returns 2048 samples per channel: the core through + /// the SBR filter banks, delayed by [`SBR_DELAY`](crate::aac::SBR_DELAY) samples, with the high + /// band rebuilt from each frame's SBR data. Until the stream's first SBR header, and for a + /// frame whose SBR data is missing or damaged, the high band is empty. Parametric stereo + /// (HE-AAC v2) is not decoded: its mono core stays mono. + pub fn set_sbr(&mut self, on: bool) { + if on != self.sbr_out.is_some() { + self.sbr_out = on.then(Vec::new); + } + } /// Output channels of the current element layout. pub fn channels(&self) -> usize { self.channels.len() @@ -158,30 +184,61 @@ impl Decoder { c.overlap.fill(0.0); c.prev_shape = WindowShape::Sine; } + for e in self.sbr_out.iter_mut().flatten() { + e.reset(); + } } fn sf_index(&self) -> u8 { self.asc.sf_index } - /// Decode one access unit (one `raw_data_block`) into 1024 samples per channel. An access unit - /// that fails leaves the decoder as it was, but for [`Self::sbr`]: its damage, including a damaged - /// program config element, does not reach the access units after it. + /// Decode one access unit (one `raw_data_block`) into 1024 samples per channel (2048 with + /// [`Self::set_sbr`] on). An access unit that fails leaves the decoder as it was, but for + /// [`Self::sbr`]: its damage, including a damaged program config element, does not reach the + /// access units after it. Damaged SBR data alone does not fail it. pub fn decode(&mut self, au: &[u8]) -> Result>> { let before = (self.asc.pce.clone(), self.layout.clone(), self.offsets.clone(), self.channels.clone(), self.noise_state); - let decoded = self.decode_block(au, true); - if decoded.is_err() { - (self.asc.pce, self.layout, self.offsets, self.channels, self.noise_state) = before; + match self.decode_block(au, true) { + Ok((pcm, sbr_refs)) => Ok(self.apply_sbr(au, pcm, &sbr_refs)), + Err(e) => { + (self.asc.pce, self.layout, self.offsets, self.channels, self.noise_state) = before; + Err(e) + } } - decoded } - fn decode_block(&mut self, au: &[u8], allow_reconfig: bool) -> Result>> { + /// Run each layout entry's channels through its SBR decoder, when SBR is on. + fn apply_sbr(&mut self, au: &[u8], pcm: Vec>, refs: &[Option]) -> Vec> { + let rate = self.sample_rate(); + let Some(elements) = self.sbr_out.as_mut() else { + return pcm; + }; + // a new layout (a program config element) starts the SBR decoders afresh: their history + // belongs to the old element order, even where the channel counts match + if elements.len() != self.layout.len() || self.sbr_layout != self.layout { + *elements = self.layout.iter().map(|l| sbr::Element::new(l.0.channels(), rate)).collect(); + self.sbr_layout.clone_from(&self.layout); + } + let mut out = Vec::with_capacity(pcm.len()); + for (i, e) in elements.iter_mut().enumerate() { + let first = self.offsets.get(i).copied().unwrap_or(0); + let core = pcm.get(first..first + e.channels()).unwrap_or_default(); + let payload = refs.get(i).copied().flatten().map(|(start, end, crc)| sbr::Payload { data: au, start, end, crc }); + out.extend(e.process(core, payload)); + } + out + } + + fn decode_block(&mut self, au: &[u8], allow_reconfig: bool) -> Result { let mut br = BitReader::new(au); let nch = self.channels.len(); let mut saw_element = false; let mut out: Vec>> = vec![None; nch]; let mut used = vec![false; self.layout.len()]; + // SBR data belongs to the single channel or pair element before its fill element + let mut sbr_refs: Vec> = vec![None; self.layout.len()]; + let mut last_slot = None; loop { let id = br.read_bits(3)?; saw_element |= matches!(id, ID_SCE | ID_CPE | ID_LFE); @@ -191,6 +248,7 @@ impl Decoder { let ty = if id == ID_SCE { ElementType::Sce } else { ElementType::Lfe }; let slot = self.find_slot(ty, tag, &used)?; used[slot] = true; + last_slot = (id == ID_SCE).then_some(slot); let mut ics = self.decode_ics(&mut br, None)?; let ch = self.offsets[slot]; self.fill_noise(&mut ics, None); @@ -200,6 +258,7 @@ impl Decoder { let tag = br.read_bits(4)? as u8; let slot = self.find_slot(ElementType::Cpe, tag, &used)?; used[slot] = true; + last_slot = Some(slot); let ch = self.offsets[slot]; let [a, b] = self.decode_cpe(&mut br, ch)?; out[ch] = Some(a); @@ -237,8 +296,13 @@ impl Decoder { } if count > 0 { let ext = br.read_bits(4)?; - self.sbr_seen |= matches!(ext, EXT_SBR_DATA | EXT_SBR_DATA_CRC); + let is_sbr = matches!(ext, EXT_SBR_DATA | EXT_SBR_DATA_CRC); + self.sbr_seen |= is_sbr; + let start = br.position(); br.skip(count * 8 - 4)?; + if let (true, Some(slot)) = (is_sbr, last_slot.take()) { + sbr_refs[slot] = Some((start, br.position(), ext == EXT_SBR_DATA_CRC)); + } } } _ => break, // ID_END @@ -251,7 +315,7 @@ impl Decoder { for o in out { res.push(o.ok_or(Error::Bitstream("access unit is missing a channel element"))?); } - Ok(res) + Ok((res, sbr_refs)) } fn find_slot(&self, ty: ElementType, tag: u8, used: &[bool]) -> Result { @@ -637,6 +701,85 @@ mod tests { assert_eq!(dec.sample_rate(), 22_050); } + /// With SBR on, an access unit decodes to 2048 samples, and over a silent core the high band + /// its SBR fill element describes is there; with SBR off, the core alone (silence). + #[test] + fn sbr_fill_elements_rebuild_the_high_band() { + let au = sbr_au(0); + // AAC-LC, 22.05 kHz, mono + let asc = AudioSpecificConfig::lc(7, 1); + let mut dec = Decoder::from_config(asc.clone()).unwrap(); + let core = dec.decode(&au).unwrap(); + assert_eq!((core[0].len(), dec.sample_rate()), (1024, 22_050)); + assert!(dec.sbr() && core[0].iter().all(|&s| s == 0.0)); + let mut dec = Decoder::from_config(asc).unwrap(); + dec.set_sbr(true); + assert_eq!(dec.sample_rate(), 44_100); + let mut out = Vec::new(); + for _ in 0..4 { + out.extend(dec.decode(&au).unwrap().remove(0)); + } + assert_eq!(out.len(), 4 * 2048); + let rms = (out[2048..].iter().map(|s| s * s).sum::() / (out.len() - 2048) as f32).sqrt(); + assert!(rms > 1e-4, "{rms}"); + } + + /// A program config that changes the layout restarts the SBR decoders, even where the + /// channel counts match, and one in an access unit that fails leaves their history alone: the + /// access units after it decode as if it had not been there. + #[test] + fn a_new_layout_restarts_sbr_and_a_failed_one_keeps_it() { + let au = sbr_au(0); + let one = ProgramConfig { front: vec![(false, 0)], ..Default::default() }; + let six = ProgramConfig { front: vec![(false, 0), (true, 0)], back: vec![(true, 1)], lfe: vec![0], ..Default::default() }; + let decoder = || { + let mut d = Decoder::from_config(AudioSpecificConfig { pce: Some(one.clone()), ..AudioSpecificConfig::lc(7, 0) }).unwrap(); + d.set_sbr(true); + d + }; + let (mut kept, mut failed) = (decoder(), decoder()); + for _ in 0..3 { + assert_eq!(kept.decode(&au).unwrap(), failed.decode(&au).unwrap()); + } + // the new layout's elements are missing from this access unit + assert!(failed.decode(&silent_au(Some(&six), &[(ID_SCE, 0)])).is_err()); + for _ in 0..2 { + assert_eq!(kept.decode(&au).unwrap(), failed.decode(&au).unwrap()); + } + // the same channel count under another tag is a new layout: the SBR decoder starts over, + // as in a decoder that had that layout from the start + let tag1 = ProgramConfig { front: vec![(false, 1)], ..Default::default() }; + let mut fresh = Decoder::from_config(AudioSpecificConfig { pce: Some(tag1.clone()), ..AudioSpecificConfig::lc(7, 0) }).unwrap(); + fresh.set_sbr(true); + kept.set_layout(tag1.layout()); + assert_eq!(kept.decode(&sbr_au(1)).unwrap(), fresh.decode(&sbr_au(1)).unwrap()); + } + + /// One access unit of SCE `tag` with an SBR fill element of [`sbr::test_payload`] (44.1 kHz + /// output). + fn sbr_au(tag: u32) -> Vec { + let payload = sbr::test_payload(44_100, |_| 40, |_| 0).unwrap(); + let mut bw = BitWriter::new(); + bw.write_bits(ID_SCE, 3); + bw.write_bits(tag, 4); + bw.write_bits(0, 8 + 11 + 3); // global_gain, ics_info (max_sfb 0), no pulse/TNS/gain control + bw.write_bits(ID_FIL, 3); + let count = (4 + 8 * payload.len()).div_ceil(8); + if count < 15 { + bw.write_bits(count as u32, 4); + } else { + bw.write_bits(15, 4); + bw.write_bits((count - 14) as u32, 8); + } + bw.write_bits(EXT_SBR_DATA, 4); + for &b in &payload { + bw.write_bits(u32::from(b), 8); + } + bw.write_bits(0, 4); + bw.write_bits(ID_END, 3); + bw.finish() + } + /// An access unit whose elements are all empty: `elements` are (id, tag), then ID_END. fn silent_au(pce: Option<&ProgramConfig>, elements: &[(u32, u32)]) -> Vec { let mut bw = BitWriter::new(); diff --git a/crates/audio/src/aac/encoder/mod.rs b/crates/audio/src/aac/encoder/mod.rs index 5a0ea5ac8..c641b790b 100644 --- a/crates/audio/src/aac/encoder/mod.rs +++ b/crates/audio/src/aac/encoder/mod.rs @@ -162,6 +162,16 @@ const HISTORY: usize = 2048; /// Bits per channel in a silent frame, rounded up: the smallest frame the encoder writes. const MIN_FRAME_BITS_PER_CHANNEL: u64 = 48; +/// The lowest and highest constant bit rate (bits per second) [`Encoder::new`] accepts for a +/// stream without ADTS headers: at least 1 kbit/s per channel and a silent frame, at most the +/// 6144 bits per channel and frame of the decoder buffer. +pub(crate) fn cbr_range(sample_rate: u32, channels: usize) -> (u32, u32) { + let (rate, channels) = (u64::from(sample_rate), channels as u64); + let silent = (MIN_FRAME_BITS_PER_CHANNEL * channels * rate).div_ceil(FRAME_LEN as u64); + let most = 6144 * channels * rate / FRAME_LEN as u64; + (u32::try_from(silent.max(1000 * channels)).unwrap_or(u32::MAX), u32::try_from(most).unwrap_or(u32::MAX)) +} + /// AAC-LC encoder. pub struct Encoder { cfg: EncoderConfig, diff --git a/crates/audio/src/aac/huffman.rs b/crates/audio/src/aac/huffman.rs index adac5c9c9..0c29673e7 100644 --- a/crates/audio/src/aac/huffman.rs +++ b/crates/audio/src/aac/huffman.rs @@ -62,12 +62,12 @@ const SUB: u32 = 1 << 31; /// entry with `SUB` set is `SUB | offset << 5 | width` and selects a sub-table of `1 << width` /// leaves indexed by the next `width` bits. An entry of 0 is a bit pattern that no codeword starts /// with (the codebooks are complete, so this does not occur). -struct Lut { +pub(crate) struct Lut { t: Vec, } impl Lut { - fn build(codes: &[(u32, u8)]) -> Lut { + pub(crate) fn build(codes: &[(u32, u8)]) -> Lut { let mut t = vec![0u32; 1 << L1]; // sub-table width per first-level prefix let mut width = vec![0u32; 1 << L1]; @@ -103,7 +103,7 @@ impl Lut { /// The symbol of the codeword at the reader position; `Eof` if the codeword is cut off. #[inline] - fn decode(&self, br: &mut BitReader) -> Result { + pub(crate) fn decode(&self, br: &mut BitReader) -> Result { let bits = br.peek(32); let mut e = self.t[(bits >> (32 - L1)) as usize]; if e & SUB != 0 { diff --git a/crates/audio/src/aac/mod.rs b/crates/audio/src/aac/mod.rs index 0a5f639a4..fb79ae743 100644 --- a/crates/audio/src/aac/mod.rs +++ b/crates/audio/src/aac/mod.rs @@ -5,13 +5,13 @@ //! model, window switching, M/S, TNS, bit reservoir, CBR/VBR, 1–8 channels. //! - [`Decoder`] (feature `aac-decode`): `AudioSpecificConfig` + raw access units → planar f32. //! Supports M/S, intensity stereo, PNS, TNS, pulse data, both window shapes and every window -//! sequence. +//! sequence, and the SBR tool of HE-AAC ([`Decoder::set_sbr`]). //! - [`latm_config`] / [`latm_split`] (feature `aac-decode`): the configuration and payload of //! LOAS / LATM frames (AAC in MPEG transport streams), for the decoder. //! -//! HE-AAC (SBR/PS), Main/SSR/LTP profiles, coupling channels and 960-sample frames are not -//! supported; for HE-AAC only the AAC-LC core is decoded (at the core rate; [`Decoder::sbr`] says -//! when). +//! Parametric stereo (HE-AAC v2), Main/SSR/LTP profiles, coupling channels and 960-sample frames +//! are not supported. HE-AAC decodes at twice the core rate with [`Decoder::set_sbr`] on +//! ([`Decoder::sbr`] says when a stream uses SBR), and as its AAC-LC core otherwise. //! //! From FilmCraft's `filmcraft-aac` (storytold/filmcraft@8fda2f2). Public items keep their names, //! so FilmCraft can re-export this module. @@ -53,21 +53,31 @@ mod ics; #[cfg(feature = "aac-decode")] mod latm; mod mdct; +#[cfg(feature = "aac-decode")] +mod sbr; mod tables; mod tns; #[cfg(feature = "aac-decode")] pub(crate) use config::split_adts_upto; -pub use config::{AdtsHeader, AudioSpecificConfig, ElementType, ProgramConfig, config_layout, split_adts}; +pub use config::{AdtsConfig, AdtsConfigError, AdtsHeader, AudioSpecificConfig, ElementType, ProgramConfig, config_layout, split_adts}; #[cfg(feature = "aac-decode")] pub use decoder::Decoder; #[cfg(feature = "aac-encode")] +pub(crate) use encoder::cbr_range; +#[cfg(feature = "aac-encode")] pub use encoder::{BitrateMode, Encoder, EncoderConfig, EncoderStats}; #[cfg(feature = "aac-decode")] pub use latm::{LatmConfig, latm_config, latm_split}; pub use mdct::WindowShape; pub use tables::{SAMPLE_RATES, sample_rate_index}; +/// Output samples by which [`Decoder`] with [`Decoder::set_sbr`] on delays the core: the SBR filter +/// banks and the SBR frame's lag behind the core frame. HE-AAC encoders count it in the priming the +/// container signals. +#[cfg(feature = "aac-decode")] +pub const SBR_DELAY: usize = sbr::DELAY; + /// Samples per AAC-LC frame (per channel). pub const FRAME_LEN: usize = 1024; @@ -197,9 +207,9 @@ const UPSAMPLE_DELAY: usize = 481; #[cfg(feature = "aac-decode")] const UPSAMPLE_HISTORY: usize = UPSAMPLE_DELAY + UPSAMPLE_HALF; -/// 2× upsampler for HE-AAC. The decoder reconstructs the AAC-LC core at half the output rate and -/// not yet the SBR high band, so the core is interpolated to the output rate: timing and pitch are -/// right and the band above the core's Nyquist stays empty. Even outputs are the input samples, +/// 2× upsampler: interpolates an AAC-LC core to twice its rate, leaving the band above the core's +/// Nyquist empty. [`Decoder::set_sbr`] decodes HE-AAC with its high band instead; this stays for +/// callers that want the core alone at the output rate. Even outputs are the input samples, /// odd ones a Blackman-windowed sinc half-band interpolation over 2 × `UPSAMPLE_HALF` inputs. /// The output is delayed by `UPSAMPLE_DELAY` input samples, as an SBR decoder's is; the state /// carries across packets, so a seek primed with the preceding packet gives the same samples as diff --git a/crates/audio/src/aac/sbr/bitstream.rs b/crates/audio/src/aac/sbr/bitstream.rs new file mode 100644 index 000000000..0e8e70c3a --- /dev/null +++ b/crates/audio/src/aac/sbr/bitstream.rs @@ -0,0 +1,493 @@ +//! The SBR bitstream (ISO/IEC 14496-3 §4.4.2.8 syntax, §4.5.2.8 semantics): the header and the +//! single and channel pair elements of an `EXT_SBR_DATA` fill element. + +use std::sync::OnceLock; + +use craft_media::bitstream::BitReader; + +use super::freq::FreqTables; +use super::tables::{ + F_ENV_1_5DB, F_ENV_3_0DB, F_ENV_BAL_1_5DB, F_ENV_BAL_3_0DB, T_ENV_1_5DB, T_ENV_3_0DB, T_ENV_BAL_1_5DB, T_ENV_BAL_3_0DB, T_NOISE_3_0DB, + T_NOISE_BAL_3_0DB, +}; +use super::{Header, MAX_ENV}; +use crate::aac::huffman::Lut; +use crate::aac::{Error, Result}; + +/// `bs_frame_class`: fixed leading and trailing frame borders. +pub(crate) const FIXFIX: u8 = 0; +/// Fixed leading border, variable trailing border. +pub(crate) const FIXVAR: u8 = 1; +/// Variable leading border, fixed trailing border. +pub(crate) const VARFIX: u8 = 2; +/// Variable leading and trailing borders. +pub(crate) const VARVAR: u8 = 3; + +/// SBR time slots per frame (`numTimeSlots`; two QMF slots each). +const TIME_SLOTS: usize = 16; + +/// Decoding tables for the SBR Huffman codes, in [`Book`] order. +fn luts() -> &'static [Lut; 10] { + static LUTS: OnceLock<[Lut; 10]> = OnceLock::new(); + LUTS.get_or_init(|| { + [ + Lut::build(&T_ENV_1_5DB), + Lut::build(&F_ENV_1_5DB), + Lut::build(&T_ENV_BAL_1_5DB), + Lut::build(&F_ENV_BAL_1_5DB), + Lut::build(&T_ENV_3_0DB), + Lut::build(&F_ENV_3_0DB), + Lut::build(&T_ENV_BAL_3_0DB), + Lut::build(&F_ENV_BAL_3_0DB), + Lut::build(&T_NOISE_3_0DB), + Lut::build(&T_NOISE_BAL_3_0DB), + ] + }) +} + +/// One of the SBR Huffman codes, and its largest absolute value. +#[derive(Clone, Copy)] +struct Book(usize, i32); + +const T_ENV_15: Book = Book(0, 60); +const F_ENV_15: Book = Book(1, 60); +const T_BAL_15: Book = Book(2, 24); +const F_BAL_15: Book = Book(3, 24); +const T_ENV_30: Book = Book(4, 31); +const F_ENV_30: Book = Book(5, 31); +const T_BAL_30: Book = Book(6, 12); +const F_BAL_30: Book = Book(7, 12); +const T_NOISE: Book = Book(8, 31); +const T_NOISE_BAL: Book = Book(9, 12); + +fn huff(br: &mut BitReader, book: Book) -> Result { + Ok(luts()[book.0].decode(br)? as i32 - book.1) +} + +fn bits(br: &mut BitReader, n: u32) -> Result { + Ok(br.read_bits(n)? as usize) +} + +impl Header { + /// `sbr_header()`, with the defaults of the fields its extra parts leave out. + pub(crate) fn read(br: &mut BitReader) -> Result

{ + let amp_res = br.read_flag()?; + let start_freq = br.read_bits(4)? as u8; + let stop_freq = br.read_bits(4)? as u8; + let xover_band = br.read_bits(3)? as u8; + br.skip(2)?; // bs_reserved + let (extra_1, extra_2) = (br.read_flag()?, br.read_flag()?); + let mut h = Header { + amp_res, + start_freq, + stop_freq, + xover_band, + freq_scale: 2, + alter_scale: true, + noise_bands: 2, + limiter_bands: 2, + limiter_gains: 2, + interpol_freq: true, + smoothing_mode: true, + }; + if extra_1 { + h.freq_scale = br.read_bits(2)? as u8; + h.alter_scale = br.read_flag()?; + h.noise_bands = br.read_bits(2)? as u8; + } + if extra_2 { + h.limiter_bands = br.read_bits(2)? as u8; + h.limiter_gains = br.read_bits(2)? as u8; + h.interpol_freq = br.read_flag()?; + h.smoothing_mode = br.read_flag()?; + } + Ok(h) + } +} + +/// The time/frequency grid of one channel in one frame (`sbr_grid()`, §4.6.18.3.3). +#[derive(Clone, Debug, PartialEq, Eq)] +pub(crate) struct Grid { + pub(crate) class: u8, + /// Envelopes (1 to 5). + pub(crate) num_env: usize, + /// Envelope borders in time slots; `t_env[num_env]` may reach past the frame (up to 19). + pub(crate) t_env: [usize; MAX_ENV + 1], + /// Noise floors (1 or 2). + pub(crate) num_noise: usize, + /// Noise floor borders. + pub(crate) t_noise: [usize; 3], + /// High frequency resolution per envelope. + pub(crate) freq_res: [bool; MAX_ENV], + /// The transient envelope `l_A`, if any (it may be `num_env`: the next frame's first). + pub(crate) la: Option, + /// 3 dB envelope steps (the header's choice, but 1.5 dB for a single FIXFIX envelope). + pub(crate) amp_res: bool, +} + +/// `ceil(log2(num_env + 1))`: the width of `bs_pointer`. +fn pointer_bits(num_env: usize) -> u32 { + usize::BITS - num_env.leading_zeros() +} + +fn read_rel(br: &mut BitReader, n: usize) -> Result<[usize; 3]> { + let mut rel = [0; 3]; + for r in rel.iter_mut().take(n) { + *r = 2 * bits(br, 2)? + 2; + } + Ok(rel) +} + +impl Grid { + fn read(br: &mut BitReader, amp_res: bool) -> Result { + let bad = || Error::Bitstream("invalid SBR time grid"); + let class = br.read_bits(2)? as u8; + let mut t = [0i64; MAX_ENV + 1]; + let mut freq_res = [false; MAX_ENV]; + let (n, pointer) = match class { + FIXFIX => { + let n = 1usize << br.read_bits(2)?; + if n > 4 { + return Err(bad()); + } + freq_res = [br.read_flag()?; MAX_ENV]; + for (l, t) in t.iter_mut().enumerate().take(n + 1) { + *t = (l * TIME_SLOTS / n) as i64; + } + (n, 0) + } + FIXVAR | VARFIX => { + let abs = bits(br, 2)?; + let rels = bits(br, 2)?; + let n = rels + 1; + let rel = read_rel(br, rels)?; + let pointer = bits(br, pointer_bits(n))?; + for l in 0..n { + freq_res[if class == FIXVAR { n - 1 - l } else { l }] = br.read_flag()?; + } + if class == FIXVAR { + t[n] = (abs + TIME_SLOTS) as i64; + for l in (1..n).rev() { + t[l] = t[l + 1] - rel[n - 1 - l] as i64; + } + } else { + (t[0], t[n]) = (abs as i64, TIME_SLOTS as i64); + for l in 1..n { + t[l] = t[l - 1] + rel[l - 1] as i64; + } + } + (n, pointer) + } + _ => { + let abs0 = bits(br, 2)?; + let abs1 = bits(br, 2)? + TIME_SLOTS; + let (rels0, rels1) = (bits(br, 2)?, bits(br, 2)?); + let n = rels0 + rels1 + 1; + if n > MAX_ENV { + return Err(bad()); + } + let (rel0, rel1) = (read_rel(br, rels0)?, read_rel(br, rels1)?); + let pointer = bits(br, pointer_bits(n))?; + for r in freq_res.iter_mut().take(n) { + *r = br.read_flag()?; + } + (t[0], t[n]) = (abs0 as i64, abs1 as i64); + for l in 1..=rels0 { + t[l] = t[l - 1] + rel0[l - 1] as i64; + } + for l in (rels0 + 1..n).rev() { + t[l] = t[l + 1] - rel1[n - 1 - l] as i64; + } + (n, pointer) + } + }; + if t[0] < 0 || t[..=n].windows(2).any(|w| w[0] >= w[1]) { + return Err(bad()); + } + // l_A: in ISO/IEC 14496-3 a VARFIX pointer of 1, like 0, marks no transient envelope + let la = match (pointer, class) { + (0, _) | (1, VARFIX) => None, + (p, FIXVAR | VARVAR) => Some((n + 1).checked_sub(p).ok_or_else(bad)?), + (p, _) => Some(p - 1), + }; + if la.is_some_and(|la| la > n) { + return Err(bad()); + } + let t_env: [usize; MAX_ENV + 1] = std::array::from_fn(|i| t[i] as usize); + let (num_noise, t_noise) = if n == 1 { + (1, [t_env[0], t_env[1], 0]) + } else { + let middle = match (class, pointer) { + (FIXFIX, _) => n / 2, + (VARFIX, 0) => 1, + (VARFIX, 1) => n - 1, + (VARFIX, p) => p - 1, + (_, 0 | 1) => n - 1, + (_, p) => (n + 1).checked_sub(p).ok_or_else(bad)?, + }; + if middle == 0 || middle >= n { + return Err(bad()); + } + (2, [t_env[0], t_env[middle], t_env[n]]) + }; + Ok(Grid { class, num_env: n, t_env, num_noise, t_noise, freq_res, la, amp_res: amp_res && !(class == FIXFIX && n == 1) }) + } +} + +/// The data of one channel as transmitted: delta-coded envelope and noise floor values (the +/// first frequency-coded value of each is absolute), not yet scaled for a balance channel. +#[derive(Clone, Debug)] +pub(crate) struct ChannelData { + pub(crate) grid: Grid, + pub(crate) df_env: [bool; MAX_ENV], + pub(crate) df_noise: [bool; 2], + pub(crate) invf: [u8; 5], + pub(crate) env: [[i32; 64]; MAX_ENV], + pub(crate) noise: [[i32; 5]; 2], + /// `bs_add_harmonic` per high-resolution band, when `bs_add_harmonic_flag` is set. + pub(crate) harmonic: Option<[bool; 64]>, +} + +impl ChannelData { + fn new(grid: Grid) -> ChannelData { + ChannelData { + grid, + df_env: [false; MAX_ENV], + df_noise: [false; 2], + invf: [0; 5], + env: [[0; 64]; MAX_ENV], + noise: [[0; 5]; 2], + harmonic: None, + } + } + + fn read_dtdf(&mut self, br: &mut BitReader) -> Result<()> { + for d in self.df_env.iter_mut().take(self.grid.num_env) { + *d = br.read_flag()?; + } + for d in self.df_noise.iter_mut().take(self.grid.num_noise) { + *d = br.read_flag()?; + } + Ok(()) + } + + fn read_invf(&mut self, br: &mut BitReader, t: &FreqTables) -> Result<()> { + for m in self.invf.iter_mut().take(t.n_q()) { + *m = br.read_bits(2)? as u8; + } + Ok(()) + } + + /// `sbr_envelope()`; `balance` for the second channel of a coupled pair. + fn read_envelope(&mut self, br: &mut BitReader, t: &FreqTables, balance: bool) -> Result<()> { + let (t_book, f_book, start_bits) = match (self.grid.amp_res, balance) { + (false, false) => (T_ENV_15, F_ENV_15, 7), + (false, true) => (T_BAL_15, F_BAL_15, 6), + (true, false) => (T_ENV_30, F_ENV_30, 6), + (true, true) => (T_BAL_30, F_BAL_30, 5), + }; + for l in 0..self.grid.num_env { + let n = t.bands(self.grid.freq_res[l]).len() - 1; + let row = &mut self.env[l]; + if self.df_env[l] { + for v in row.iter_mut().take(n) { + *v = huff(br, t_book)?; + } + } else { + row[0] = br.read_bits(start_bits)? as i32; + for v in row.iter_mut().take(n).skip(1) { + *v = huff(br, f_book)?; + } + } + } + Ok(()) + } + + /// `sbr_noise()`. + fn read_noise(&mut self, br: &mut BitReader, t: &FreqTables, balance: bool) -> Result<()> { + let (t_book, f_book) = if balance { (T_NOISE_BAL, F_BAL_30) } else { (T_NOISE, F_ENV_30) }; + for l in 0..self.grid.num_noise { + let row = &mut self.noise[l]; + if self.df_noise[l] { + for v in row.iter_mut().take(t.n_q()) { + *v = huff(br, t_book)?; + } + } else { + row[0] = br.read_bits(5)? as i32; + for v in row.iter_mut().take(t.n_q()).skip(1) { + *v = huff(br, f_book)?; + } + } + } + Ok(()) + } + + /// `bs_add_harmonic_flag` and `sbr_sinusoidal_coding()`. + fn read_harmonics(&mut self, br: &mut BitReader, t: &FreqTables) -> Result<()> { + if br.read_flag()? { + let mut h = [false; 64]; + for v in h.iter_mut().take(t.n_high()) { + *v = br.read_flag()?; + } + self.harmonic = Some(h); + } + Ok(()) + } +} + +/// Skip `bs_extended_data` (parametric stereo, for one, is not decoded). +fn skip_extended(br: &mut BitReader) -> Result<()> { + if br.read_flag()? { + let mut count = bits(br, 4)?; + if count == 15 { + count += bits(br, 8)?; + } + br.skip(8 * count)?; + } + Ok(()) +} + +/// The SBR data of one channel element. +#[derive(Clone, Debug)] +pub(crate) struct Frame { + /// The pair codes its second channel as a balance against the first. + pub(crate) coupling: bool, + pub(crate) channels: Vec, +} + +/// `sbr_single_channel_element()` (`channels` 1) or `sbr_channel_pair_element()` (2). +pub(crate) fn read_data(br: &mut BitReader, t: &FreqTables, amp_res: bool, channels: usize) -> Result { + if br.read_flag()? { + br.skip(4 * channels)?; // bs_reserved + } + if channels == 1 { + let mut c = ChannelData::new(Grid::read(br, amp_res)?); + c.read_dtdf(br)?; + c.read_invf(br, t)?; + c.read_envelope(br, t, false)?; + c.read_noise(br, t, false)?; + c.read_harmonics(br, t)?; + skip_extended(br)?; + return Ok(Frame { coupling: false, channels: vec![c] }); + } + let coupling = br.read_flag()?; + let (mut a, mut b); + if coupling { + a = ChannelData::new(Grid::read(br, amp_res)?); + b = ChannelData::new(a.grid.clone()); + a.read_dtdf(br)?; + b.read_dtdf(br)?; + a.read_invf(br, t)?; + b.invf = a.invf; + a.read_envelope(br, t, false)?; + a.read_noise(br, t, false)?; + b.read_envelope(br, t, true)?; + b.read_noise(br, t, true)?; + } else { + a = ChannelData::new(Grid::read(br, amp_res)?); + b = ChannelData::new(Grid::read(br, amp_res)?); + a.read_dtdf(br)?; + b.read_dtdf(br)?; + a.read_invf(br, t)?; + b.read_invf(br, t)?; + a.read_envelope(br, t, false)?; + b.read_envelope(br, t, false)?; + a.read_noise(br, t, false)?; + b.read_noise(br, t, false)?; + } + a.read_harmonics(br, t)?; + b.read_harmonics(br, t)?; + skip_extended(br)?; + Ok(Frame { coupling, channels: vec![a, b] }) +} + +#[cfg(test)] +mod tests { + use craft_media::bitstream::BitWriter; + + use super::*; + + /// Every SBR Huffman table is a complete prefix code, with the shortest codeword for a zero + /// delta, and its decoder reads back every codeword. + #[test] + fn huffman_tables_are_complete_prefix_codes() { + let tables: [&[(u32, u8)]; 10] = [ + &T_ENV_1_5DB, + &F_ENV_1_5DB, + &T_ENV_BAL_1_5DB, + &F_ENV_BAL_1_5DB, + &T_ENV_3_0DB, + &F_ENV_3_0DB, + &T_ENV_BAL_3_0DB, + &F_ENV_BAL_3_0DB, + &T_NOISE_3_0DB, + &T_NOISE_BAL_3_0DB, + ]; + let lavs = [60, 60, 24, 24, 31, 31, 12, 12, 31, 12]; + for (i, (codes, lav)) in tables.iter().zip(lavs).enumerate() { + assert_eq!(codes.len(), 2 * lav + 1, "table {i}"); + let max = u32::from(codes.iter().map(|c| c.1).max().unwrap()); + assert_eq!(codes.iter().map(|&(_, l)| 1u64 << (max - u32::from(l))).sum::(), 1 << max, "table {i}: Kraft sum"); + let shortest = codes.iter().map(|c| c.1).min().unwrap(); + assert_eq!(codes[lav].1, shortest, "table {i}"); + for (sym, &(code, len)) in codes.iter().enumerate() { + let mut bw = BitWriter::new(); + bw.write_bits(code, u32::from(len)); + bw.write_bits(0x5A5A5A, 24); + let data = bw.finish(); + let mut br = BitReader::new(&data); + assert_eq!(huff(&mut br, Book(i, lav as i32)).unwrap(), sym as i32 - lav as i32, "table {i}"); + assert_eq!(br.position(), usize::from(len)); + } + } + } + + fn grid(bits: &[(u32, u32)]) -> Result { + let mut bw = BitWriter::new(); + for &(v, n) in bits { + bw.write_bits(v, n); + } + bw.write_bits(0, 16); + let data = bw.finish(); + Grid::read(&mut BitReader::new(&data), true) + } + + /// The four frame classes give the borders, noise floors and transient envelope of + /// §4.6.18.3.3; a FIXFIX frame of one envelope uses 1.5 dB steps. + #[test] + fn grids_of_each_frame_class() { + // FIXFIX, 4 envelopes, high resolution + let g = grid(&[(FIXFIX as u32, 2), (2, 2), (1, 1)]).unwrap(); + assert_eq!((g.num_env, &g.t_env[..5], g.num_noise, g.t_noise, g.la, g.amp_res), (4, &[0, 4, 8, 12, 16][..], 2, [0, 8, 16], None, true)); + let g = grid(&[(FIXFIX as u32, 2), (0, 2), (0, 1)]).unwrap(); + assert_eq!((g.num_env, g.t_noise, g.amp_res), (1, [0, 16, 0], false)); + // FIXVAR: trailing border 16 + 2, two relative borders (4, 2) back from it, pointer 1, + // resolutions sent last envelope first + let g = grid(&[(FIXVAR as u32, 2), (2, 2), (2, 2), (1, 2), (0, 2), (1, 2), (0, 1), (1, 1), (1, 1)]).unwrap(); + assert_eq!((g.num_env, &g.t_env[..4]), (3, &[0, 12, 14, 18][..])); + assert_eq!((g.la, g.t_noise, g.freq_res[..3].to_vec()), (Some(3), [0, 14, 18], vec![true, true, false])); + // VARFIX: leading border 1, relative 2, pointer 2: l_A = 1 + let g = grid(&[(VARFIX as u32, 2), (1, 2), (1, 2), (0, 2), (2, 2), (1, 1), (0, 1)]).unwrap(); + assert_eq!((g.num_env, &g.t_env[..3], g.la, g.t_noise), (2, &[1, 3, 16][..], Some(1), [1, 3, 16])); + // the same with pointer 1: no transient envelope, the noise floors still split at 3 + let g = grid(&[(VARFIX as u32, 2), (1, 2), (1, 2), (0, 2), (1, 2), (1, 1), (0, 1)]).unwrap(); + assert_eq!((g.la, g.t_noise), (None, [1, 3, 16])); + // VARVAR: 1 .. 17 with one border from each end + let g = grid(&[(VARVAR as u32, 2), (1, 2), (1, 2), (1, 2), (1, 2), (0, 2), (0, 2), (0, 2), (0, 1), (0, 1), (0, 1)]).unwrap(); + assert_eq!((g.num_env, &g.t_env[..4], g.la), (3, &[1, 3, 15, 17][..], None)); + } + + /// Grids whose borders do not increase, too many envelopes and a pointer past the envelopes + /// are refused. + #[test] + fn invalid_grids_are_refused() { + // FIXFIX with 8 envelopes + assert!(grid(&[(FIXFIX as u32, 2), (3, 2), (0, 1)]).is_err()); + // VARVAR with 7 envelopes + assert!(grid(&[(VARVAR as u32, 2), (0, 2), (0, 2), (3, 2), (3, 2)]).is_err()); + // FIXVAR whose relative borders run back past the start: 16 - (8 + 8 + 8) + assert!(grid(&[(FIXVAR as u32, 2), (0, 2), (3, 2), (3, 2), (3, 2), (3, 2), (0, 3), (0, 4)]).is_err()); + // VARFIX, 2 envelopes, pointer 3: the middle border would be the trailing one + assert!(grid(&[(VARFIX as u32, 2), (0, 2), (1, 2), (0, 2), (3, 2), (0, 1), (0, 1)]).is_err()); + } +} diff --git a/crates/audio/src/aac/sbr/freq.rs b/crates/audio/src/aac/sbr/freq.rs new file mode 100644 index 000000000..7bb372dc4 --- /dev/null +++ b/crates/audio/src/aac/sbr/freq.rs @@ -0,0 +1,372 @@ +//! Frequency band tables (ISO/IEC 14496-3 §4.6.18.3.2) and the patches and limiter bands built on +//! them (§4.6.18.6.3, §4.6.18.7.3). Every table holds QMF channel borders. + +use super::Header; + +/// The most patches the HF generator may build. +const MAX_PATCHES: usize = 5; + +/// The tables of one SBR header at one output rate. +#[derive(Clone, Debug, PartialEq, Eq)] +pub(crate) struct FreqTables { + /// First QMF channel of the master table (the start of the SBR range before the crossover). + pub(crate) k0: usize, + /// First QMF channel of the SBR range. + pub(crate) kx: usize, + /// Channels in the SBR range (`kx + m` is its end). + pub(crate) m: usize, + /// High-resolution envelope bands (`n_high + 1` borders). + pub(crate) high: Vec, + /// Low-resolution envelope bands. + pub(crate) low: Vec, + /// Noise floor bands (1 to 5). + pub(crate) noise: Vec, + /// Limiter bands. + pub(crate) lim: Vec, + /// `(first source channel, channels)` of each patch, in order from `kx` up. + pub(crate) patches: Vec<(usize, usize)>, +} + +/// Round half up, as the specification's `NINT`. +fn nint(x: f64) -> i64 { + (x + 0.5).floor() as i64 +} + +/// The start channel offsets of `bs_start_freq`, by output rate (§4.6.18.3.2.1). +fn start_offsets(fs: u32) -> &'static [i8; 16] { + match fs { + 0..=16_000 => &[-8, -7, -6, -5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7], + 16_001..=22_050 => &[-5, -4, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7, 9, 11, 13], + 22_051..=24_000 => &[-5, -3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7, 9, 11, 13, 16], + 24_001..=32_000 => &[-6, -4, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7, 9, 11, 13, 16], + 32_001..=64_000 => &[-4, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7, 9, 11, 13, 16, 20], + _ => &[-2, -1, 0, 1, 2, 3, 4, 5, 6, 7, 9, 11, 13, 16, 20, 24], + } +} + +/// `NINT(f · 128 / fs)` for the 3/4/5 kHz (start) or 6/8/10 kHz (stop) minimum at `fs`. +fn min_channel(fs: u32, low: f64, mid: f64, high: f64) -> i64 { + let f = if fs < 32_000 { + low + } else if fs < 64_000 { + mid + } else { + high + }; + nint(f * 128.0 / f64::from(fs)) +} + +/// The start and stop channels `k0`, `k2` of the master table. +fn start_stop(h: &Header, fs: u32) -> Option<(usize, usize)> { + let k0 = min_channel(fs, 3000.0, 4000.0, 5000.0) + i64::from(start_offsets(fs)[usize::from(h.start_freq & 15)]); + let k2 = match h.stop_freq { + 14 => 2 * k0, + 15 => 3 * k0, + s => { + let stop_min = min_channel(fs, 6000.0, 8000.0, 10_000.0); + if stop_min <= 0 { + return None; + } + let ratio = 64.0 / stop_min as f64; + let at = |p: i64| nint(stop_min as f64 * ratio.powf(p as f64 / 13.0)); + let mut dk: Vec = (0..13).map(|p| at(p + 1) - at(p)).collect(); + dk.sort_unstable(); + stop_min + dk.iter().take(usize::from(s)).sum::() + } + } + .min(64); + // the low band reaches only the analysis bank's 32 channels (`kx` is checked against them + // once the crossover is known) + (1..=32).contains(&k0).then_some(())?; + (k2 > k0).then_some((k0 as usize, k2 as usize)) +} + +/// The master table for `bs_freq_scale` 0: bands of one or two channels. +fn master_linear(k0: usize, k2: usize, alter_scale: bool) -> Option> { + let (dk, bands) = if alter_scale { (2, 2 * nint((k2 - k0) as f64 / 4.0) as usize) } else { (1, 2 * ((k2 - k0) / 2)) }; + if bands == 0 { + return None; + } + let mut v = vec![dk as i64; bands]; + let mut diff = k2 as i64 - (k0 + bands * dk) as i64; + // take the difference from the lowest bands (too wide) or give it to the highest (too narrow) + let mut k = if diff < 0 { 0 } else { bands - 1 }; + while diff != 0 { + let step = if diff < 0 { 1 } else { -1 }; + *v.get_mut(k)? -= step; + diff += step; + k = if step == 1 { k + 1 } else { k.checked_sub(1)? }; + } + cumulative(k0, &v) +} + +/// The borders `start, start + d[0], …`; `None` unless every band is at least a channel wide. +fn cumulative(start: usize, d: &[i64]) -> Option> { + let mut out = Vec::with_capacity(d.len() + 1); + out.push(start); + let mut at = start as i64; + for &d in d { + if d <= 0 { + return None; + } + at += d; + out.push(at as usize); + } + Some(out) +} + +/// The band widths of `2 · NINT(bands · log2(hi / lo) / (2 · warp))` logarithmic bands. +fn log_widths(lo: usize, hi: usize, bands: f64, warp: f64) -> Vec { + let ratio = hi as f64 / lo as f64; + let n = 2 * nint(bands * ratio.log2() / (2.0 * warp)); + let at = |k: i64| nint(lo as f64 * ratio.powf(k as f64 / n as f64)); + let mut d: Vec = (0..n.max(0)).map(|k| at(k + 1) - at(k)).collect(); + d.sort_unstable(); + d +} + +/// The master table for `bs_freq_scale` 1–3: 12, 10 or 8 bands per octave, warped above +/// `2 · k0` by `bs_alter_scale`. +fn master_log(k0: usize, k2: usize, freq_scale: u8, alter_scale: bool) -> Option> { + let bands = [12.0, 10.0, 8.0][usize::from(freq_scale.clamp(1, 3) - 1)]; + let two_regions = k2 as f64 / k0 as f64 > 2.2449; + let k1 = if two_regions { 2 * k0 } else { k2 }; + let d0 = log_widths(k0, k1, bands, 1.0); + let mut table = cumulative(k0, &d0)?; + if table.len() < 2 { + return None; + } + if two_regions { + let mut d1 = log_widths(k1, k2, bands, if alter_scale { 1.3 } else { 1.0 }); + let max0 = d0.iter().copied().max()?; + if d1.first().is_some_and(|&d| d < max0) { + // widen the first upper band to the widest lower one, at the cost of the last + let change = max0 - d1[0]; + d1[0] = max0; + *d1.last_mut()? -= change; + d1.sort_unstable(); + } + table.extend(cumulative(k1, &d1)?.into_iter().skip(1)); + } + Some(table) +} + +impl FreqTables { + /// The tables of `h` at output rate `fs`, or `None` if the header asks for an invalid layout. + pub(crate) fn new(h: &Header, fs: u32) -> Option { + let (k0, k2) = start_stop(h, fs)?; + let master = if h.freq_scale == 0 { master_linear(k0, k2, h.alter_scale)? } else { master_log(k0, k2, h.freq_scale, h.alter_scale)? }; + let n_master = master.len().checked_sub(1)?; + let xover = usize::from(h.xover_band); + if xover >= n_master { + return None; + } + let high: Vec = master.get(xover..)?.to_vec(); + let n_high = high.len() - 1; + let (kx, end) = (high[0], *high.last()?); + if kx > 32 || end > 64 || end <= kx { + return None; + } + let m = end - kx; + // every other border, keeping the first (and the second when the count is odd) + let n_low = n_high - n_high / 2; + let low: Vec = (0..=n_low).map(|k| high[if k == 0 { 0 } else { 2 * k - (n_high & 1) }]).collect(); + let n_q = if h.noise_bands == 0 { 1 } else { (nint(f64::from(h.noise_bands) * (end as f64 / kx as f64).log2()) as usize).max(1) }; + if n_q > 5 || n_q > n_low { + return None; + } + let mut noise = vec![low[0]]; + let mut i = 0; + for k in 1..=n_q { + i += (n_low - i) / (n_q + 1 - k); + noise.push(*low.get(i)?); + } + let patches = patches(&master, k0, kx, m, fs)?; + let lim = limiter(&low, &patches, kx, h.limiter_bands); + Some(FreqTables { k0, kx, m, high, low, noise, lim, patches }) + } + + /// High-resolution envelope bands. + pub(crate) fn n_high(&self) -> usize { + self.high.len() - 1 + } + + /// Noise floor bands. + pub(crate) fn n_q(&self) -> usize { + self.noise.len() - 1 + } + + /// The envelope band table of a resolution (`true`: high). + pub(crate) fn bands(&self, high_res: bool) -> &[usize] { + if high_res { &self.high } else { &self.low } + } +} + +/// Map the low band onto the SBR range (§4.6.18.6.3): runs of source channels, each ending below +/// `k0`, that together cover `kx .. kx + m`. +fn patches(master: &[usize], k0: usize, kx: usize, m: usize, fs: u32) -> Option> { + let n_master = master.len() - 1; + let goal = nint(2.048e6 / f64::from(fs)).max(0) as usize; + let mut k = if goal < kx + m { master.iter().position(|&f| f >= goal).unwrap_or(n_master) } else { n_master }; + let (mut msb, mut usb) = (k0, kx); + let mut out: Vec<(usize, usize)> = Vec::new(); + // each pass either adds a patch or lowers `msb` to `kx`; a stream that does neither twice + // running cannot finish. One patch over the limit may still be a short last one that is + // dropped below. + for _ in 0..2 * (MAX_PATCHES + 2) { + // the highest master border the patch can reach from below k0 + let mut j = k + 1; + let (sb, odd) = loop { + j = j.checked_sub(1)?; + let sb = *master.get(j)?; + let odd = (sb + k0) % 2; + if sb + odd < k0 + msb { + break (sb, odd); + } + }; + let count = sb.saturating_sub(usb); + if count > 0 { + let start = (k0 - odd).checked_sub(count)?; + out.push((start, count)); + if out.len() > MAX_PATCHES + 1 { + return None; + } + (usb, msb) = (sb, sb); + } else { + msb = kx; + } + if master.get(k)?.saturating_sub(sb) < 3 { + k = n_master; + } + if sb == kx + m { + if out.len() > 1 && out.last().is_some_and(|p| p.1 < 3) { + out.pop(); + } + return (!out.is_empty() && out.len() <= MAX_PATCHES).then_some(out); + } + } + None +} + +/// The limiter band table (§4.6.18.7.3): one band over the range, or about 1.2, 2 or 3 bands per +/// octave on the low-resolution and patch borders. +fn limiter(low: &[usize], patches: &[(usize, usize)], kx: usize, limiter_bands: u8) -> Vec { + let (Some(&first), Some(&last)) = (low.first(), low.last()) else { + return Vec::new(); + }; + if limiter_bands == 0 { + return vec![first, last]; + } + let per_octave = [1.2, 2.0, 3.0][usize::from(limiter_bands.clamp(1, 3) - 1)]; + let mut borders = vec![kx]; + for p in patches { + borders.push(borders.last().copied().unwrap_or(kx) + p.1); + } + let mut t: Vec = low.to_vec(); + t.extend(borders.iter().skip(1).take(patches.len().saturating_sub(1))); + t.sort_unstable(); + let is_border = |v: usize| borders.contains(&v); + let mut k = 1; + while k < t.len() { + let octaves = if t[k - 1] == 0 { f64::INFINITY } else { (t[k] as f64 / t[k - 1] as f64).log2() }; + if octaves * per_octave >= 0.49 { + k += 1; + } else if t[k] == t[k - 1] || !is_border(t[k]) { + t.remove(k); + } else if !is_border(t[k - 1]) { + t.remove(k - 1); + } else { + k += 1; + } + } + if t.len() < 2 { + return vec![first, last]; + } + t +} + +#[cfg(test)] +mod tests { + use super::*; + + fn header(start: u8, stop: u8, xover: u8, freq_scale: u8, alter: bool, noise: u8) -> Header { + Header { + amp_res: true, + start_freq: start, + stop_freq: stop, + xover_band: xover, + freq_scale, + alter_scale: alter, + noise_bands: noise, + limiter_bands: 2, + limiter_gains: 2, + interpol_freq: true, + smoothing_mode: true, + } + } + + /// Every header field combination at every output rate gives tables that are strictly + /// increasing, start at `kx`, end at `kx + m`, keep the low band within the analysis bank and + /// patch from below `k0` (or is refused). + #[test] + fn every_header_gives_consistent_tables_or_none() { + let mut valid = 0; + for fs in [16_000, 22_050, 24_000, 32_000, 44_100, 48_000, 64_000, 88_200, 96_000] { + for start in 0..16 { + for stop in 0..16 { + for xover in 0..8 { + for fsc in 0..4 { + for alter in [false, true] { + for noise in 0..4 { + let Some(t) = FreqTables::new(&header(start, stop, xover, fsc, alter, noise), fs) else { continue }; + valid += 1; + for v in [&t.high, &t.low, &t.noise, &t.lim] { + assert!(v.windows(2).all(|w| w[0] < w[1]), "{v:?}"); + assert_eq!(v[0], t.kx); + } + for v in [&t.high, &t.low, &t.noise] { + assert_eq!(*v.last().unwrap(), t.kx + t.m); + } + assert!(t.kx <= 32 && t.kx + t.m <= 64 && t.n_q() <= 5); + // a last patch of under 3 channels is dropped, and the limiter + // bands may then stop short of the range's end + assert!(t.patches.iter().map(|p| p.1).sum::() <= t.m); + assert!(t.lim.len() >= 2 && *t.lim.last().unwrap() <= t.kx + t.m); + assert!(t.patches.iter().all(|&(s, n)| s + n <= t.k0)); + } + } + } + } + } + } + } + assert!(valid > 10_000, "{valid}"); + } + + /// The highest start a header can give, 32 at 24 kHz (start 15), is valid: with stop 14 + /// (twice the start) and linear bands it covers channels 32 to 64. + #[test] + fn a_start_at_channel_32_is_accepted() { + let t = FreqTables::new(&header(15, 14, 0, 0, false, 2), 24_000).unwrap(); + assert_eq!((t.k0, t.kx, t.m), (32, 32, 32)); + } + + /// At 96 kHz this header builds patches of 6, 6, 6, 6, 5 and 2 channels: the short sixth is + /// dropped, as any short last patch is, and the five left are within the limit. + #[test] + fn a_short_sixth_patch_is_dropped_before_the_limit() { + let t = FreqTables::new(&header(3, 9, 0, 0, false, 2), 96_000).unwrap(); + assert_eq!((t.k0, t.kx + t.m), (8, 39)); + assert_eq!(t.patches.iter().map(|p| p.1).collect::>(), [6, 6, 6, 6, 5]); + } + + /// A typical 44.1 kHz header: start channel 12 + offset, stop near 64, three noise bands. + #[test] + fn a_44k_header_has_the_expected_layout() { + let t = FreqTables::new(&header(5, 9, 0, 2, true, 2), 44_100).unwrap(); + // startMin 12 + offset 2 = 14 + assert_eq!((t.k0, t.kx), (14, 14)); + assert_eq!(t.low.len() - 1, t.n_high() - t.n_high() / 2); + assert!(t.n_q() >= 1); + } +} diff --git a/crates/audio/src/aac/sbr/hf.rs b/crates/audio/src/aac/sbr/hf.rs new file mode 100644 index 000000000..1c2fd81ac --- /dev/null +++ b/crates/audio/src/aac/sbr/hf.rs @@ -0,0 +1,333 @@ +//! HF generation (ISO/IEC 14496-3 §4.6.18.6) and HF adjustment (§4.6.18.7) of one channel. + +use super::bitstream::Grid; +use super::freq::FreqTables; +use super::tables::NOISE; +use super::{Cx, Header, MAX_ENV, T_HF_ADJ, X_SLOTS}; + +/// Low band subband samples of a frame: slots -8 .. 32 relative to the frame (`X_Low`). +pub(crate) type LowBand = [[Cx; 32]; X_SLOTS]; +/// High band subband samples, at slot `l + T_HF_ADJ` for QMF slot `l` of the frame's envelopes. +pub(crate) type HighBand = [[Cx; 64]; X_SLOTS]; + +/// The per-channel state the HF tools carry from frame to frame. +#[derive(Clone, Debug)] +pub(crate) struct History { + /// `bs_invf_mode` and chirp factor per noise band. + invf: [u8; 5], + bw: [f32; 5], + /// The last frame's transient envelope was its last (`l_A = L_E`): no noise in this frame's + /// first. + la_at_end: bool, + /// Sinusoids present in the last envelope, by channel above `kx`. + sine: [bool; 64], + /// Gains and noise levels of the last four slots, newest first (smoothing). + gains: [[f32; 64]; 4], + noise_levels: [[f32; 64]; 4], + /// The smoothing history holds data of the current tables. + primed: bool, + noise_index: usize, + sine_index: usize, +} + +impl History { + pub(crate) fn new() -> History { + History { + invf: [0; 5], + bw: [0.0; 5], + la_at_end: false, + sine: [false; 64], + gains: [[0.0; 64]; 4], + noise_levels: [[0.0; 64]; 4], + primed: false, + noise_index: 0, + sine_index: 0, + } + } + + /// Forget what depends on the frequency tables (after a header that changes them). + pub(crate) fn reset_tables(&mut self) { + self.sine = [false; 64]; + self.primed = false; + self.noise_index = 0; + } +} + +/// The chirp factor of a noise band from its inverse filtering mode and the last frame's. +fn chirp(mode: u8, prev_mode: u8, prev_bw: f32) -> f32 { + let new_bw = match (mode, prev_mode) { + (0, 1) => 0.6, + (0, _) => 0.0, + (1, 0) => 0.6, + (1, _) => 0.75, + (2, _) => 0.9, + _ => 0.98, + }; + let bw = if new_bw < prev_bw { 0.75 * new_bw + 0.25 * prev_bw } else { 0.90625 * new_bw + 0.09375 * prev_bw }; + if bw < 0.015625 { 0.0 } else { bw } +} + +/// Second order linear prediction of one low band channel over the frame and the 6 slots +/// before it (the covariance method, §4.6.18.6.2): `(α0, α1)`. +fn predictor(x: &LowBand, k: usize) -> (Cx, Cx) { + // φ(i, j) = Σ x(n − i + 2) · conj(x(n − j + 2)), n = 0 .. 37 + let phi = |i: usize, j: usize| { + let (mut re, mut im) = (0f64, 0f64); + for n in 0..SLOTS_PLUS_6 { + let (a, b) = (x[n + 2 - i][k], x[n + 2 - j][k]); + re += f64::from(a.re) * f64::from(b.re) + f64::from(a.im) * f64::from(b.im); + im += f64::from(a.im) * f64::from(b.re) - f64::from(a.re) * f64::from(b.im); + } + (re, im) + }; + let (p01, p02, p11, p12, p22) = (phi(0, 1), phi(0, 2), phi(1, 1).0, phi(1, 2), phi(2, 2).0); + let d = p22 * p11 - (p12.0 * p12.0 + p12.1 * p12.1) / (1.0 + 1e-6); + let a1 = if d != 0.0 { + // (φ01 · φ12 − φ02 · φ11) / d + ((p01.0 * p12.0 - p01.1 * p12.1 - p02.0 * p11) / d, (p01.0 * p12.1 + p01.1 * p12.0 - p02.1 * p11) / d) + } else { + (0.0, 0.0) + }; + let a0 = if p11 != 0.0 { + // −(φ01 + α1 · conj(φ12)) / φ11 + (-(p01.0 + a1.0 * p12.0 + a1.1 * p12.1) / p11, -(p01.1 + a1.1 * p12.0 - a1.0 * p12.1) / p11) + } else { + (0.0, 0.0) + }; + let mag = |a: (f64, f64)| (a.0 * a.0 + a.1 * a.1).sqrt(); + if mag(a0) >= 4.0 || mag(a1) >= 4.0 { + return (Cx::default(), Cx::default()); + } + (Cx { re: a0.0 as f32, im: a0.1 as f32 }, Cx { re: a1.0 as f32, im: a1.1 as f32 }) +} + +/// Slots of the covariance sums: the frame's 32 and 6 before it. +const SLOTS_PLUS_6: usize = super::SLOTS + 6; + +/// The index of the band of `bands` holding channel `k` (bands are increasing borders). +fn band_of(bands: &[usize], k: usize) -> usize { + bands.partition_point(|&b| b <= k).saturating_sub(1).min(bands.len().saturating_sub(2)) +} + +/// Patch the low band into the SBR range, inverse filtered (§4.6.18.6.3), for QMF slots +/// `from .. to` of the frame. +pub(crate) fn generate(x: &LowBand, t: &FreqTables, invf: &[u8; 5], h: &mut History, (from, to): (usize, usize), out: &mut HighBand) { + let mut bw = [0f32; 5]; + for i in 0..t.n_q() { + bw[i] = chirp(invf[i], h.invf[i], h.bw[i]); + } + (h.invf, h.bw) = (*invf, bw); + let mut pred: [Option<(Cx, Cx)>; 32] = [None; 32]; + let mut k = t.kx; + for &(start, count) in &t.patches { + for p in start..start + count { + let (a0, a1) = *pred[p].get_or_insert_with(|| predictor(x, p)); + let g = bw[band_of(&t.noise, k)]; + let (c0, c1) = (a0.scale(g), a1.scale(g * g)); + for l in from..to { + let i = l + T_HF_ADJ; + out[i][k] = x[i][p] + c0 * x[i - 1][p] + c1 * x[i - 2][p]; + } + k += 1; + } + } +} + +/// Dequantised envelope and noise floor values of one channel in one frame. +#[derive(Clone, Debug)] +pub(crate) struct Levels { + /// Envelope energy per envelope and band (of that envelope's resolution). + pub(crate) energy: [[f32; 64]; MAX_ENV], + /// Noise to signal ratio per noise floor and noise band. + pub(crate) noise: [[f32; 5]; 2], +} + +/// Gains for one envelope, per channel of the SBR range. +struct EnvelopeGains { + gain: [f32; 64], + noise: [f32; 64], + sine: [f32; 64], +} + +const LIMITER_GAINS: [f32; 4] = [0.70795, 1.0, 1.41254, 1e10]; +/// `h_smooth`, as the specification gives it. +#[allow(clippy::excessive_precision)] +const SMOOTHING: [f32; 5] = [0.33333333333333, 0.30150283239582, 0.21816949906249, 0.11516383427084, 0.03183050093751]; +const EPS: f32 = 1.0; +const EPS0: f32 = 1e-12; + +/// What the adjustment of one channel's frame works from. +pub(crate) struct Frame<'a> { + /// The generated high band. + pub(crate) x: &'a HighBand, + pub(crate) t: &'a FreqTables, + pub(crate) hdr: &'a Header, + pub(crate) g: &'a Grid, + pub(crate) lv: &'a Levels, + /// `bs_add_harmonic`, when the frame codes sinusoids. + pub(crate) harmonic: Option<&'a [bool; 64]>, +} + +/// The gains of envelope `l` (§4.6.18.7.3–4): `delta` is 0 for a transient envelope, which gets +/// no noise. +fn envelope_gains(f: &Frame, l: usize, sine_index: &[bool; 64], delta: bool) -> EnvelopeGains { + let Frame { x, t, hdr, g, lv, .. } = *f; + let (kx, m) = (t.kx, t.m); + let bands = t.bands(g.freq_res[l]); + let (from, to) = (2 * g.t_env[l] + T_HF_ADJ, 2 * g.t_env[l + 1] + T_HF_ADJ); + let slots = (to - from) as f32; + let floor = usize::from(g.num_noise == 2 && g.t_env[l] >= g.t_noise[1]); + // per channel: original energy, current energy, noise ratio, sinusoid in its band + let mut e_orig = [0f32; 64]; + let mut e_curr = [0f32; 64]; + let mut q = [0f32; 64]; + let mut s_band = [false; 64]; + for j in 0..bands.len() - 1 { + let (lo, hi) = (bands[j] - kx, bands[j + 1] - kx); + let any_sine = sine_index[lo..hi].iter().any(|&s| s); + let mut sum = 0f32; + for c in lo..hi { + let ch = c + kx; + let e: f32 = (from..to).map(|i| x[i][ch].norm_sqr()).sum::() / slots; + e_curr[c] = e; + sum += e; + e_orig[c] = lv.energy[l][j]; + q[c] = lv.noise[floor][band_of(&t.noise, ch)]; + s_band[c] = any_sine; + } + if !hdr.interpol_freq { + e_curr[lo..hi].fill(sum / (hi - lo) as f32); + } + } + let mut out = EnvelopeGains { gain: [0.0; 64], noise: [0.0; 64], sine: [0.0; 64] }; + for c in 0..m { + // `q / (1 + q)` first: energy and noise ratio are each clamped, their product is not + let (eo, ec, q) = (e_orig[c], e_curr[c], q[c]); + let share = q / (1.0 + q); + out.noise[c] = (eo * share).sqrt(); + if sine_index[c] { + out.sine[c] = (eo / (1.0 + q)).sqrt(); + } + out.gain[c] = if s_band[c] { (eo / (EPS + ec) * share).sqrt() } else { (eo / ((EPS + ec) * (1.0 + if delta { q } else { 0.0 }))).sqrt() }; + } + // limit per limiter band, then boost back the energy the limiter took + let lim_gain = LIMITER_GAINS[usize::from(hdr.limiter_gains & 3)]; + for w in t.lim.windows(2) { + let (lo, hi) = (w[0] - kx, w[1] - kx); + let (so, sc) = (e_orig[lo..hi].iter().sum::(), e_curr[lo..hi].iter().sum::()); + let max = (lim_gain * ((EPS0 + so) / (EPS0 + sc)).sqrt()).min(1e5); + let mut den = 0f32; + for c in lo..hi { + if out.gain[c] > max { + out.noise[c] *= max / out.gain[c]; + out.gain[c] = max; + } + den += e_curr[c] * out.gain[c] * out.gain[c]; + if out.sine[c] != 0.0 { + den += out.sine[c] * out.sine[c]; + } else if delta { + den += out.noise[c] * out.noise[c]; + } + } + let boost = ((EPS0 + so) / (EPS0 + den)).sqrt().min(1.584_893_2); + for c in lo..hi { + out.gain[c] *= boost; + out.noise[c] *= boost; + out.sine[c] *= boost; + } + } + out +} + +/// Adjust the generated high band to the transmitted envelopes and add noise and sinusoids +/// (§4.6.18.7), for the frame's envelopes, into `y`. +pub(crate) fn adjust(f: &Frame, h: &mut History, y: &mut HighBand) { + let Frame { x, t, hdr, g, harmonic, .. } = *f; + let (kx, m) = (t.kx, t.m); + let smooth = !hdr.smoothing_mode; + let prev_la_at_end = h.la_at_end; + for l in 0..g.num_env { + // sinusoids start at the transient envelope, unless one was already playing in the band + let mut sine_index = [false; 64]; + if let Some(harm) = harmonic { + for i in 0..t.n_high() { + let c = (t.high[i] + t.high[i + 1]) / 2 - kx; + sine_index[c] = harm[i] && (g.la.is_none_or(|la| l >= la) || h.sine[c]); + } + } + let transient = g.la == Some(l) || (l == 0 && prev_la_at_end); + let e = envelope_gains(f, l, &sine_index, !transient); + if !h.primed { + h.gains = [e.gain; 4]; + h.noise_levels = [e.noise; 4]; + h.primed = true; + } + for i in 2 * g.t_env[l]..2 * g.t_env[l + 1] { + let (mut gain, mut noise) = (e.gain, e.noise); + if smooth && !transient { + for c in 0..m { + gain[c] = SMOOTHING[0] * e.gain[c] + (0..4).map(|j| SMOOTHING[j + 1] * h.gains[j][c]).sum::(); + noise[c] = SMOOTHING[0] * e.noise[c] + (0..4).map(|j| SMOOTHING[j + 1] * h.noise_levels[j][c]).sum::(); + } + } + h.gains.rotate_right(1); + h.noise_levels.rotate_right(1); + (h.gains[0], h.noise_levels[0]) = (e.gain, e.noise); + let (sin_re, sin_im) = [(1.0, 0.0), (0.0, 1.0), (-1.0, 0.0), (0.0, -1.0)][h.sine_index]; + let row = &mut y[i + T_HF_ADJ]; + for c in 0..m { + let ch = c + kx; + let mut v = x[i + T_HF_ADJ][ch].scale(gain[c]); + h.noise_index = (h.noise_index + 1) & 511; + if e.sine[c] != 0.0 { + let sign = if ch % 2 == 0 { 1.0 } else { -1.0 }; + v.re += e.sine[c] * sin_re; + v.im += e.sine[c] * sin_im * sign; + } else if !transient { + let [nr, ni] = NOISE[h.noise_index]; + v.re += noise[c] * nr; + v.im += noise[c] * ni; + } + row[ch] = v; + } + h.sine_index = (h.sine_index + 1) & 3; + } + if l + 1 == g.num_env { + h.sine = sine_index; + } + } + h.la_at_end = g.la == Some(g.num_env); +} + +#[cfg(test)] +mod tests { + use super::*; + + /// The chirp factor follows §4.6.18.6.2: rising quickly, falling slowly, and zero below 1/64. + #[test] + fn chirp_factors() { + assert!((chirp(3, 3, 0.0) - 0.90625 * 0.98).abs() < 1e-6); + assert!((chirp(0, 3, 0.98) - 0.25 * 0.98).abs() < 1e-6); + assert_eq!(chirp(0, 0, 0.05), 0.0); + assert!((chirp(1, 0, 0.0) - 0.90625 * 0.6).abs() < 1e-6); + } + + /// A channel that is a decaying complex exponential is predicted exactly by its first order + /// term. + #[test] + fn prediction_of_a_resonance() { + let mut x: LowBand = [[Cx::default(); 32]; X_SLOTS]; + let r = Cx { re: 0.9 * 0.3f32.cos(), im: 0.9 * 0.3f32.sin() }; + let mut v = Cx { re: 1.0, im: 0.0 }; + for slot in x.iter_mut() { + slot[3] = v; + v = v * r; + } + let (a0, a1) = predictor(&x, 3); + // x(n) + α0·x(n−1) + α1·x(n−2) = 0 with α0 = −r, α1 = 0 + assert!((a0.re + r.re).abs() < 1e-3 && (a0.im + r.im).abs() < 1e-3, "{a0:?}"); + assert!(a1.re.abs() < 1e-3 && a1.im.abs() < 1e-3, "{a1:?}"); + // a silent channel has no predictor + assert_eq!(predictor(&x, 4), (Cx::default(), Cx::default())); + } +} diff --git a/crates/audio/src/aac/sbr/mod.rs b/crates/audio/src/aac/sbr/mod.rs new file mode 100644 index 000000000..752cd03a0 --- /dev/null +++ b/crates/audio/src/aac/sbr/mod.rs @@ -0,0 +1,544 @@ +//! Spectral band replication (HE-AAC, ISO/IEC 14496-3 §4.6.18): rebuilds the band above the +//! AAC-LC core from the core's own low band and the envelope data of the stream's SBR fill +//! elements, at twice the core rate. +//! +//! Each channel's core output goes through a 32-band QMF analysis bank. The HF generator copies +//! (patches) low band channels into the SBR range, inverse filtered; the HF adjuster scales them +//! to the transmitted envelopes and adds noise and sinusoids; a 64-band QMF synthesis bank merges +//! the delayed low band and the new high band. A frame without usable SBR data (none sent yet, or +//! damaged) still passes through the banks, so the output keeps its rate and delay. +//! +//! Written from the SBR description and tables of ETSI ES 201 980 (DRM, Annex I has the tables it +//! shares with ISO/IEC 14496-3) and the decoder process of ISO/IEC 14496-3. Parametric stereo +//! (HE-AAC v2) is not decoded: its extension is skipped and the mono core is extended alone. + +mod bitstream; +mod freq; +mod hf; +mod qmf; +mod tables; + +use std::ops::{Add, Mul}; + +use craft_media::bitstream::BitReader; + +use self::bitstream::{ChannelData, Frame}; +use self::freq::FreqTables; +use self::hf::{HighBand, History, Levels, LowBand}; +use self::qmf::{Analysis, Synthesis}; + +/// QMF slots per frame: 1024 core samples in slots of 32. +pub(crate) const SLOTS: usize = 32; +/// The most envelopes in a frame. +pub(crate) const MAX_ENV: usize = 5; +/// Slots of the last frame kept ahead of this one's low band (`t_HFGen`). +const T_HF_GEN: usize = 8; +/// The SBR frame lags the core frame by `T_HF_GEN − T_HF_ADJ` slots. +pub(crate) const T_HF_ADJ: usize = 2; +/// Slots in the low and high band buffers: the frame and `T_HF_GEN` before it. +pub(crate) const X_SLOTS: usize = SLOTS + T_HF_GEN; +/// The 16-bit full scale the specification's energies are in. +const PCM_SCALE: f32 = 32768.0; +/// `NOISE_FLOOR_OFFSET`. +const NOISE_OFFSET: i32 = 6; + +/// Output samples by which the SBR decoder delays the core: the filter banks, and the SBR frame +/// lagging the core frame by `T_HF_GEN − T_HF_ADJ` slots of 64. +pub const DELAY: usize = qmf::DELAY + 64 * (T_HF_GEN - T_HF_ADJ); + +/// A complex subband sample. +#[derive(Clone, Copy, Default, Debug, PartialEq)] +pub(crate) struct Cx { + pub(crate) re: f32, + pub(crate) im: f32, +} + +impl Cx { + fn scale(self, g: f32) -> Cx { + Cx { re: self.re * g, im: self.im * g } + } + fn norm_sqr(self) -> f32 { + self.re * self.re + self.im * self.im + } +} + +impl Add for Cx { + type Output = Cx; + fn add(self, o: Cx) -> Cx { + Cx { re: self.re + o.re, im: self.im + o.im } + } +} + +impl Mul for Cx { + type Output = Cx; + fn mul(self, o: Cx) -> Cx { + Cx { re: self.re * o.re - self.im * o.im, im: self.re * o.im + self.im * o.re } + } +} + +/// `sbr_header()` (§4.5.2.8.1). +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub(crate) struct Header { + pub(crate) amp_res: bool, + pub(crate) start_freq: u8, + pub(crate) stop_freq: u8, + pub(crate) xover_band: u8, + pub(crate) freq_scale: u8, + pub(crate) alter_scale: bool, + pub(crate) noise_bands: u8, + pub(crate) limiter_bands: u8, + pub(crate) limiter_gains: u8, + pub(crate) interpol_freq: bool, + pub(crate) smoothing_mode: bool, +} + +impl Header { + /// Whether `o` keeps the frequency band tables (a change of any of these fields is an SBR + /// reset). + fn same_tables(&self, o: &Header) -> bool { + (self.start_freq, self.stop_freq, self.freq_scale, self.alter_scale, self.xover_band, self.noise_bands) + == (o.start_freq, o.stop_freq, o.freq_scale, o.alter_scale, o.xover_band, o.noise_bands) + } +} + +/// Where an element's SBR data is: the access unit, and the bit range of the fill element's +/// payload after its extension type. +#[derive(Clone, Copy, Debug)] +pub(crate) struct Payload<'a> { + pub(crate) data: &'a [u8], + pub(crate) start: usize, + pub(crate) end: usize, + /// `EXT_SBR_DATA_CRC`: a 10-bit CRC leads the data. + pub(crate) crc: bool, +} + +/// Quantised envelope values per envelope and band, and noise floor values per floor and band. +type Quantised = ([[i32; 64]; MAX_ENV], [[i32; 5]; 2]); + +/// One channel's state across frames. +#[derive(Clone)] +struct Channel { + analysis: Analysis, + synthesis: Synthesis, + /// Low band of the frame, with the last `T_HF_GEN` slots of the frame before ahead of it. + x_low: Box, + /// Adjusted high band of the last frame (its envelopes may reach into this one). + y_prev: Box, + /// Where `y_prev` holds data: QMF channels, and the end of its slots. + y_prev_range: (usize, usize), + y_prev_end: usize, + /// Quantised envelope values of the last envelope, and its resolution; noise floor values of + /// the last noise floor (time-direction delta coding). + env: [i32; 64], + env_high_res: bool, + noise: [i32; 5], + history: History, +} + +impl Channel { + fn new() -> Channel { + Channel { + analysis: Analysis::new(), + synthesis: Synthesis::new(), + x_low: Box::new([[Cx::default(); 32]; X_SLOTS]), + y_prev: Box::new([[Cx::default(); 64]; X_SLOTS]), + y_prev_range: (0, 0), + y_prev_end: 0, + env: [0; 64], + env_high_res: false, + noise: [0; 5], + history: History::new(), + } + } + + /// Decode the delta-coded envelope and noise floor values (§4.6.18.3.4); `scale` is 2 for a + /// balance channel. + fn absolute(&mut self, d: &ChannelData, t: &FreqTables, scale: i32) -> Quantised { + let mut env = [[0i32; 64]; MAX_ENV]; + for l in 0..d.grid.num_env { + let res = d.grid.freq_res[l]; + let n = t.bands(res).len() - 1; + if d.df_env[l] { + for k in 0..n { + // the value this band refers to in the last envelope, mapped across resolutions + let reference = if res == self.env_high_res { + k + } else if res { + hf_band_of(t.bands(false), t.high[k]) + } else { + t.high.iter().position(|&b| b == t.low[k]).unwrap_or(0) + }; + env[l][k] = self.env.get(reference).copied().unwrap_or(0) + scale * d.env[l][k]; + } + } else { + env[l][0] = scale * d.env[l][0]; + for k in 1..n { + env[l][k] = env[l][k - 1] + scale * d.env[l][k]; + } + } + (self.env, self.env_high_res) = (env[l], res); + } + let mut noise = [[0i32; 5]; 2]; + for l in 0..d.grid.num_noise { + for k in 0..t.n_q() { + let delta = scale * d.noise[l][k]; + noise[l][k] = if d.df_noise[l] { + self.noise[k] + delta + } else if k == 0 { + delta + } else { + noise[l][k - 1] + delta + }; + } + self.noise = noise[l]; + } + (env, noise) + } +} + +/// The index of the band of `bands` holding channel `k`. +fn hf_band_of(bands: &[usize], k: usize) -> usize { + bands.partition_point(|&b| b <= k).saturating_sub(1) +} + +/// `2^x`, kept to a range whose energies stay finite. +fn exp2(x: f32) -> f32 { + x.clamp(-64.0, 64.0).exp2() +} + +/// Dequantise one channel (§4.6.18.3.5), or a coupled pair from its level and balance channels. +fn dequantise(frame: &Frame, values: &[Quantised], t: &FreqTables) -> Vec { + let g = &frame.channels[0].grid; + let amp = if g.amp_res { 1.0 } else { 0.5 }; + let mut out = vec![Levels { energy: [[0.0; 64]; MAX_ENV], noise: [[0.0; 5]; 2] }; values.len()]; + if frame.coupling && values.len() == 2 { + let pan = if g.amp_res { 12.0 } else { 24.0 }; + let ((e0, q0), (e1, q1)) = (&values[0], &values[1]); + for l in 0..g.num_env { + for k in 0..t.bands(g.freq_res[l]).len() - 1 { + let level = 64.0 * exp2(amp * e0[l][k] as f32 + 1.0); + let balance = amp * (pan - e1[l][k] as f32); + out[0].energy[l][k] = level / (1.0 + exp2(balance)); + out[1].energy[l][k] = level / (1.0 + exp2(-balance)); + } + } + for l in 0..g.num_noise { + for k in 0..t.n_q() { + let level = exp2((NOISE_OFFSET - q0[l][k] + 1) as f32); + let balance = (12 - q1[l][k]) as f32; + out[0].noise[l][k] = level / (1.0 + exp2(balance)); + out[1].noise[l][k] = level / (1.0 + exp2(-balance)); + } + } + return out; + } + for ((lv, (e, q)), d) in out.iter_mut().zip(values).zip(&frame.channels) { + let g = &d.grid; + let amp = if g.amp_res { 1.0 } else { 0.5 }; + for l in 0..g.num_env { + for k in 0..t.bands(g.freq_res[l]).len() - 1 { + lv.energy[l][k] = 64.0 * exp2(amp * e[l][k] as f32); + } + } + for l in 0..g.num_noise { + for k in 0..t.n_q() { + lv.noise[l][k] = exp2((NOISE_OFFSET - q[l][k]) as f32); + } + } + } + out +} + +/// The SBR decoder of one channel element: a single channel (SCE, or an LFE, which has no SBR +/// data) or a channel pair (CPE). +#[derive(Clone)] +pub(crate) struct Element { + /// Output sampling rate. + rate: u32, + header: Option
, + tables: Option, + channels: Vec, +} + +impl Element { + pub(crate) fn new(channels: usize, rate: u32) -> Element { + Element { rate, header: None, tables: None, channels: (0..channels).map(|_| Channel::new()).collect() } + } + + pub(crate) fn channels(&self) -> usize { + self.channels.len() + } + + /// Forget the signal (a seek). The header is kept: it holds until the stream sends another. + pub(crate) fn reset(&mut self) { + for c in &mut self.channels { + *c = Channel::new(); + } + } + + /// Read an element's SBR data: `None` without a header to read it by. + fn read(&mut self, p: Payload) -> Option { + let mut br = BitReader::new(p.data); + br.seek_bits(p.start); + if p.crc { + br.skip(10).ok()?; + } + if br.read_flag().ok()? { + let h = Header::read(&mut br).ok()?; + // a header that runs past its fill element is damaged: keep the one in use + (br.position() <= p.end).then_some(())?; + if self.header != Some(h) { + if self.header.is_none_or(|o| !o.same_tables(&h)) { + for c in &mut self.channels { + c.history.reset_tables(); + } + } + self.tables = FreqTables::new(&h, self.rate); + self.header = Some(h); + } + } + let (h, t) = (self.header?, self.tables.as_ref()?); + let frame = bitstream::read_data(&mut br, t, h.amp_res, self.channels.len()).ok()?; + (br.position() <= p.end).then_some(frame) + } + + /// Decode one frame: 1024 core samples per channel in, 2048 out. + pub(crate) fn process(&mut self, core: &[Vec], payload: Option) -> Vec> { + for (c, pcm) in self.channels.iter_mut().zip(core) { + c.x_low.copy_within(SLOTS..X_SLOTS, 0); + let scaled: Vec = pcm.iter().map(|s| s * PCM_SCALE).collect(); + c.analysis.process(&scaled, &mut c.x_low[T_HF_GEN..]); + } + let frame = payload.and_then(|p| self.read(p)); + let mut y: Vec> = self.channels.iter().map(|_| Box::new([[Cx::default(); 64]; X_SLOTS])).collect(); + // per channel: kx, m, and the first and end slot of `y` this frame fills + let mut current: Vec> = vec![None; self.channels.len()]; + if let (Some(frame), Some(h), Some(t)) = (&frame, self.header, self.tables.as_ref()) { + let values: Vec<_> = self + .channels + .iter_mut() + .zip(&frame.channels) + .enumerate() + .map(|(i, (c, d))| c.absolute(d, t, if frame.coupling && i == 1 { 2 } else { 1 })) + .collect(); + let levels = dequantise(frame, &values, t); + for (((c, d), (lv, y)), cur) in self.channels.iter_mut().zip(&frame.channels).zip(levels.iter().zip(&mut y)).zip(&mut current) { + let g = &d.grid; + let slots = (2 * g.t_env[0], 2 * g.t_env[g.num_env]); + let mut x_high: Box = Box::new([[Cx::default(); 64]; X_SLOTS]); + hf::generate(&c.x_low, t, &d.invf, &mut c.history, slots, &mut x_high); + let f = hf::Frame { x: &x_high, t, hdr: &h, g, lv, harmonic: d.harmonic.as_ref() }; + hf::adjust(&f, &mut c.history, y); + *cur = Some((t.kx, t.m, slots.0 + T_HF_ADJ, slots.1 + T_HF_ADJ)); + } + } + let mut out = vec![vec![0f32; 2 * 1024]; self.channels.len()]; + for (((c, y), out), current) in self.channels.iter_mut().zip(y).zip(&mut out).zip(current) { + for (l, block) in out.as_chunks_mut::<64>().0.iter_mut().enumerate() { + let i = l + T_HF_ADJ; + let mut x = [Cx::default(); 64]; + // the high band of this slot: this frame's envelopes, the last frame's reaching + // into this one, or none + let high = match current { + Some((kx, m, from, to)) if (from..to).contains(&i) => Some((&y[i], kx, m)), + _ if i + SLOTS < c.y_prev_end => Some((&c.y_prev[i + SLOTS], c.y_prev_range.0, c.y_prev_range.1)), + _ => None, + }; + let low = high.map_or(current.map_or(32, |cur| cur.0), |h| h.1); + x[..low].copy_from_slice(&c.x_low[i][..low]); + if let Some((row, kx, m)) = high { + x[kx..kx + m].copy_from_slice(&row[kx..kx + m]); + } + c.synthesis.process(&x, 64, block); + } + for s in out.iter_mut() { + *s /= PCM_SCALE; + } + c.y_prev = y; + (c.y_prev_range, c.y_prev_end) = current.map_or(((0, 0), 0), |cur| ((cur.0, cur.1), cur.3)); + } + out + } +} + +/// The header of [`test_payload`]. +#[cfg(test)] +pub(crate) const TEST_HEADER: Header = Header { + amp_res: true, + start_freq: 5, + stop_freq: 9, + xover_band: 0, + freq_scale: 2, + alter_scale: true, + noise_bands: 2, + limiter_bands: 2, + limiter_gains: 2, + interpol_freq: true, + smoothing_mode: true, +}; + +/// An SBR single channel element for tests at output rate `rate`: [`TEST_HEADER`] (3 dB steps, +/// start 5, stop 9, the default extra fields), then one FIXFIX envelope at high resolution with +/// band `k` at `env(k)` (1.5 dB steps, as a single FIXFIX envelope has) and noise band `k` at +/// `noise(k)`. +#[cfg(test)] +pub(crate) fn test_payload(rate: u32, env: impl Fn(usize) -> i32, noise: impl Fn(usize) -> i32) -> Option> { + use craft_media::bitstream::BitWriter; + let t = FreqTables::new(&TEST_HEADER, rate)?; + let (env, noise): (Vec, Vec) = ((0..t.n_high()).map(env).collect(), (0..t.n_q()).map(noise).collect()); + let mut bw = BitWriter::new(); + bw.write_bits(1, 1); // bs_header_flag + for (v, n) in [(1, 1), (5, 4), (9, 4), (0, 3), (0, 2), (0, 1), (0, 1)] { + bw.write_bits(v, n); // amp_res, start, stop, xover, reserved, no extra fields + } + bw.write_bits(0, 1); // bs_data_extra + bw.write_bits(u32::from(bitstream::FIXFIX), 2); + bw.write_bits(0, 2); // one envelope + bw.write_bits(1, 1); // high resolution + bw.write_bits(0, 2); // frequency-direction envelope and noise floor + for _ in &noise { + bw.write_bits(0, 2); // no inverse filtering + } + bw.write_bits(env[0] as u32, 7); + for w in env.windows(2) { + let (code, len) = tables::F_ENV_1_5DB[(w[1] - w[0] + 60) as usize]; + bw.write_bits(code, u32::from(len)); + } + bw.write_bits(noise[0] as u32, 5); + for w in noise.windows(2) { + let (code, len) = tables::F_ENV_3_0DB[(w[1] - w[0] + 31) as usize]; + bw.write_bits(code, u32::from(len)); + } + bw.write_bits(0, 2); // no sinusoids, no extended data + Some(bw.finish()) +} + +#[cfg(test)] +mod tests { + use std::f64::consts::PI; + + use super::*; + + const RATE: u32 = 44_100; + + fn tables() -> FreqTables { + FreqTables::new(&TEST_HEADER, RATE).unwrap() + } + + /// Mean energy per DFT bin of `x` between `lo` and `hi` QMF channels (64 per half rate). + fn energy(x: &[f32], lo: usize, hi: usize) -> f64 { + let n = x.len(); + let (b0, b1) = (lo * n / 128, hi * n / 128); + let mut e = 0f64; + for b in b0 + 2..b1 - 2 { + let (mut re, mut im) = (0f64, 0f64); + for (i, &s) in x.iter().enumerate() { + let w = 0.5 - 0.5 * (2.0 * PI * i as f64 / n as f64).cos(); + let ph = 2.0 * PI * (b * i) as f64 / n as f64; + re += f64::from(s) * w * ph.cos(); + im += f64::from(s) * w * ph.sin(); + } + e += re * re + im * im; + } + e / (b1 - b0 - 4) as f64 + } + + fn run(e: &mut Element, data: &[u8], frames: usize) -> Vec { + let core = vec![vec![0f32; 1024]]; + let mut out = Vec::new(); + for _ in 0..frames { + let p = Payload { data, start: 0, end: 8 * data.len(), crc: false }; + let block = e.process(&core, Some(p)); + assert_eq!((block.len(), block[0].len()), (1, 2048)); + assert!(block[0].iter().all(|s| s.is_finite())); + out.extend_from_slice(&block[0]); + } + out + } + + /// Over a silent core the high band is the noise the envelope asks for: the upper bands, + /// 12 steps of 1.5 dB above the lower ones, come out 18 dB louder, and nothing below `kx`. + #[test] + fn the_high_band_follows_the_envelope() { + let t = tables(); + let mid = t.n_high() / 2; + let data = test_payload(RATE, |k| if k < mid { 30 } else { 42 }, |_| 0).unwrap(); + let mut e = Element::new(1, RATE); + let out = run(&mut e, &data, 6); + let tail = &out[out.len() - 4096..]; + let (lower, upper) = (energy(tail, t.high[0], t.high[mid]), energy(tail, t.high[mid], t.high[t.n_high()])); + let step = 10.0 * (upper / lower).log10(); + assert!((step - 18.06).abs() < 1.5, "{step:.2} dB"); + let below = energy(tail, 1, t.kx - 1); + assert!(10.0 * (lower / below.max(1e-30)).log10() > 60.0); + // the same envelope without its header is not read: an element that has seen none + // passes the core through alone + let mut fresh = Element::new(1, RATE); + let mut no_header = data.clone(); + no_header[0] &= 0x7F; + assert!(run(&mut fresh, &no_header, 3).iter().all(|&s| s == 0.0)); + } + + /// Damaged SBR data (flipped bits, cut short, or noise) never panics and keeps the output + /// finite and of full length. + #[test] + fn damaged_sbr_data_is_survived() { + let good = test_payload(RATE, |k| 20 + k as i32, |_| 3).unwrap(); + let mut seed = 0x1234_5678u32; + let mut next = || { + seed ^= seed << 13; + seed ^= seed >> 17; + seed ^= seed << 5; + seed + }; + let mut e = Element::new(1, RATE); + let mut pair = Element::new(2, RATE); + for i in 0..600 { + let mut data = good.clone(); + for _ in 0..1 + next() % 6 { + let bit = next() as usize % (8 * data.len()); + data[bit / 8] ^= 0x80 >> (bit % 8); + } + if i % 3 == 0 { + data.truncate(1 + next() as usize % data.len()); + } + if i % 5 == 0 { + data = (0..1 + next() % 40).map(|_| next() as u8).collect(); + data[0] |= 0x80; + } + run(&mut e, &data, 1); + let p = Payload { data: &data, start: 0, end: 8 * data.len(), crc: i % 2 == 0 }; + let out = pair.process(&[vec![0.1; 1024], vec![-0.1; 1024]], Some(p)); + assert!(out.iter().all(|c| c.len() == 2048 && c.iter().all(|s| s.is_finite()))); + } + } + + /// The largest envelope (127 steps of 1.5 dB) with the largest noise floor (each band 31 + /// steps below the last) keeps the output finite: the gains are formed so that no + /// intermediate product of two clamped levels overflows. + #[test] + fn extreme_levels_stay_finite() { + let rate = + [22_050, 24_000, 32_000, 44_100, 48_000].into_iter().find(|&r| FreqTables::new(&TEST_HEADER, r).is_some_and(|t| t.n_q() >= 3)).unwrap(); + let data = test_payload(rate, |_| 127, |k| -31 * k as i32).unwrap(); + let mut e = Element::new(1, rate); + run(&mut e, &data, 3); + } + + /// A header that runs past the end of its fill element is not taken: the one in use stays, + /// and the frame is dropped. + #[test] + fn a_header_cut_short_is_not_taken() { + let data = test_payload(RATE, |_| 30, |_| 0).unwrap(); + let mut e = Element::new(1, RATE); + run(&mut e, &data, 1); + assert_eq!(e.header, Some(TEST_HEADER)); + // another start frequency (bits 2..6), in an element that ends 6 bits in + let mut cut = data.clone(); + cut[0] ^= 0x04; + assert!(e.read(Payload { data: &cut, start: 0, end: 6, crc: false }).is_none()); + assert_eq!(e.header, Some(TEST_HEADER)); + assert!(e.read(Payload { data: &cut, start: 0, end: 8 * cut.len(), crc: false }).is_some()); + assert_ne!(e.header, Some(TEST_HEADER)); + } +} diff --git a/crates/audio/src/aac/sbr/qmf.rs b/crates/audio/src/aac/sbr/qmf.rs new file mode 100644 index 000000000..de947f303 --- /dev/null +++ b/crates/audio/src/aac/sbr/qmf.rs @@ -0,0 +1,155 @@ +//! The SBR filter banks (ISO/IEC 14496-3 §4.6.18.4): a 32-band complex QMF analysis bank at the +//! core rate and a 64-band complex QMF synthesis bank at the output rate, sharing the window `c`. +//! +//! Each bank is a polyphase window followed by a complex modulation: analysis +//! `2·exp(iÏ€(k + ½)(2n − ½)/64)`, synthesis `exp(iÏ€(k + ½)(2n − 251)/128)/64`. With the window's +//! indexing, the two phase offsets put the banks' combined response on its centre tap, so the low +//! band comes back unchanged (to about -60 dB, the window's own limit), 576 output samples late. +//! The synthesis bank holds back its last two output samples, which makes it 578: with the SBR +//! frame's lag, the 962 samples by which HE-AAC decoders (ffmpeg's among them) delay the core, and +//! which the priming encoders signal counts on. + +use std::f64::consts::PI; +use std::sync::OnceLock; + +use super::Cx; +use super::tables::QMF_WINDOW; + +/// Output samples by which analysis followed by synthesis delays the low band. +pub(crate) const DELAY: usize = 578; + +/// Modulation matrices, computed once. +struct Kernels { + /// Analysis, at `[k][n]`: 32 Ă— 64. + analysis: Vec<[Cx; 64]>, + /// Synthesis, at `[n][k]`: 128 Ă— 64. + synthesis: Vec<[Cx; 64]>, +} + +fn kernels() -> &'static Kernels { + static K: OnceLock = OnceLock::new(); + K.get_or_init(|| { + let cx = |phase: f64, scale: f64| Cx { re: (phase.cos() * scale) as f32, im: (phase.sin() * scale) as f32 }; + let analysis = (0..32).map(|k| std::array::from_fn(|n| cx(PI * (k as f64 + 0.5) * (2.0 * n as f64 - 0.5) / 64.0, 2.0))).collect(); + let synthesis = (0..128).map(|n| std::array::from_fn(|k| cx(PI * (k as f64 + 0.5) * (2.0 * n as f64 - 251.0) / 128.0, 1.0 / 64.0))).collect(); + Kernels { analysis, synthesis } + }) +} + +/// The 32-band analysis bank of one channel. +#[derive(Clone)] +pub(crate) struct Analysis { + /// The last 320 input samples, newest first. + x: Box<[f32; 320]>, +} + +impl Analysis { + pub(crate) fn new() -> Analysis { + Analysis { x: Box::new([0.0; 320]) } + } + + /// Split `32 · out.len()` input samples into `out.len()` slots of 32 subbands. + pub(crate) fn process(&mut self, input: &[f32], out: &mut [[Cx; 32]]) { + let k = kernels(); + for (block, slot) in input.as_chunks::<32>().0.iter().zip(out.iter_mut()) { + self.x.copy_within(0..288, 32); + for (n, &s) in block.iter().enumerate() { + self.x[31 - n] = s; + } + let mut u = [0f32; 64]; + for (n, u) in u.iter_mut().enumerate() { + *u = (0..5).map(|j| self.x[n + 64 * j] * QMF_WINDOW[2 * (n + 64 * j)]).sum(); + } + for (sb, row) in slot.iter_mut().zip(&k.analysis) { + let (mut re, mut im) = (0f32, 0f32); + for (u, m) in u.iter().zip(row) { + re += u * m.re; + im += u * m.im; + } + *sb = Cx { re, im }; + } + } + } +} + +/// The 64-band synthesis bank of one channel. +#[derive(Clone)] +pub(crate) struct Synthesis { + v: Box<[f32; 1280]>, + /// The last two output samples, which the next slot puts out first. + held: [f32; 2], +} + +impl Synthesis { + pub(crate) fn new() -> Synthesis { + Synthesis { v: Box::new([0.0; 1280]), held: [0.0; 2] } + } + + /// Merge one slot of 64 subbands into 64 output samples (`out[..64]`, of which it needs 64). + /// Bands from `bands` up are taken as zero. + pub(crate) fn process(&mut self, x: &[Cx; 64], bands: usize, out: &mut [f32]) { + let k = kernels(); + let bands = bands.min(64); + self.v.copy_within(0..1152, 128); + for (n, row) in k.synthesis.iter().enumerate() { + self.v[n] = x[..bands].iter().zip(row).map(|(x, m)| x.re * m.re - x.im * m.im).sum(); + } + let mut slot = [0f32; 66]; + slot[..2].copy_from_slice(&self.held); + for (n, o) in slot[2..].iter_mut().enumerate() { + let mut s = 0f32; + for i in 0..5 { + s += self.v[256 * i + n] * QMF_WINDOW[128 * i + n]; + s += self.v[256 * i + 192 + n] * QMF_WINDOW[128 * i + 64 + n]; + } + *o = s; + } + self.held = [slot[64], slot[65]]; + for (o, s) in out.iter_mut().zip(&slot[..64]) { + *o = *s; + } + } +} + +#[cfg(test)] +mod tests { + use super::super::SLOTS; + use super::*; + + fn tones(t: f64) -> f64 { + [0.0371, 0.0913, 0.1527, 0.2211, 0.2903, 0.3519, 0.4127].iter().enumerate().map(|(j, f)| (2.0 * PI * f * t + j as f64).sin() * 0.1).sum() + } + + /// Tones across the core band through analysis and synthesis (low 32 bands only) come out at + /// twice the rate, [`DELAY`] samples late, with the error 60 dB down. + #[test] + fn analysis_then_synthesis_reconstructs_the_input() { + let input: Vec = (0..8 * 1024).map(|i| tones(i as f64) as f32).collect(); + let (mut a, mut s) = (Analysis::new(), Synthesis::new()); + let mut out = Vec::new(); + let mut slots = [[Cx::default(); 32]; SLOTS]; + for frame in input.as_chunks::<1024>().0 { + a.process(frame, &mut slots); + for slot in &slots { + let mut x = [Cx::default(); 64]; + x[..32].copy_from_slice(slot); + let mut o = [0f32; 64]; + s.process(&x, 32, &mut o); + out.extend_from_slice(&o); + } + } + assert_eq!(out.len(), 2 * input.len()); + let snr_at = |delay: usize| { + let (mut err, mut sig) = (0f64, 0f64); + for (i, &o) in out.iter().enumerate().skip(3000) { + let want = tones((i as f64 - delay as f64) / 2.0); + err += (o as f64 - want).powi(2); + sig += want * want; + } + 10.0 * (sig / err).log10() + }; + let snr = snr_at(DELAY); + assert!(snr > 60.0, "reconstruction SNR {snr:.1} dB"); + assert!(snr_at(DELAY - 1) < 40.0 && snr_at(DELAY + 1) < 40.0); + } +} diff --git a/crates/audio/src/aac/sbr/tables.rs b/crates/audio/src/aac/sbr/tables.rs new file mode 100644 index 000000000..7a648120e --- /dev/null +++ b/crates/audio/src/aac/sbr/tables.rs @@ -0,0 +1,357 @@ +//! SBR tables (ETSI ES 201 980 V1.2.2 Annex I, the SBR tables it shares with ISO/IEC 14496-3 §4.A.6). +//! +//! Transcribed from the specification's tables, which give the window to 14 digits (more than an +//! `f32` holds). Annex I misprints the labels of `c[297]` and `c[305]`; the values are in order. +//! Each Huffman table is a complete prefix code (see the tests in `sbr::bitstream`). + +#![allow(clippy::excessive_precision)] + +/// The QMF bank window `c` (640 coefficients). +#[rustfmt::skip] +pub(crate) static QMF_WINDOW: [f32; 640] = [ + -0.00055252865047, -0.00056176925738, -0.00049475180896, -0.00048752279712, -0.00048937912498, -0.00050407143497, -0.00052265642972, -0.00054665656337, + -0.00056778025613, -0.00058709304852, -0.00061327473938, -0.00063124935319, -0.00065403333621, -0.00067776907764, -0.00069416146273, -0.00071577364744, + -0.00072550431222, -0.00074409418541, -0.00074905980532, -0.00076813719270, -0.00077248485949, -0.00078343322877, -0.00077798694927, -0.00078036647100, + -0.00078014496257, -0.00077579773310, -0.00076307935757, -0.00075300014201, -0.00073193571525, -0.00072153919876, -0.00069179375372, -0.00066504150893, + -0.00063415949025, -0.00059461189330, -0.00055645763906, -0.00051455722108, -0.00046063254803, -0.00040951214522, -0.00035011758756, -0.00028969811748, + -0.00020983373440, -0.00014463809349, -0.00006173344072, 0.00001349497418, 0.00010943831274, 0.00020430170688, 0.00029495311041, 0.00040265402160, + 0.00051073884952, 0.00062393761391, 0.00074580258865, 0.00086084433262, 0.00098859883015, 0.00112501551307, 0.00125778846475, 0.00139024948272, + 0.00154432198471, 0.00168680832531, 0.00183482654224, 0.00198411407369, 0.00214615835557, 0.00230172547746, 0.00246256169126, 0.00262017586902, + 0.00278704643465, 0.00294694477165, 0.00311254206525, 0.00327396134847, 0.00344188741828, 0.00360082681231, 0.00376039229104, 0.00392074323703, + 0.00408197531935, 0.00422642692270, 0.00437307196781, 0.00452098527825, 0.00466064606118, 0.00479325608498, 0.00491376035745, 0.00503930226013, + 0.00514073539032, 0.00524611661324, 0.00534716811982, 0.00541967759307, 0.00548760401507, 0.00554757145088, 0.00559380230045, 0.00562206432097, + 0.00564551969164, 0.00563891995151, 0.00562661141932, 0.00559171286630, 0.00554043639400, 0.00547537830770, 0.00538389758970, 0.00527157587272, + 0.00513822754514, 0.00498396877629, 0.00481094690600, 0.00460395301471, 0.00438018617447, 0.00412516423270, 0.00384564081246, 0.00354012465507, + 0.00320918858098, 0.00284467578623, 0.00245085400321, 0.00202741761850, 0.00157846825768, 0.00109023290512, 0.00058322642480, 0.00002760451905, + -0.00054642808664, -0.00115681355227, -0.00180394725893, -0.00248267236449, -0.00319337783900, -0.00394011240522, -0.00472225962400, -0.00553372111088, + -0.00637922932685, -0.00726158168517, -0.00817982333726, -0.00913253296085, -0.01011502154986, -0.01113155480321, -0.01218499959508, -0.01327182200351, + 0.01439046660792, 0.01554055533423, 0.01673247129989, 0.01794333813443, 0.01918724313698, 0.02045317933555, 0.02174675502535, 0.02306801692862, + 0.02441609920285, 0.02578758475467, 0.02718594296329, 0.02860721736385, 0.03005026574279, 0.03150176087389, 0.03297540810337, 0.03446209487686, + 0.03596975605542, 0.03748128504252, 0.03900536794745, 0.04053491705584, 0.04206490946367, 0.04360975421304, 0.04514884056413, 0.04668430272642, + 0.04821657200672, 0.04973857556014, 0.05125561555216, 0.05276307465207, 0.05424527683589, 0.05571736482138, 0.05716164501299, 0.05859156836260, + 0.05998374801761, 0.06134551717207, 0.06268578081172, 0.06397158980681, 0.06522471064380, 0.06643675122104, 0.06760759851228, 0.06870438283512, + 0.06976302447127, 0.07076287107266, 0.07170026731102, 0.07256825833083, 0.07336202550803, 0.07410036424342, 0.07474525581194, 0.07531373362019, + 0.07580083586584, 0.07619924793396, 0.07649921704119, 0.07670934904245, 0.07681739756964, 0.07682300113923, 0.07672049241746, 0.07650507183194, + 0.07617483218536, 0.07573057565061, 0.07515762552870, 0.07446643947564, 0.07364060057620, 0.07267746427299, 0.07158263647903, 0.07035330735093, + 0.06896640131951, 0.06745250215166, 0.06576906686508, 0.06394448059633, 0.06196027790387, 0.05981665708090, 0.05751526919867, 0.05504600343009, + 0.05240938217366, 0.04959786763445, 0.04663033051701, 0.04347687821958, 0.04014582784127, 0.03664181168133, 0.03295839306691, 0.02908240060125, + 0.02503075618909, 0.02079970728622, 0.01637012582228, 0.01176238327857, 0.00696368621617, 0.00197656014503, -0.00320868968304, -0.00857117491366, + -0.01412888273558, -0.01988341292573, -0.02582272888064, -0.03195312745332, -0.03827765720822, -0.04478068215856, -0.05148041767934, -0.05837053268336, + -0.06544098531359, -0.07269433008129, -0.08013729344279, -0.08775475365593, -0.09555333528914, -0.10353295311463, -0.11168269317730, -0.12000779846800, + -0.12850028503878, -0.13715517611934, -0.14597664911870, -0.15496070710605, -0.16409588556669, -0.17338081721706, -0.18281725485142, -0.19239667457267, + -0.20212501768103, -0.21197358538056, -0.22196526964149, -0.23206908706791, -0.24230168845974, -0.25264803095722, -0.26310532994603, -0.27366340405625, + -0.28432141891085, -0.29507167170646, -0.30590985751916, -0.31682789136456, -0.32781137272105, -0.33887226938665, -0.34999141229310, -0.36115899031355, + 0.37237955463061, 0.38363500139043, 0.39492117615675, 0.40623176767625, 0.41756968968409, 0.42891199207373, 0.44025537543665, 0.45159965356824, + 0.46293080852757, 0.47424532146115, 0.48552530911099, 0.49677082545707, 0.50798175000434, 0.51912349702391, 0.53022408956855, 0.54125534487322, + 0.55220512585061, 0.56307891401370, 0.57385241316923, 0.58454032354679, 0.59511230862496, 0.60557835389180, 0.61591099320291, 0.62612426956055, + 0.63619801077286, 0.64612696959461, 0.65590163024671, 0.66551398801627, 0.67496631901712, 0.68423532934598, 0.69332823767032, 0.70223887193539, + 0.71094104263095, 0.71944626349561, 0.72774489002994, 0.73582117582769, 0.74368278636488, 0.75131374561237, 0.75870807608242, 0.76586748650939, + 0.77277808813327, 0.77942875190216, 0.78583531203920, 0.79197358416424, 0.79784664137700, 0.80344857518505, 0.80876950044491, 0.81381912706217, + 0.81857760046468, 0.82304198905409, 0.82722753473360, 0.83110384571520, 0.83469373618402, 0.83797173378865, 0.84095413924722, 0.84362382812005, + 0.84598184698206, 0.84803157770763, 0.84978051984268, 0.85119715249343, 0.85230470352147, 0.85310209497017, 0.85357205739107, 0.85373856005937, + 0.85357205739107, 0.85310209497017, 0.85230470352147, 0.85119715249343, 0.84978051984268, 0.84803157770763, 0.84598184698206, 0.84362382812005, + 0.84095413924722, 0.83797173378865, 0.83469373618402, 0.83110384571520, 0.82722753473360, 0.82304198905409, 0.81857760046468, 0.81381912706217, + 0.80876950044491, 0.80344857518505, 0.79784664137700, 0.79197358416424, 0.78583531203920, 0.77942875190216, 0.77277808813327, 0.76586748650939, + 0.75870807608242, 0.75131374561237, 0.74368278636488, 0.73582117582769, 0.72774489002994, 0.71944626349561, 0.71094104263095, 0.70223887193539, + 0.69332823767032, 0.68423532934598, 0.67496631901712, 0.66551398801627, 0.65590163024671, 0.64612696959461, 0.63619801077286, 0.62612426956055, + 0.61591099320291, 0.60557835389180, 0.59511230862496, 0.58454032354679, 0.57385241316923, 0.56307891401370, 0.55220512585061, 0.54125534487322, + 0.53022408956855, 0.51912349702391, 0.50798175000434, 0.49677082545707, 0.48552530911099, 0.47424532146115, 0.46293080852757, 0.45159965356824, + 0.44025537543665, 0.42891199207373, 0.41756968968409, 0.40623176767625, 0.39492117615675, 0.38363500139043, 0.37237955463061, 0.36115899031355, + -0.34999141229310, -0.33887226938665, -0.32781137272105, -0.31682789136456, -0.30590985751916, -0.29507167170646, -0.28432141891085, -0.27366340405625, + -0.26310532994603, -0.25264803095722, -0.24230168845974, -0.23206908706791, -0.22196526964149, -0.21197358538056, -0.20212501768103, -0.19239667457267, + -0.18281725485142, -0.17338081721706, -0.16409588556669, -0.15496070710605, -0.14597664911870, -0.13715517611934, -0.12850028503878, -0.12000779846800, + -0.11168269317730, -0.10353295311463, -0.09555333528914, -0.08775475365593, -0.08013729344279, -0.07269433008129, -0.06544098531359, -0.05837053268336, + -0.05148041767934, -0.04478068215856, -0.03827765720822, -0.03195312745332, -0.02582272888064, -0.01988341292573, -0.01412888273558, -0.00857117491366, + -0.00320868968304, 0.00197656014503, 0.00696368621617, 0.01176238327857, 0.01637012582228, 0.02079970728622, 0.02503075618909, 0.02908240060125, + 0.03295839306691, 0.03664181168133, 0.04014582784127, 0.04347687821958, 0.04663033051701, 0.04959786763445, 0.05240938217366, 0.05504600343009, + 0.05751526919867, 0.05981665708090, 0.06196027790387, 0.06394448059633, 0.06576906686508, 0.06745250215166, 0.06896640131951, 0.07035330735093, + 0.07158263647903, 0.07267746427299, 0.07364060057620, 0.07446643947564, 0.07515762552870, 0.07573057565061, 0.07617483218536, 0.07650507183194, + 0.07672049241746, 0.07682300113923, 0.07681739756964, 0.07670934904245, 0.07649921704119, 0.07619924793396, 0.07580083586584, 0.07531373362019, + 0.07474525581194, 0.07410036424342, 0.07336202550803, 0.07256825833083, 0.07170026731102, 0.07076287107266, 0.06976302447127, 0.06870438283512, + 0.06760759851228, 0.06643675122104, 0.06522471064380, 0.06397158980681, 0.06268578081172, 0.06134551717207, 0.05998374801761, 0.05859156836260, + 0.05716164501299, 0.05571736482138, 0.05424527683589, 0.05276307465207, 0.05125561555216, 0.04973857556014, 0.04821657200672, 0.04668430272642, + 0.04514884056413, 0.04360975421304, 0.04206490946367, 0.04053491705584, 0.03900536794745, 0.03748128504252, 0.03596975605542, 0.03446209487686, + 0.03297540810337, 0.03150176087389, 0.03005026574279, 0.02860721736385, 0.02718594296329, 0.02578758475467, 0.02441609920285, 0.02306801692862, + 0.02174675502535, 0.02045317933555, 0.01918724313698, 0.01794333813443, 0.01673247129989, 0.01554055533423, 0.01439046660792, 0.01327182200351, + -0.01218499959508, -0.01113155480321, -0.01011502154986, -0.00913253296085, -0.00817982333726, -0.00726158168517, -0.00637922932685, -0.00553372111088, + -0.00472225962400, -0.00394011240522, -0.00319337783900, -0.00248267236449, -0.00180394725893, -0.00115681355227, -0.00054642808664, 0.00002760451905, + 0.00058322642480, 0.00109023290512, 0.00157846825768, 0.00202741761850, 0.00245085400321, 0.00284467578623, 0.00320918858098, 0.00354012465507, + 0.00384564081246, 0.00412516423270, 0.00438018617447, 0.00460395301471, 0.00481094690600, 0.00498396877629, 0.00513822754514, 0.00527157587272, + 0.00538389758970, 0.00547537830770, 0.00554043639400, 0.00559171286630, 0.00562661141932, 0.00563891995151, 0.00564551969164, 0.00562206432097, + 0.00559380230045, 0.00554757145088, 0.00548760401507, 0.00541967759307, 0.00534716811982, 0.00524611661324, 0.00514073539032, 0.00503930226013, + 0.00491376035745, 0.00479325608498, 0.00466064606118, 0.00452098527825, 0.00437307196781, 0.00422642692270, 0.00408197531935, 0.00392074323703, + 0.00376039229104, 0.00360082681231, 0.00344188741828, 0.00327396134847, 0.00311254206525, 0.00294694477165, 0.00278704643465, 0.00262017586902, + 0.00246256169126, 0.00230172547746, 0.00214615835557, 0.00198411407369, 0.00183482654224, 0.00168680832531, 0.00154432198471, 0.00139024948272, + 0.00125778846475, 0.00112501551307, 0.00098859883015, 0.00086084433262, 0.00074580258865, 0.00062393761391, 0.00051073884952, 0.00040265402160, + 0.00029495311041, 0.00020430170688, 0.00010943831274, 0.00001349497418, -0.00006173344072, -0.00014463809349, -0.00020983373440, -0.00028969811748, + -0.00035011758756, -0.00040951214522, -0.00046063254803, -0.00051455722108, -0.00055645763906, -0.00059461189330, -0.00063415949025, -0.00066504150893, + -0.00069179375372, -0.00072153919876, -0.00073193571525, -0.00075300014201, -0.00076307935757, -0.00077579773310, -0.00078014496257, -0.00078036647100, + -0.00077798694927, -0.00078343322877, -0.00077248485949, -0.00076813719270, -0.00074905980532, -0.00074409418541, -0.00072550431222, -0.00071577364744, + -0.00069416146273, -0.00067776907764, -0.00065403333621, -0.00063124935319, -0.00061327473938, -0.00058709304852, -0.00056778025613, -0.00054665656337, + -0.00052265642972, -0.00050407143497, -0.00048937912498, -0.00048752279712, -0.00049475180896, -0.00056176925738, -0.00055252865047, 0.00000000000000, +]; + +/// The SBR noise table `V` (Table I.13): 512 unit-variance complex values (re, im). +#[rustfmt::skip] +pub(crate) static NOISE: [[f32; 2]; 512] = [ + [-0.99948153278296, -0.59483417516607], [0.97113454393991, -0.67528515225647], [0.14130051758487, -0.95090983575689], [-0.47005496701697, -0.37340549728647], + [0.80705063769351, 0.29653668284408], [-0.38981478896926, 0.89572605717087], [-0.01053049862020, -0.66959058036166], [-0.91266367957293, -0.11522938140034], + [0.54840422910309, 0.75221367176302], [0.40009252867955, -0.98929400334421], [-0.99867974711855, -0.88147068645358], [-0.95531076805040, 0.90908757154593], + [-0.45725933317144, -0.56716323646760], [-0.72929675029275, -0.98008272727324], [0.75622801399036, 0.20950329995549], [0.07069442601050, -0.78247898470706], + [0.74496252926055, -0.91169004445807], [-0.96440182703856, -0.94739918296622], [0.30424629369539, -0.49438267012479], [0.66565033746925, 0.64652935542491], + [0.91697008020594, 0.17514097332009], [-0.70774918760427, 0.52548653416543], [-0.70051415345560, -0.45340028808763], [-0.99496513054797, -0.90071908066973], + [0.98164490790123, -0.77463155528697], [-0.54671580548181, -0.02570928536004], [-0.01689629065389, 0.00287506445732], [-0.86110349531986, 0.42548583726477], + [-0.98892980586032, -0.87881132267556], [0.51756627678691, 0.66926784710139], [-0.99635026409640, -0.58107730574765], [-0.99969370862163, 0.98369989360250], + [0.55266258627194, 0.59449057465591], [0.34581177741673, 0.94879421061866], [0.62664209577999, -0.74402970906471], [-0.77149701404973, -0.33883658042801], + [-0.91592244254432, 0.03687901376713], [-0.76285492357887, -0.91371867919124], [0.79788337195331, -0.93180971199849], [0.54473080610200, -0.11919206037186], + [-0.85639281671058, 0.42429854760451], [-0.92882402971423, 0.27871809078609], [-0.11708371046774, -0.99800843444966], [0.21356749817493, -0.90716295627033], + [-0.76191692573909, 0.99768118356265], [0.98111043100884, -0.95854459734407], [-0.85913269895572, 0.95766566168880], [-0.93307242253692, 0.49431757696466], + [0.30485754879632, -0.70540034357529], [0.85289650925190, 0.46766131791044], [0.91328082618125, -0.99839597361769], [-0.05890199924154, 0.70741827819497], + [0.28398686150148, 0.34633555702188], [0.95258164539612, -0.54893416026939], [-0.78566324168507, -0.75568541079691], [-0.95789495447877, -0.20423194696966], + [0.82411158711197, 0.96654618432562], [-0.65185446735885, -0.88734990773289], [-0.93643603134666, 0.99870790442385], [0.91427159529618, -0.98290505544444], + [-0.70395684036886, 0.58796798221039], [0.00563771969365, 0.61768196727244], [0.89065051931895, 0.52783352697585], [-0.68683707712762, 0.80806944710339], + [0.72165342518718, -0.69259857349564], [-0.62928247730667, 0.13627037407335], [0.29938434065514, -0.46051329682246], [-0.91781958879280, -0.74012716684186], + [0.99298717043688, 0.40816610075661], [0.82368298622748, -0.74036047190173], [-0.98512833386833, -0.99972330709594], [-0.95915368242257, -0.99237800466040], + [-0.21411126572790, -0.93424819052545], [-0.68821476106884, -0.26892306315457], [0.91851997982317, 0.09358228901785], [-0.96062769559127, 0.36099095133739], + [0.51646184922287, -0.71373332873917], [0.61130721139669, 0.46950141175917], [0.47336129371299, -0.27333178296162], [0.90998308703519, 0.96715662938132], + [0.44844799194357, 0.99211574628306], [0.66614891079092, 0.96590176169121], [0.74922239129237, -0.89879858826087], [-0.99571588506485, 0.52785521494349], + [0.97401082477563, -0.16855870075190], [0.72683747733879, -0.48060774432251], [0.95432193457128, 0.68849603408441], [-0.72962208425191, -0.76608443420917], + [-0.85359479233537, 0.88738125901579], [-0.81412430338535, -0.97480768049637], [-0.87930772356786, 0.74748307690436], [-0.71573331064977, -0.98570608178923], + [0.83524300028228, 0.83702537075163], [-0.48086065601423, -0.98848504923531], [0.97139128574778, 0.80093621198236], [0.51992825347895, 0.80247631400510], + [-0.00848591195325, -0.76670128000486], [-0.70294374303036, 0.55359910445577], [-0.95894428168140, -0.43265504344783], [0.97079252950321, 0.09325857238682], + [-0.92404293670797, 0.85507704027855], [-0.69506469500450, 0.98633412625459], [0.26559203620024, 0.73314307966524], [0.28038443336943, 0.14537913654427], + [-0.74138124825523, 0.99310339807762], [-0.01752795995444, -0.82616635284178], [-0.55126773094930, -0.98898543862153], [0.97960898850996, -0.94021446752851], + [-0.99196309146936, 0.67019017358456], [-0.67684928085260, 0.12631491649378], [0.09140039465500, -0.20537731453108], [-0.71658965751996, -0.97788200391224], + [0.81014640078925, 0.53722648362443], [0.40616991671205, -0.26469008598449], [-0.67680188682972, 0.94502052337695], [0.86849774348749, -0.18333598647899], + [-0.99500381284851, -0.02634122068550], [0.84329189340667, 0.10406957462213], [-0.09215968531446, 0.69540012101253], [0.99956173327206, -0.12358542001404], + [-0.79732779473535, -0.91582524736159], [0.96349973642406, 0.96640458041000], [-0.79942778496547, 0.64323902822857], [-0.11566039853896, 0.28587846253726], + [-0.39922954514662, 0.94129601616966], [0.99089197565987, -0.92062625581587], [0.28631285179909, -0.91035047143603], [-0.83302725605608, -0.67330410892084], + [0.95404443402072, 0.49162765398743], [-0.06449863579434, 0.03250560813135], [-0.99575054486311, 0.42389784469507], [-0.65501142790847, 0.82546114655624], + [-0.81254441908887, -0.51627234660629], [-0.99646369485481, 0.84490533520752], [0.00287840603348, 0.64768261158166], [0.70176989408455, -0.20453028573322], + [0.96361882270190, 0.40706967140989], [-0.68883758192426, 0.91338958840772], [-0.34875585502238, 0.71472290693300], [0.91980081243087, 0.66507455644919], + [-0.99009048343881, 0.85868021604848], [0.68865791458395, 0.55660316809678], [-0.99484402129368, -0.20052559254934], [0.94214511408023, -0.99696425367461], + [-0.67414626793544, 0.49548221180078], [-0.47339353684664, -0.85904328834047], [0.14323651387360, -0.94145598222488], [-0.29268293575672, 0.05759224927952], + [0.43793861458754, -0.78904969892724], [-0.36345126374441, 0.64874435357162], [-0.08750604656825, 0.97686944362527], [-0.96495267812511, -0.53960305946511], + [0.55526940659947, 0.78891523734774], [0.73538215752630, 0.96452072373404], [-0.30889773919437, -0.80664389776860], [0.03574995626194, -0.97325616900959], + [0.98720684660488, 0.48409133691962], [-0.81689296271203, -0.90827703628298], [0.67866860118215, 0.81284503870856], [-0.15808569732583, 0.85279555024382], + [0.80723395114371, -0.24717418514605], [0.47788757329038, -0.46333147839295], [0.96367554763201, 0.38486749303242], [-0.99143875716818, -0.24945277239809], + [0.83081876925833, -0.94780851414763], [-0.58753191905341, 0.01290772389163], [0.95538108220960, -0.85557052096538], [-0.96490920476211, -0.64020970923102], + [-0.97327101028521, 0.12378128133110], [0.91400366022124, 0.57972471346930], [-0.99925837363824, 0.71084847864067], [-0.86875903507313, -0.20291699203564], + [-0.26240034795124, -0.68264554369108], [-0.24664412953388, -0.87642273115183], [0.02416275806869, 0.27192914288905], [0.82068619590515, -0.85087787994476], + [0.88547373760759, -0.89636802901469], [-0.18173078152226, -0.26152145156800], [0.09355476558534, 0.54845123045604], [-0.54668414224090, 0.95980774020221], + [0.37050990604091, -0.59910140383171], [-0.70373594262891, 0.91227665827081], [-0.34600785879594, -0.99441426144200], [-0.68774481731008, -0.30238837956299], + [-0.26843291251234, 0.83115668004362], [0.49072334613242, -0.45359708737775], [0.38975993093975, 0.95515358099121], [-0.97757125224150, 0.05305894580606], + [-0.17325552859616, -0.92770672250494], [0.99948035025744, 0.58285545563426], [-0.64946246527458, 0.68645507104960], [-0.12016920576437, -0.57147322153312], + [-0.58947456517751, -0.34847132454388], [-0.41815140454465, 0.16276422358861], [0.99885650204884, 0.11136095490444], [-0.56649614128386, -0.90494866361587], + [0.94138021032330, 0.35281916733018], [-0.75725076534641, 0.53650549640587], [0.20541973692630, -0.94435144369918], [0.99980371023351, 0.79835913565599], + [0.29078277605775, 0.35393777921520], [-0.62858772103030, 0.38765693387102], [0.43440904467688, -0.98546330463232], [-0.98298583762390, 0.21021524625209], + [0.19513029146934, -0.94239832251867], [-0.95476662400101, 0.98364554179143], [0.93379635304810, -0.70881994583682], [-0.85235410573336, -0.08342347966410], + [-0.86425093011245, -0.45795025029466], [0.38879779059045, 0.97274429344593], [0.92045124735495, -0.62433652524220], [0.89162532251878, 0.54950955570563], + [-0.36834336949252, 0.96458298020975], [0.93891760988045, -0.89968353740388], [0.99267657565094, -0.03757034316958], [-0.94063471614176, 0.41332338538963], + [0.99740224117019, -0.16830494996370], [-0.35899413170555, -0.46633226649613], [0.05237237274947, -0.25640361602661], [0.36703583957424, -0.38653265641875], + [0.91653180367913, -0.30587628726597], [0.69000803499316, 0.90952171386132], [-0.38658751133527, 0.99501571208985], [-0.29250814029851, 0.37444994344615], + [-0.60182204677608, 0.86779651036123], [-0.97418588163217, 0.96468523666475], [0.88461574003963, 0.57508405276414], [0.05198933055162, 0.21269661669964], + [-0.53499621979720, 0.97241553731237], [-0.49429560226497, 0.98183865291903], [-0.98935142339139, -0.40249159006933], [-0.98081380091130, -0.72856895534041], + [-0.27338148835532, 0.99950922447209], [0.06310802338302, -0.54539587529618], [-0.20461677199539, -0.14209977628489], [0.66223843141647, 0.72528579940326], + [-0.84764345483665, 0.02372316801261], [-0.89039863483811, 0.88866581484602], [0.95903308477986, 0.76744927173873], [0.73504123909879, -0.03747203173192], + [-0.31744434966056, -0.36834111883652], [-0.34110827591623, 0.40211222807691], [0.47803883714199, -0.39423219786288], [0.98299195879514, 0.01989791390047], + [-0.30963073129751, -0.18076720599336], [0.99992588229018, -0.26281872094289], [-0.93149731080767, -0.98313162570490], [0.99923472302773, -0.80142993767554], + [-0.26024169633417, -0.75999759855752], [-0.35712514743563, 0.19298963768574], [-0.99899084509530, 0.74645156992493], [0.86557171579452, 0.55593866696299], + [0.33408042438752, 0.86185953874709], [0.99010736374716, 0.04602397576623], [-0.66694269691195, -0.91643611810148], [0.64016792079480, 0.15649530836856], + [0.99570534804836, 0.45844586038111], [-0.63431466947340, 0.21079116459234], [-0.07706847005931, -0.89581437101329], [0.98590090577724, 0.88241721133981], + [0.80099335254678, -0.36851896710853], [0.78368131392666, 0.45506999802597], [0.08707806671691, 0.80938994918745], [-0.86811883080712, 0.39347308654705], + [-0.39466529740375, -0.66809432114456], [0.97875325649683, -0.72467840967746], [-0.95038560288864, 0.89563219587625], [0.17005239424212, 0.54683053962658], + [-0.76910792026848, -0.96226617549298], [0.99743281016846, 0.42697157037567], [0.95437383549973, 0.97002324109952], [0.99578905365569, -0.54106826257356], + [0.28058259829990, -0.85361420634036], [0.85256524470573, -0.64567607735589], [-0.50608540105128, -0.65846015480300], [-0.97210735183243, -0.23095213067791], + [0.95424048234441, -0.99240147091219], [-0.96926570524023, 0.73775654896574], [0.30872163214726, 0.41514960556126], [-0.24523839572639, 0.63206633394807], + [-0.33813265086024, -0.38661779441897], [-0.05826828420146, -0.06940774188029], [-0.22898461455054, 0.97054853316316], [-0.18509915019881, 0.47565762892084], + [-0.10488238045009, -0.87769947402394], [-0.71886586182037, 0.78030982480538], [0.99793873738654, 0.90041310491497], [0.57563307626120, -0.91034337352097], + [0.28909646383717, 0.96307783970534], [0.42188998312520, 0.48148651230437], [0.93335049681047, -0.43537023883588], [-0.97087374418267, 0.86636445711364], + [0.36722871286923, 0.65291654172961], [-0.81093025665696, 0.08778370229363], [-0.26240603062237, -0.92774095379098], [0.83996497984604, 0.55839849139647], + [-0.99909615720225, -0.96024605713970], [0.74649464155061, 0.12144893606462], [-0.74774595569805, -0.26898062008959], [0.95781667469567, -0.79047927052628], + [0.95472308713099, -0.08588776019550], [0.48708332746299, 0.99999041579432], [0.46332038247497, 0.10964126185063], [-0.76497004940162, 0.89210929242238], + [0.57397389364339, 0.35289703373760], [0.75374316974495, 0.96705214651335], [-0.59174397685714, -0.89405370422752], [0.75087906691890, -0.29612672982396], + [-0.98607857336230, 0.25034911730023], [-0.40761056640505, -0.90045573444695], [0.66929266740477, 0.98629493401748], [-0.97463695257310, -0.00190223301301], + [0.90145509409859, 0.99781390365446], [-0.87259289048043, 0.99233587353666], [-0.91529461447692, -0.15698707534206], [-0.03305738840705, -0.37205262859764], + [0.07223051368337, -0.88805001733626], [0.99498012188353, 0.97094358113387], [-0.74904939500519, 0.99985483641521], [0.04585228574211, 0.99812337444082], + [-0.89054954257993, -0.31791913188064], [-0.83782144651251, 0.97637632547466], [0.33454804933804, -0.86231516800408], [-0.99707579362824, 0.93237990079441], + [-0.22827527843994, 0.18874759397997], [0.67248046289143, -0.03646211390569], [-0.05146538187944, -0.92599700120679], [0.99947295749905, 0.93625229707912], + [0.66951124390363, 0.98905825623893], [-0.99602956559179, -0.44654715757688], [0.82104905483590, 0.99540741724928], [0.99186510988782, 0.72023001312947], + [-0.65284592392918, 0.52186723253637], [0.93885443798188, -0.74895312615259], [0.96735248738388, 0.90891816978629], [-0.22225968841114, 0.57124029781228], + [-0.44132783753414, -0.92688840659280], [-0.85694974219574, 0.88844532719844], [0.91783042091762, -0.46356892383970], [0.72556974415690, -0.99899555770747], + [-0.99711581834508, 0.58211560180426], [0.77638976371966, 0.94321834873819], [0.07717324253925, 0.58638399856595], [-0.56049829194163, 0.82522301569036], + [0.98398893639988, 0.39467440420569], [0.47546946844938, 0.68613044836811], [0.65675089314631, 0.18331637134880], [0.03273375457980, -0.74933109564108], + [-0.38684144784738, 0.51337349030406], [-0.97346267944545, -0.96549364384098], [-0.53282156061942, -0.91423265091354], [0.99817310731176, 0.61133572482148], + [-0.50254500772635, -0.88829338134294], [0.01995873238855, 0.85223515096765], [0.99930381973804, 0.94578896296649], [0.82907767600783, -0.06323442598128], + [-0.58660709669728, 0.96840773806582], [-0.17573736667267, -0.48166920859485], [0.83434292401346, -0.13023450646997], [0.05946491307025, 0.20511047074866], + [0.81505484574602, -0.94685947861369], [-0.44976380954860, 0.40894572671545], [-0.89746474625671, 0.99846578838537], [0.39677256130792, -0.74854668609359], + [-0.07588948563079, 0.74096214084170], [0.76343198951445, 0.41746629422634], [-0.74490104699626, 0.94725911744610], [0.64880119792759, 0.41336660830571], + [0.62319537462542, -0.93098313552599], [0.42215817594807, -0.07712787385208], [0.02704554141885, -0.05417518053666], [0.80001773566818, 0.91542195141039], + [-0.79351832348816, -0.36208897989136], [0.63872359151636, 0.08128252493444], [0.52890520960295, 0.60048872455592], [0.74238552914587, 0.04491915291044], + [0.99096131449250, -0.19451182854402], [-0.80412329643109, -0.88513818199457], [-0.64612616129736, 0.72198674804544], [0.11657770663191, -0.83662833815041], + [-0.95053182488101, -0.96939905138082], [-0.62228872928622, 0.82767262846661], [0.03004475787316, -0.99738896333384], [-0.97987214341034, 0.36526129686425], + [-0.99986980746200, -0.36021610299715], [0.89110648599879, -0.97894250343044], [0.10407960510582, 0.77357793811619], [0.95964737821728, -0.35435818285502], + [0.50843233159162, 0.96107691266205], [0.17006334670615, -0.76854025314829], [0.25872675063360, 0.99893303933816], [-0.01115998681937, 0.98496019742444], + [-0.79598702973261, 0.97138411318894], [-0.99264708948101, -0.99542822402536], [-0.99829663752818, 0.01877138824311], [-0.70801016548184, 0.33680685948117], + [-0.70467057786826, 0.93272777501857], [0.99846021905254, -0.98725746254433], [-0.63364968534650, -0.16473594423746], [-0.16258217500792, -0.95939125400802], + [-0.43645594360633, -0.94805030113284], [-0.99848471702976, 0.96245166923809], [-0.16796458968998, -0.98987511890470], [-0.87979225745213, -0.71725725041680], + [0.44183099021786, -0.93568974498761], [0.93310180125532, -0.99913308068246], [-0.93941931782002, -0.56409379640356], [-0.88590003188677, 0.47624600491382], + [0.99971463703691, -0.83889954253462], [-0.75376385639978, 0.00814643438625], [0.93887685615875, -0.11284528204636], [0.85126435782309, 0.52349251543547], + [0.39701421446381, 0.81779634174316], [-0.37024464187437, -0.87071656222959], [-0.36024828242896, 0.34655735648287], [-0.93388812549209, -0.84476541096429], + [-0.65298804552119, -0.18439575450921], [0.11960319006843, 0.99899346780168], [0.94292565553160, 0.83163906518293], [0.75081145286948, -0.35533223142265], + [0.56721979748394, -0.24076836414499], [0.46857766746029, -0.30140233457198], [0.97312313923635, -0.99548191630031], [-0.38299976567017, 0.98516909715427], + [0.41025800019463, 0.02116736935734], [0.09638062008048, 0.04411984381457], [-0.85283249275397, 0.91475563922421], [0.88866808958124, -0.99735267083226], + [-0.48202429536989, -0.96805608884164], [0.27572582416567, 0.58634753335832], [-0.65889129659168, 0.58835634138583], [0.98838086953732, 0.99994349600236], + [-0.20651349620689, 0.54593044066355], [-0.62126416356920, -0.59893681700392], [0.20320105410437, -0.86879180355289], [-0.97790548600584, 0.96290806999242], + [0.11112534735126, 0.21484763313301], [-0.41368337314182, 0.28216837680365], [0.24133038992960, 0.51294362630238], [-0.66393410674885, -0.08249679629081], + [-0.53697829178752, -0.97649903936228], [-0.97224737889348, 0.22081333579837], [0.87392477144549, -0.12796173740361], [0.19050361015753, 0.01602615387195], + [-0.46353441212724, -0.95249041539006], [-0.07064096339021, -0.94479803205886], [-0.92444085484466, -0.10457590187436], [-0.83822593578728, -0.01695043208885], + [0.75214681811150, -0.99955681042665], [-0.42102998829339, 0.99720941999394], [-0.72094786237696, -0.35008961934255], [0.78843311019251, 0.52851398958271], + [0.97394027897442, -0.26695944086561], [0.99206463477946, -0.57010120849429], [0.76789609461795, -0.76519356730966], [-0.82002421836409, -0.73530179553767], + [0.81924990025724, 0.99698425250579], [-0.26719850873357, 0.68903369776193], [-0.43311260380975, 0.85321815947490], [0.99194979673836, 0.91876249766422], + [-0.80692001248487, -0.32627540663214], [0.43080003649976, -0.21919095636638], [0.67709491937357, -0.95478075822906], [0.56151770568316, -0.70693811747778], + [0.10831862810749, -0.08628837174592], [0.91229417540436, -0.65987351408410], [-0.48972893932274, 0.56289246362686], [-0.89033658689697, -0.71656563987082], + [0.65269447475094, 0.65916004833932], [0.67439478141121, -0.81684380846796], [-0.47770832416973, -0.16789556203025], [-0.99715979260878, -0.93565784007648], + [-0.90889593602546, 0.62034397054380], [-0.06618622548177, -0.23812217221359], [0.99430266919728, 0.18812555317553], [0.97686402381843, -0.28664534366620], + [0.94813650221268, -0.97506640027128], [-0.95434497492853, -0.79607978501983], [-0.49104783137150, 0.32895214359663], [0.99881175120751, 0.88993983831354], + [0.50449166760303, -0.85995072408434], [0.47162891065108, -0.18680204049569], [-0.62081581361840, 0.75000676218956], [-0.43867015250812, 0.99998069244322], + [0.98630563232075, -0.53578899600662], [-0.61510362277374, -0.89515019899997], [-0.03841517601843, -0.69888815681179], [-0.30102157304644, -0.07667808922205], + [0.41881284182683, 0.02188098922282], [-0.86135454941237, 0.98947480909359], [0.67226861393788, -0.13494389011014], [-0.70737398842068, -0.76547349325992], + [0.94044946687963, 0.09026201157416], [-0.82386352534327, 0.08924768823676], [-0.32070666698656, 0.50143421908753], [0.57593163224487, -0.98966422921509], + [-0.36326018419965, 0.07440243123228], [0.99979044674350, -0.14130287347405], [-0.92366023326932, -0.97979298068180], [-0.44607178518598, -0.54233252016394], + [0.44226800932956, 0.71326756742752], [0.03671907158312, 0.63606389366675], [0.52175424682195, -0.85396826735705], [-0.94701139690956, -0.01826348194255], + [-0.98759606946049, 0.82288714303073], [0.87434794743625, 0.89399495655433], [-0.93412041758744, 0.41374052024363], [0.96063943315511, 0.93116709541280], + [0.97534253457837, 0.86150930812689], [0.99642466504163, 0.70190043427512], [-0.94705089665984, -0.29580042814306], [0.91599807087376, -0.98147830385781], +]; + +/// `t_huffman_env_1_5dB` (`(codeword, length)` by index): envelope, time direction, 1.5 dB steps (LAV 60). +#[rustfmt::skip] +pub(crate) static T_ENV_1_5DB: [(u32, u8); 121] = [ + (0x3ffd6, 18), (0x3ffd7, 18), (0x3ffd8, 18), (0x3ffd9, 18), (0x3ffda, 18), (0x3ffdb, 18), (0x7ffb8, 19), (0x7ffb9, 19), + (0x7ffba, 19), (0x7ffbb, 19), (0x7ffbc, 19), (0x7ffbd, 19), (0x7ffbe, 19), (0x7ffbf, 19), (0x7ffc0, 19), (0x7ffc1, 19), + (0x7ffc2, 19), (0x7ffc3, 19), (0x7ffc4, 19), (0x7ffc5, 19), (0x7ffc6, 19), (0x7ffc7, 19), (0x7ffc8, 19), (0x7ffc9, 19), + (0x7ffca, 19), (0x7ffcb, 19), (0x7ffcc, 19), (0x7ffcd, 19), (0x7ffce, 19), (0x7ffcf, 19), (0x7ffd0, 19), (0x7ffd1, 19), + (0x7ffd2, 19), (0x7ffd3, 19), (0x1ffe6, 17), (0x3ffd4, 18), (0xfff0, 16), (0x1ffe9, 17), (0x3ffd5, 18), (0x1ffe7, 17), + (0xfff1, 16), (0xffec, 16), (0xffed, 16), (0xffee, 16), (0x7ff4, 15), (0x3ff9, 14), (0x3ff7, 14), (0x1ffa, 13), + (0x1ff9, 13), (0xffb, 12), (0x7fc, 11), (0x3fc, 10), (0x1fd, 9), (0xfd, 8), (0x7d, 7), (0x3d, 6), + (0x1d, 5), (0xd, 4), (0x5, 3), (0x1, 2), (0x0, 2), (0x4, 3), (0xc, 4), (0x1c, 5), + (0x3c, 6), (0x7c, 7), (0xfc, 8), (0x1fc, 9), (0x3fd, 10), (0xffa, 12), (0x1ff8, 13), (0x3ff6, 14), + (0x3ff8, 14), (0x7ff5, 15), (0xffef, 16), (0x1ffe8, 17), (0xfff2, 16), (0x7ffd4, 19), (0x7ffd5, 19), (0x7ffd6, 19), + (0x7ffd7, 19), (0x7ffd8, 19), (0x7ffd9, 19), (0x7ffda, 19), (0x7ffdb, 19), (0x7ffdc, 19), (0x7ffdd, 19), (0x7ffde, 19), + (0x7ffdf, 19), (0x7ffe0, 19), (0x7ffe1, 19), (0x7ffe2, 19), (0x7ffe3, 19), (0x7ffe4, 19), (0x7ffe5, 19), (0x7ffe6, 19), + (0x7ffe7, 19), (0x7ffe8, 19), (0x7ffe9, 19), (0x7ffea, 19), (0x7ffeb, 19), (0x7ffec, 19), (0x7ffed, 19), (0x7ffee, 19), + (0x7ffef, 19), (0x7fff0, 19), (0x7fff1, 19), (0x7fff2, 19), (0x7fff3, 19), (0x7fff4, 19), (0x7fff5, 19), (0x7fff6, 19), + (0x7fff7, 19), (0x7fff8, 19), (0x7fff9, 19), (0x7fffa, 19), (0x7fffb, 19), (0x7fffc, 19), (0x7fffd, 19), (0x7fffe, 19), + (0x7ffff, 19), +]; + +/// `f_huffman_env_1_5dB` (`(codeword, length)` by index): envelope, frequency direction, 1.5 dB steps (LAV 60). +#[rustfmt::skip] +pub(crate) static F_ENV_1_5DB: [(u32, u8); 121] = [ + (0x7ffe7, 19), (0x7ffe8, 19), (0xfffd2, 20), (0xfffd3, 20), (0xfffd4, 20), (0xfffd5, 20), (0xfffd6, 20), (0xfffd7, 20), + (0xfffd8, 20), (0x7ffda, 19), (0xfffd9, 20), (0xfffda, 20), (0xfffdb, 20), (0xfffdc, 20), (0x7ffdb, 19), (0xfffdd, 20), + (0x7ffdc, 19), (0x7ffdd, 19), (0xfffde, 20), (0x3ffe4, 18), (0xfffdf, 20), (0xfffe0, 20), (0xfffe1, 20), (0x7ffde, 19), + (0xfffe2, 20), (0xfffe3, 20), (0xfffe4, 20), (0x7ffdf, 19), (0xfffe5, 20), (0x7ffe0, 19), (0x3ffe8, 18), (0x7ffe1, 19), + (0x3ffe0, 18), (0x3ffe9, 18), (0x1ffef, 17), (0x3ffe5, 18), (0x1ffec, 17), (0x1ffed, 17), (0x1ffee, 17), (0xfff4, 16), + (0xfff3, 16), (0xfff0, 16), (0x7ff7, 15), (0x7ff6, 15), (0x3ffa, 14), (0x1ffa, 13), (0x1ff9, 13), (0xffa, 12), + (0xff8, 12), (0x7f9, 11), (0x3fb, 10), (0x1fc, 9), (0x1fa, 9), (0xfb, 8), (0x7c, 7), (0x3c, 6), + (0x1c, 5), (0xc, 4), (0x5, 3), (0x1, 2), (0x0, 2), (0x4, 3), (0xd, 4), (0x1d, 5), + (0x3d, 6), (0xfa, 8), (0xfc, 8), (0x1fb, 9), (0x3fa, 10), (0x7f8, 11), (0x7fa, 11), (0x7fb, 11), + (0xff9, 12), (0xffb, 12), (0x1ff8, 13), (0x1ffb, 13), (0x3ff8, 14), (0x3ff9, 14), (0xfff1, 16), (0xfff2, 16), + (0x1ffea, 17), (0x1ffeb, 17), (0x3ffe1, 18), (0x3ffe2, 18), (0x3ffea, 18), (0x3ffe3, 18), (0x3ffe6, 18), (0x3ffe7, 18), + (0x3ffeb, 18), (0xfffe6, 20), (0x7ffe2, 19), (0xfffe7, 20), (0xfffe8, 20), (0xfffe9, 20), (0xfffea, 20), (0xfffeb, 20), + (0xfffec, 20), (0x7ffe3, 19), (0xfffed, 20), (0xfffee, 20), (0xfffef, 20), (0xffff0, 20), (0x7ffe4, 19), (0xffff1, 20), + (0x3ffec, 18), (0xffff2, 20), (0xffff3, 20), (0x7ffe5, 19), (0x7ffe6, 19), (0xffff4, 20), (0xffff5, 20), (0xffff6, 20), + (0xffff7, 20), (0xffff8, 20), (0xffff9, 20), (0xffffa, 20), (0xffffb, 20), (0xffffc, 20), (0xffffd, 20), (0xffffe, 20), + (0xfffff, 20), +]; + +/// `t_huffman_env_bal_1_5dB` (`(codeword, length)` by index): envelope balance, time direction, 1.5 dB steps (LAV 24). +#[rustfmt::skip] +pub(crate) static T_ENV_BAL_1_5DB: [(u32, u8); 49] = [ + (0xffe4, 16), (0xffe5, 16), (0xffe6, 16), (0xffe7, 16), (0xffe8, 16), (0xffe9, 16), (0xffea, 16), (0xffeb, 16), + (0xffec, 16), (0xffed, 16), (0xffee, 16), (0xffef, 16), (0xfff0, 16), (0xfff1, 16), (0xfff2, 16), (0xfff3, 16), + (0xfff4, 16), (0xffe2, 16), (0xffc, 12), (0x7fc, 11), (0x1fe, 9), (0x7e, 7), (0x1e, 5), (0x6, 3), + (0x0, 1), (0x2, 2), (0xe, 4), (0x3e, 6), (0xfe, 8), (0x7fd, 11), (0xffd, 12), (0x7ff0, 15), + (0xffe3, 16), (0xfff5, 16), (0xfff6, 16), (0xfff7, 16), (0xfff8, 16), (0xfff9, 16), (0xfffa, 16), (0x1fff6, 17), + (0x1fff7, 17), (0x1fff8, 17), (0x1fff9, 17), (0x1fffa, 17), (0x1fffb, 17), (0x1fffc, 17), (0x1fffd, 17), (0x1fffe, 17), + (0x1ffff, 17), +]; + +/// `f_huffman_env_bal_1_5dB` (`(codeword, length)` by index): envelope balance, frequency direction, 1.5 dB steps (LAV 24). +#[rustfmt::skip] +pub(crate) static F_ENV_BAL_1_5DB: [(u32, u8); 49] = [ + (0x3ffe2, 18), (0x3ffe3, 18), (0x3ffe4, 18), (0x3ffe5, 18), (0x3ffe6, 18), (0x3ffe7, 18), (0x3ffe8, 18), (0x3ffe9, 18), + (0x3ffea, 18), (0x3ffeb, 18), (0x3ffec, 18), (0x3ffed, 18), (0x3ffee, 18), (0x3ffef, 18), (0x3fff0, 18), (0xfff7, 16), + (0x1fff0, 17), (0x3ffc, 14), (0x7fe, 11), (0x7fc, 11), (0xfe, 8), (0x7e, 7), (0xe, 4), (0x2, 2), + (0x0, 1), (0x6, 3), (0x1e, 5), (0x3e, 6), (0x1fe, 9), (0x7fd, 11), (0xffe, 12), (0x7ffa, 15), + (0xfff6, 16), (0x3fff1, 18), (0x3fff2, 18), (0x3fff3, 18), (0x3fff4, 18), (0x3fff5, 18), (0x3fff6, 18), (0x3fff7, 18), + (0x3fff8, 18), (0x3fff9, 18), (0x3fffa, 18), (0x3fffb, 18), (0x3fffc, 18), (0x3fffd, 18), (0x3fffe, 18), (0x7fffe, 19), + (0x7ffff, 19), +]; + +/// `t_huffman_env_3_0dB` (`(codeword, length)` by index): envelope, time direction, 3 dB steps (LAV 31). +#[rustfmt::skip] +pub(crate) static T_ENV_3_0DB: [(u32, u8); 63] = [ + (0x3ffed, 18), (0x3ffee, 18), (0x7ffde, 19), (0x7ffdf, 19), (0x7ffe0, 19), (0x7ffe1, 19), (0x7ffe2, 19), (0x7ffe3, 19), + (0x7ffe4, 19), (0x7ffe5, 19), (0x7ffe6, 19), (0x7ffe7, 19), (0x7ffe8, 19), (0x7ffe9, 19), (0x7ffea, 19), (0x7ffeb, 19), + (0x7ffec, 19), (0x1fff4, 17), (0xfff7, 16), (0xfff9, 16), (0xfff8, 16), (0x3ffb, 14), (0x3ffa, 14), (0x3ff8, 14), + (0x1ffa, 13), (0xffc, 12), (0x7fc, 11), (0xfe, 8), (0x3e, 6), (0xe, 4), (0x2, 2), (0x0, 1), + (0x6, 3), (0x1e, 5), (0x7e, 7), (0x1fe, 9), (0x7fd, 11), (0x1ffb, 13), (0x3ff9, 14), (0x3ffc, 14), + (0x7ffa, 15), (0xfff6, 16), (0x1fff5, 17), (0x3ffec, 18), (0x7ffed, 19), (0x7ffee, 19), (0x7ffef, 19), (0x7fff0, 19), + (0x7fff1, 19), (0x7fff2, 19), (0x7fff3, 19), (0x7fff4, 19), (0x7fff5, 19), (0x7fff6, 19), (0x7fff7, 19), (0x7fff8, 19), + (0x7fff9, 19), (0x7fffa, 19), (0x7fffb, 19), (0x7fffc, 19), (0x7fffd, 19), (0x7fffe, 19), (0x7ffff, 19), +]; + +/// `f_huffman_env_3_0dB` (`(codeword, length)` by index): envelope and noise floor, frequency direction, 3 dB steps (LAV 31). +#[rustfmt::skip] +pub(crate) static F_ENV_3_0DB: [(u32, u8); 63] = [ + (0xffff0, 20), (0xffff1, 20), (0xffff2, 20), (0xffff3, 20), (0xffff4, 20), (0xffff5, 20), (0xffff6, 20), (0x3fff3, 18), + (0x7fff5, 19), (0x7ffee, 19), (0x7ffef, 19), (0x7fff6, 19), (0x3fff4, 18), (0x3fff2, 18), (0xffff7, 20), (0x7fff0, 19), + (0x1fff5, 17), (0x3fff0, 18), (0x1fff4, 17), (0xfff7, 16), (0xfff6, 16), (0x7ff8, 15), (0x3ffb, 14), (0xffd, 12), + (0x7fd, 11), (0x3fd, 10), (0x1fd, 9), (0xfd, 8), (0x3e, 6), (0xe, 4), (0x2, 2), (0x0, 1), + (0x6, 3), (0x1e, 5), (0xfc, 8), (0x1fc, 9), (0x3fc, 10), (0x7fc, 11), (0xffc, 12), (0x1ffc, 13), + (0x3ffa, 14), (0x7ff9, 15), (0x7ffa, 15), (0xfff8, 16), (0xfff9, 16), (0x1fff6, 17), (0x1fff7, 17), (0x3fff5, 18), + (0x3fff6, 18), (0x3fff1, 18), (0xffff8, 20), (0x7fff1, 19), (0x7fff2, 19), (0x7fff3, 19), (0xffff9, 20), (0x7fff7, 19), + (0x7fff4, 19), (0xffffa, 20), (0xffffb, 20), (0xffffc, 20), (0xffffd, 20), (0xffffe, 20), (0xfffff, 20), +]; + +/// `t_huffman_env_bal_3_0dB` (`(codeword, length)` by index): envelope balance, time direction, 3 dB steps (LAV 12). +#[rustfmt::skip] +pub(crate) static T_ENV_BAL_3_0DB: [(u32, u8); 25] = [ + (0x1ff2, 13), (0x1ff3, 13), (0x1ff4, 13), (0x1ff5, 13), (0x1ff6, 13), (0x1ff7, 13), (0x1ff8, 13), (0xff8, 12), + (0xfe, 8), (0x7e, 7), (0xe, 4), (0x6, 3), (0x0, 1), (0x2, 2), (0x1e, 5), (0x3e, 6), + (0x1fe, 9), (0x1ff9, 13), (0x1ffa, 13), (0x1ffb, 13), (0x1ffc, 13), (0x1ffd, 13), (0x1ffe, 13), (0x3ffe, 14), + (0x3fff, 14), +]; + +/// `f_huffman_env_bal_3_0dB` (`(codeword, length)` by index): envelope and noise floor balance, frequency direction, 3 dB steps (LAV 12). +#[rustfmt::skip] +pub(crate) static F_ENV_BAL_3_0DB: [(u32, u8); 25] = [ + (0x1ff7, 13), (0x1ff8, 13), (0x1ff9, 13), (0x1ffa, 13), (0x1ffb, 13), (0x3ff8, 14), (0x3ff9, 14), (0x7fc, 11), + (0xfe, 8), (0x7e, 7), (0xe, 4), (0x2, 2), (0x0, 1), (0x6, 3), (0x1e, 5), (0x3e, 6), + (0x1fe, 9), (0xffa, 12), (0x1ff6, 13), (0x3ffa, 14), (0x3ffb, 14), (0x3ffc, 14), (0x3ffd, 14), (0x3ffe, 14), + (0x3fff, 14), +]; + +/// `t_huffman_noise_3_0dB` (`(codeword, length)` by index): noise floor, time direction (LAV 31). +#[rustfmt::skip] +pub(crate) static T_NOISE_3_0DB: [(u32, u8); 63] = [ + (0x1fce, 13), (0x1fcf, 13), (0x1fd0, 13), (0x1fd1, 13), (0x1fd2, 13), (0x1fd3, 13), (0x1fd4, 13), (0x1fd5, 13), + (0x1fd6, 13), (0x1fd7, 13), (0x1fd8, 13), (0x1fd9, 13), (0x1fda, 13), (0x1fdb, 13), (0x1fdc, 13), (0x1fdd, 13), + (0x1fde, 13), (0x1fdf, 13), (0x1fe0, 13), (0x1fe1, 13), (0x1fe2, 13), (0x1fe3, 13), (0x1fe4, 13), (0x1fe5, 13), + (0x1fe6, 13), (0x1fe7, 13), (0x7f2, 11), (0xfd, 8), (0x3e, 6), (0xe, 4), (0x6, 3), (0x0, 1), + (0x2, 2), (0x1e, 5), (0xfc, 8), (0x3f8, 10), (0x1fcc, 13), (0x1fe8, 13), (0x1fe9, 13), (0x1fea, 13), + (0x1feb, 13), (0x1fec, 13), (0x1fcd, 13), (0x1fed, 13), (0x1fee, 13), (0x1fef, 13), (0x1ff0, 13), (0x1ff1, 13), + (0x1ff2, 13), (0x1ff3, 13), (0x1ff4, 13), (0x1ff5, 13), (0x1ff6, 13), (0x1ff7, 13), (0x1ff8, 13), (0x1ff9, 13), + (0x1ffa, 13), (0x1ffb, 13), (0x1ffc, 13), (0x1ffd, 13), (0x1ffe, 13), (0x3ffe, 14), (0x3fff, 14), +]; + +/// `t_huffman_noise_bal_3_0dB` (`(codeword, length)` by index): noise floor balance, time direction (LAV 12). +#[rustfmt::skip] +pub(crate) static T_NOISE_BAL_3_0DB: [(u32, u8); 25] = [ + (0xec, 8), (0xed, 8), (0xee, 8), (0xef, 8), (0xf0, 8), (0xf1, 8), (0xf2, 8), (0xf3, 8), + (0xf4, 8), (0xf5, 8), (0x1c, 5), (0x2, 2), (0x0, 1), (0x6, 3), (0x3a, 6), (0xf6, 8), + (0xf7, 8), (0xf8, 8), (0xf9, 8), (0xfa, 8), (0xfb, 8), (0xfc, 8), (0xfd, 8), (0xfe, 8), + (0xff, 8), +]; diff --git a/crates/audio/src/ac3/mod.rs b/crates/audio/src/ac3/mod.rs index a8e8920d2..6bb19df8d 100644 --- a/crates/audio/src/ac3/mod.rs +++ b/crates/audio/src/ac3/mod.rs @@ -61,7 +61,7 @@ pub enum Error { pub type Result = std::result::Result; /// Data rate in kb/s by frmsizecod / 2 (Table 5.18). -const BITRATES: [u32; 19] = [32, 40, 48, 56, 64, 80, 96, 112, 128, 160, 192, 224, 256, 320, 384, 448, 512, 576, 640]; +pub(crate) const BITRATES: [u32; 19] = [32, 40, 48, 56, 64, 80, 96, 112, 128, 160, 192, 224, 256, 320, 384, 448, 512, 576, 640]; /// Full-bandwidth channels per acmod (Table 5.8). const NFCHANS: [usize; 8] = [2, 1, 2, 3, 3, 4, 4, 5]; diff --git a/crates/audio/src/essence_pcm.rs b/crates/audio/src/essence_pcm.rs new file mode 100644 index 000000000..dfb119b8b --- /dev/null +++ b/crates/audio/src/essence_pcm.rs @@ -0,0 +1,189 @@ +//! Compatibility WAVE and AIFF wrappers for embedded PCM essence in interchange files. +//! PCM inside the model is interleaved little-endian signed integers (8-bit unsigned for WAVE). + +pub use crate::{pcm_aiff_header as aiff_header, pcm_wav_file as wav_file, pcm_wav_header as wav_header}; + +/// PCM, channels, sample rate and bits of a WAVE file (integer PCM only). +pub fn parse_wav(b: &[u8]) -> Option<(Vec, u16, u32, u16)> { + if b.len() < 12 || &b[..4] != b"RIFF" || &b[8..12] != b"WAVE" { + return None; + } + let mut at = 12; + let mut fmt = None; + while at + 8 <= b.len() { + let id = &b[at..at + 4]; + let len = u32::from_le_bytes(b[at + 4..at + 8].try_into().ok()?) as usize; + let body = &b[at + 8..(at + 8).saturating_add(len).min(b.len())]; + if id == b"fmt " && body.len() >= 16 { + let tag = u16::from_le_bytes([body[0], body[1]]); + let ch = u16::from_le_bytes([body[2], body[3]]); + let sr = u32::from_le_bytes(body[4..8].try_into().ok()?); + let bits = u16::from_le_bytes([body[14], body[15]]); + if tag != 1 && tag != 0xFFFE { + return None; + } + if tag == 0xFFFE { + // The first two bytes of the sub-format GUID carry the real format code. + let sub = body.get(24..26)?; + if u16::from_le_bytes([sub[0], sub[1]]) != 1 { + return None; + } + } + fmt = Some((ch, sr, bits)); + } else if id == b"data" { + let (ch, sr, bits) = fmt?; + return Some((body.to_vec(), ch, sr, bits)); + } + at = at.checked_add(8usize.checked_add(len)?.checked_add(len & 1)?)?; + } + None +} + +/// A complete AIFF file (big-endian PCM) from little-endian PCM. +pub fn aiff_file(pcm: &[u8], channels: u16, sample_rate: u32, bits: u16) -> Vec { + let bps = bits.div_ceil(8) as usize; + let frames = pcm.len() / (bps * channels.max(1) as usize).max(1); + let mut v = aiff_header(frames as u32, channels, sample_rate, bits); + for s in pcm.chunks_exact(bps.max(1)) { + if bps == 1 { + v.push(s[0] ^ 0x80); // WAVE 8-bit is unsigned, AIFF signed + } else { + v.extend(s.iter().rev()); + } + } + if (v.len() - 8) & 1 == 1 { + v.push(0); + } + v +} + +/// PCM (converted to little-endian), channels, sample rate and bits of an AIFF / AIFF-C (`NONE` +/// / `twos` / `sowt`) file. +pub fn parse_aiff(b: &[u8]) -> Option<(Vec, u16, u32, u16)> { + if b.len() < 12 || &b[..4] != b"FORM" || !(&b[8..12] == b"AIFF" || &b[8..12] == b"AIFC") { + return None; + } + let aifc = &b[8..12] == b"AIFC"; + let mut at = 12; + let mut comm = None; + while at + 8 <= b.len() { + let id = &b[at..at + 4]; + let len = u32::from_be_bytes(b[at + 4..at + 8].try_into().ok()?) as usize; + let body = &b[at + 8..(at + 8).saturating_add(len).min(b.len())]; + if id == b"COMM" && body.len() >= 18 { + let ch = u16::from_be_bytes([body[0], body[1]]); + let bits = u16::from_be_bytes([body[6], body[7]]); + let sr = from_extended(body[8..18].try_into().ok()?).round() as u32; + // AIFF has no compression code (trailing COMM bytes mean nothing); AIFF-C must name one + // of the integer PCM codes, or the samples are not integer PCM at all + let little = if aifc { + match body.get(18..22)? { + b"NONE" | b"twos" => false, + b"sowt" => true, + _ => return None, + } + } else { + false + }; + comm = Some((ch, sr, bits, little)); + } else if id == b"SSND" && body.len() >= 8 { + let (ch, sr, bits, little) = comm?; + let off = u32::from_be_bytes(body[..4].try_into().ok()?) as usize; + let data = body.get(8usize.checked_add(off)?..)?; + let bps = bits.div_ceil(8) as usize; + let mut pcm = Vec::with_capacity(data.len()); + for s in data.chunks_exact(bps.max(1)) { + if bps == 1 { + pcm.push(s[0] ^ 0x80); + } else if little { + pcm.extend_from_slice(s); + } else { + pcm.extend(s.iter().rev()); + } + } + return Some((pcm, ch, sr, bits)); + } + at = at.checked_add(8usize.checked_add(len)?.checked_add(len & 1)?)?; + } + None +} + +fn from_extended(b: [u8; 10]) -> f64 { + let e = (u16::from_be_bytes([b[0], b[1]]) & 0x7FFF) as i32; + let m = u64::from_be_bytes(b[2..10].try_into().unwrap_or([0; 8])); + if e == 0 && m == 0 { + return 0.0; + } + m as f64 / (1u64 << 63) as f64 * 2f64.powi(e - 16383) +} + +#[cfg(test)] +mod tests { + /// A mutated EDL whose clip ran for days produced an OMF descriptor for more than 4 GiB of + /// PCM: the RIFF size field overflowed ("attempt to add with overflow"). + #[test] + fn huge_wav_header_saturates() { + let h = super::wav_header(u32::MAX, 2, u32::MAX, 24); + assert_eq!(&h[4..8], &u32::MAX.to_le_bytes()); + } + + use super::*; + + #[test] + fn accepts_truncated_essence_summaries() { + let mut wav = wav_file(&[1, 2, 3, 4], 1, 48_000, 16); + wav.truncate(46); + assert_eq!(parse_wav(&wav), Some((vec![1, 2], 1, 48_000, 16))); + let mut aiff = aiff_file(&[1, 2, 3, 4], 1, 48_000, 16); + aiff.truncate(56); + assert_eq!(parse_aiff(&aiff), Some((vec![1, 2], 1, 48_000, 16))); + } + + #[test] + fn wav_and_aiff_round_trip() { + let pcm: Vec = (0..48u8).collect(); // 8 stereo 24-bit frames + let w = wav_file(&pcm, 2, 48_000, 24); + assert_eq!(parse_wav(&w), Some((pcm.clone(), 2, 48_000, 24))); + let a = aiff_file(&pcm, 2, 44_100, 24); + assert_eq!(parse_aiff(&a), Some((pcm.clone(), 2, 44_100, 24))); + let p16: Vec = (0..32u8).collect(); + assert_eq!(parse_aiff(&aiff_file(&p16, 1, 96_000, 16)), Some((p16, 1, 96_000, 16))); + assert!(parse_wav(b"RIFF").is_none()); + assert!(parse_aiff(&a[..30]).is_none()); + for cut in 0..w.len() { + let _ = parse_wav(&w[..cut]); + let _ = parse_aiff(&a[..cut.min(a.len())]); + } + } + + #[test] + fn aifc_needs_an_integer_pcm_compression_code() { + let file = |kind: &[u8; 4], code: &[u8]| { + let mut comm = vec![0, 1, 0, 0, 0, 1, 0, 16]; + comm.extend_from_slice(&[0x40, 0x0E, 0xBB, 0x80, 0, 0, 0, 0, 0, 0]); + comm.extend_from_slice(code); + let mut body = kind.to_vec(); + body.extend_from_slice(b"COMM"); + body.extend_from_slice(&(comm.len() as u32).to_be_bytes()); + body.extend_from_slice(&comm); + if comm.len() & 1 == 1 { + body.push(0); + } + body.extend_from_slice(b"SSND"); + body.extend_from_slice(&10u32.to_be_bytes()); + body.extend_from_slice(&[0; 8]); + body.extend_from_slice(&[0x12, 0x34]); + let mut v = b"FORM".to_vec(); + v.extend_from_slice(&(body.len() as u32).to_be_bytes()); + v.extend_from_slice(&body); + v + }; + assert_eq!(parse_aiff(&file(b"AIFC", b"NONE")).map(|p| p.0), Some(vec![0x34, 0x12])); + assert_eq!(parse_aiff(&file(b"AIFC", b"twos")).map(|p| p.0), Some(vec![0x34, 0x12])); + assert_eq!(parse_aiff(&file(b"AIFC", b"sowt")).map(|p| p.0), Some(vec![0x12, 0x34])); + assert!(parse_aiff(&file(b"AIFC", b"fl32")).is_none(), "float samples are not integer PCM"); + assert!(parse_aiff(&file(b"AIFC", b"")).is_none(), "AIFF-C without a compression code"); + // plain AIFF ignores trailing COMM bytes + assert_eq!(parse_aiff(&file(b"AIFF", b"sowt")).map(|p| p.0), Some(vec![0x34, 0x12])); + } +} diff --git a/crates/audio/src/flac_decode.rs b/crates/audio/src/flac_decode.rs new file mode 100644 index 000000000..bb56e65c5 --- /dev/null +++ b/crates/audio/src/flac_decode.rs @@ -0,0 +1,603 @@ +//! A native FLAC decoder (RFC 9639): [`decode`] for whole `.flac` files and [`Decoder`] for single +//! frames (the packets of FLAC in MP4, Matroska or Ogg). +//! +//! Every frame is checked against its CRC-8 and CRC-16. Subframe types constant, verbatim, fixed +//! and LPC; Rice partitions with 4- and 5-bit parameters and escaped partitions; wasted bits; +//! left/side, side/right and mid/side stereo; fixed and variable block sizes; 4 to 32 bits per +//! sample. The STREAMINFO MD5 is not verified. +//! +//! Samples come back as `i32`, one `Vec` per channel, in the FLAC channel order. + +/// What went wrong decoding FLAC. +#[non_exhaustive] +#[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)] +pub enum DecodeError { + /// The data ends early. + #[error("truncated FLAC data")] + Truncated, + /// The data is not valid FLAC. + #[error("invalid FLAC data: {0}")] + Invalid(&'static str), + /// A frame or header failed its CRC. + #[error("FLAC CRC mismatch in {0}")] + Crc(&'static str), + /// Valid FLAC that is not decoded. + #[error("unsupported FLAC feature: {0}")] + Unsupported(&'static str), + /// The stream is longer than the sample limit. + #[error("FLAC stream exceeds the sample limit")] + TooLarge, +} + +/// `Result` of this module. +pub type Result = std::result::Result; + +/// The STREAMINFO metadata block (RFC 9639 section 8.2). +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub struct StreamInfo { + /// Smallest block size in samples. + pub min_block_size: u16, + /// Largest block size in samples. + pub max_block_size: u16, + /// Smallest frame in bytes, 0 if unknown. + pub min_frame_size: u32, + /// Largest frame in bytes, 0 if unknown. + pub max_frame_size: u32, + /// Sample rate in Hz. + pub sample_rate: u32, + /// Channel count, 1 to 8. + pub channels: u8, + /// Bits per sample, 4 to 32. + pub bits_per_sample: u8, + /// Samples per channel in the stream, 0 if unknown. + pub total_samples: u64, + /// MD5 of the unencoded audio, all zero if not set. + pub md5: [u8; 16], +} + +impl StreamInfo { + /// Parse the 34 bytes of a STREAMINFO block body. + pub fn parse(b: &[u8]) -> Result { + let b = b.get(..34).ok_or(DecodeError::Truncated)?; + let packed = u64::from_be_bytes([b[10], b[11], b[12], b[13], b[14], b[15], b[16], b[17]]); + let info = StreamInfo { + min_block_size: u16::from_be_bytes([b[0], b[1]]), + max_block_size: u16::from_be_bytes([b[2], b[3]]), + min_frame_size: u32::from_be_bytes([0, b[4], b[5], b[6]]), + max_frame_size: u32::from_be_bytes([0, b[7], b[8], b[9]]), + sample_rate: (packed >> 44) as u32, + channels: ((packed >> 41) & 7) as u8 + 1, + bits_per_sample: ((packed >> 36) & 31) as u8 + 1, + total_samples: packed & 0xF_FFFF_FFFF, + md5: b[18..34].try_into().map_err(|_| DecodeError::Truncated)?, + }; + if info.bits_per_sample < 4 { + return Err(DecodeError::Invalid("bits per sample below 4")); + } + if info.min_block_size != 0 + && (info.min_block_size < 16 && info.min_block_size != info.max_block_size || info.max_block_size < info.min_block_size) + { + return Err(DecodeError::Invalid("inconsistent block sizes")); + } + Ok(info) + } +} + +/// One decoded frame. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct Frame { + /// Sample rate in Hz. + pub sample_rate: u32, + /// Bits per sample. + pub bits_per_sample: u8, + /// `true` when `number` is the first sample's index (variable block size), `false` when it is + /// the frame's index (fixed block size). + pub variable_block_size: bool, + /// The coded number of the frame header. + pub number: u64, + /// Samples per channel in the frame. + pub block_size: usize, + /// One `Vec` per channel, after undoing any stereo decorrelation. + pub channels: Vec>, + /// Bytes the frame took, including the header and CRC footer. + pub size: usize, +} + +// ---- checksums ---- + +fn crc8(data: &[u8]) -> u8 { + let mut crc = 0u8; + for &b in data { + crc ^= b; + for _ in 0..8 { + crc = if crc & 0x80 != 0 { crc << 1 ^ 0x07 } else { crc << 1 }; + } + } + crc +} + +fn crc16(data: &[u8]) -> u16 { + let mut crc = 0u16; + for &b in data { + crc ^= u16::from(b) << 8; + for _ in 0..8 { + crc = if crc & 0x8000 != 0 { crc << 1 ^ 0x8005 } else { crc << 1 }; + } + } + crc +} + +// ---- bits ---- + +struct Bits<'a> { + data: &'a [u8], + /// Position in bits. + pos: usize, +} + +impl<'a> Bits<'a> { + fn new(data: &'a [u8]) -> Self { + Bits { data, pos: 0 } + } + + fn left(&self) -> usize { + self.data.len() * 8 - self.pos + } + + fn bit(&mut self) -> Result { + let byte = *self.data.get(self.pos / 8).ok_or(DecodeError::Truncated)?; + let v = (byte >> (7 - self.pos % 8)) & 1; + self.pos += 1; + Ok(u32::from(v)) + } + + /// `n` bits (0 to 32) as an unsigned number. + fn bits(&mut self, n: u32) -> Result { + if n as usize > self.left() { + return Err(DecodeError::Truncated); + } + let mut v = 0u64; + let mut n = n as usize; + while n > 0 { + let byte = u64::from(self.data[self.pos / 8]); + let avail = 8 - self.pos % 8; + let take = avail.min(n); + v = v << take | (byte >> (avail - take)) & ((1 << take) - 1); + self.pos += take; + n -= take; + } + Ok(v) + } + + /// `n` bits (0 to 33) as a two's complement number. + fn signed(&mut self, n: u32) -> Result { + if n == 0 { + return Ok(0); + } + let v = self.bits(n)? as i64; + Ok(if v >> (n - 1) & 1 == 1 { v - (1 << n) } else { v }) + } + + /// Zero bits up to the next one bit (which is consumed). + fn unary(&mut self) -> Result { + let mut n = 0u64; + loop { + // Skip whole zero bytes when aligned. + if self.pos.is_multiple_of(8) { + while self.data.get(self.pos / 8) == Some(&0) { + self.pos += 8; + n += 8; + } + } + if self.bit()? == 1 { + return Ok(n); + } + n += 1; + } + } + + fn align(&mut self) { + self.pos = self.pos.div_ceil(8) * 8; + } +} + +// ---- frame ---- + +const BLOCK_SIZES: [usize; 16] = [0, 192, 576, 1152, 2304, 4608, 0, 0, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768]; + +/// A frame decoder. `info` supplies what a frame header leaves to the STREAMINFO. +#[derive(Debug, Clone, Copy, Default)] +pub struct Decoder { + info: Option, +} + +impl Decoder { + /// A decoder with the stream's STREAMINFO (needed for frames whose header defers the sample + /// rate or bit depth to it). + pub fn new(info: StreamInfo) -> Decoder { + Decoder { info: Some(info) } + } + + /// A decoder without STREAMINFO: frames that defer to it are an error. + pub fn without_streaminfo() -> Decoder { + Decoder { info: None } + } + + /// Decode the frame at the start of `data`. + pub fn decode_frame(&self, data: &[u8]) -> Result { + // StreamInfo has public fields: container callers can construct it without + // parse(), so validate before it can control shifts and sample reads. + if self.info.is_some_and(|i| !(4..=32).contains(&i.bits_per_sample) || !(1..=8).contains(&i.channels) || i.sample_rate == 0) { + return Err(DecodeError::Invalid("invalid decoder STREAMINFO")); + } + let mut r = Bits::new(data); + if r.bits(15)? != 0x7FFC { + return Err(DecodeError::Invalid("missing frame sync code")); + } + let variable = r.bit()? == 1; + let (bs_code, sr_code) = (r.bits(4)? as usize, r.bits(4)? as usize); + let (ch_code, bps_code) = (r.bits(4)? as usize, r.bits(3)? as usize); + if r.bit()? != 0 { + return Err(DecodeError::Invalid("reserved bit set")); + } + let number = coded_number(&mut r)?; + let block_size = match bs_code { + 0 => return Err(DecodeError::Invalid("reserved block size")), + 6 => r.bits(8)? as usize + 1, + 7 => r.bits(16)? as usize + 1, + c => BLOCK_SIZES[c], + }; + let sample_rate = match sr_code { + 0 => self.info.map(|i| i.sample_rate).ok_or(DecodeError::Invalid("sample rate only in STREAMINFO"))?, + 1 => 88_200, + 2 => 176_400, + 3 => 192_000, + 4 => 8_000, + 5 => 16_000, + 6 => 22_050, + 7 => 24_000, + 8 => 32_000, + 9 => 44_100, + 10 => 48_000, + 11 => 96_000, + 12 => r.bits(8)? as u32 * 1000, + 13 => r.bits(16)? as u32, + 14 => r.bits(16)? as u32 * 10, + _ => return Err(DecodeError::Invalid("forbidden sample rate code")), + }; + let header_len = r.pos.div_ceil(8); + let stored_crc = r.bits(8)? as u8; + if crc8(&data[..header_len]) != stored_crc { + return Err(DecodeError::Crc("a frame header")); + } + let bits = match bps_code { + 0 => self.info.map(|i| u32::from(i.bits_per_sample)).ok_or(DecodeError::Invalid("bit depth only in STREAMINFO"))?, + 1 => 8, + 2 => 12, + 4 => 16, + 5 => 20, + 6 => 24, + 7 => 32, + _ => return Err(DecodeError::Invalid("reserved bit depth")), + }; + let (channels, mode) = match ch_code { + 0..=7 => (ch_code + 1, Stereo::Independent), + 8 => (2, Stereo::LeftSide), + 9 => (2, Stereo::SideRight), + 10 => (2, Stereo::MidSide), + _ => return Err(DecodeError::Invalid("reserved channel assignment")), + }; + let mut out: Vec> = Vec::with_capacity(channels); + for c in 0..channels { + let extra = match (mode, c) { + (Stereo::LeftSide, 1) | (Stereo::SideRight, 0) | (Stereo::MidSide, 1) => 1, + _ => 0, + }; + out.push(subframe(&mut r, block_size, bits + extra)?); + } + r.align(); + let size = r.pos / 8; + let crc = r.bits(16)? as u16; + if crc16(&data[..size]) != crc { + return Err(DecodeError::Crc("a frame")); + } + let size = size + 2; + if mode != Stereo::Independent { + let (first, rest) = out.split_at_mut(1); + let (a, b) = (&mut first[0], &mut rest[0]); + for (x, y) in a.iter_mut().zip(b.iter_mut()) { + match mode { + Stereo::LeftSide => *y = x.wrapping_sub(*y), + Stereo::SideRight => *x = x.wrapping_add(*y), + _ => { + let m = x.wrapping_shl(1) | (*y & 1); + let side = *y; + *x = m.wrapping_add(side) >> 1; + *y = m.wrapping_sub(side) >> 1; + } + } + } + } + let mut chans = Vec::with_capacity(channels); + for c in out { + let v: Result> = c.into_iter().map(|s| checked_sample(s, bits)).collect(); + chans.push(v?); + } + Ok(Frame { sample_rate, bits_per_sample: bits as u8, variable_block_size: variable, number, block_size, channels: chans, size }) + } +} + +#[derive(Clone, Copy, PartialEq, Eq)] +enum Stereo { + Independent, + LeftSide, + SideRight, + MidSide, +} + +/// The UTF-8-like coded number of a frame header (up to 36 bits, 7 bytes). +fn coded_number(r: &mut Bits) -> Result { + let first = r.bits(8)? as u8; + let extra = match first { + 0x00..=0x7F => return Ok(u64::from(first)), + 0xC0..=0xDF => 1, + 0xE0..=0xEF => 2, + 0xF0..=0xF7 => 3, + 0xF8..=0xFB => 4, + 0xFC..=0xFD => 5, + 0xFE => 6, + _ => return Err(DecodeError::Invalid("bad coded number")), + }; + let mut v = if extra == 6 { 0 } else { u64::from(first & (0x3F >> extra)) }; + for _ in 0..extra { + let b = r.bits(8)?; + if b & 0xC0 != 0x80 { + return Err(DecodeError::Invalid("bad coded number")); + } + v = v << 6 | (b & 0x3F); + } + Ok(v) +} + +/// One subframe: `bits` is the frame bit depth (plus 1 for a side channel). +fn subframe(r: &mut Bits, n: usize, bits: u32) -> Result> { + if r.bit()? != 0 { + return Err(DecodeError::Invalid("subframe header padding bit set")); + } + let kind = r.bits(6)? as u32; + let wasted = if r.bit()? == 1 { r.unary()? as u32 + 1 } else { 0 }; + if wasted >= bits { + return Err(DecodeError::Invalid("wasted bits leave no sample bits")); + } + let bps = bits - wasted; + let mut s = vec![0i64; n]; + match kind { + 0 => { + let v = r.signed(bps)?; + s.fill(v); + } + 1 => { + for v in s.iter_mut() { + *v = r.signed(bps)?; + } + } + 8..=12 => { + let order = (kind - 8) as usize; + warmup(r, &mut s, order, bps)?; + residual(r, &mut s, order)?; + for i in order..n { + let at = |k: usize| s[i - k]; + // Wrapping: hostile residuals can overflow; such samples fail the range check later. + let p = match order { + 0 => 0, + 1 => at(1), + 2 => at(1).wrapping_mul(2).wrapping_sub(at(2)), + 3 => at(1).wrapping_mul(3).wrapping_sub(at(2).wrapping_mul(3)).wrapping_add(at(3)), + _ => at(1).wrapping_mul(4).wrapping_sub(at(2).wrapping_mul(6)).wrapping_add(at(3).wrapping_mul(4)).wrapping_sub(at(4)), + }; + s[i] = s[i].wrapping_add(p); + } + } + 32..=63 => { + let order = (kind - 31) as usize; + warmup(r, &mut s, order, bps)?; + let precision = r.bits(4)? as u32 + 1; + if precision == 16 { + return Err(DecodeError::Invalid("forbidden coefficient precision")); + } + let shift = r.signed(5)?; + if shift < 0 { + return Err(DecodeError::Invalid("negative prediction shift")); + } + let mut coefs = [0i64; 32]; + for c in coefs.iter_mut().take(order) { + *c = r.signed(precision)?; + } + residual(r, &mut s, order)?; + for i in order..n { + let mut sum = 0i64; + for (k, &c) in coefs.iter().take(order).enumerate() { + sum = sum.wrapping_add(c.wrapping_mul(s[i - 1 - k])); + } + s[i] = s[i].wrapping_add(sum >> shift); + } + } + _ => return Err(DecodeError::Invalid("reserved subframe type")), + } + if wasted > 0 { + for v in s.iter_mut() { + *v = v.wrapping_shl(wasted); + } + } + Ok(s) +} + +fn warmup(r: &mut Bits, s: &mut [i64], order: usize, bps: u32) -> Result<()> { + if order > s.len() { + return Err(DecodeError::Invalid("predictor order above the block size")); + } + for v in s.iter_mut().take(order) { + *v = r.signed(bps)?; + } + Ok(()) +} + +/// The coded residual (RFC 9639 section 9.2.7), into `s[order..]`. +fn residual(r: &mut Bits, s: &mut [i64], order: usize) -> Result<()> { + let method = r.bits(2)?; + let (param_bits, escape) = match method { + 0 => (4, 15), + 1 => (5, 31), + _ => return Err(DecodeError::Invalid("reserved residual coding method")), + }; + let part_order = r.bits(4)? as u32; + let n = s.len(); + let parts = 1usize << part_order; + if !n.is_multiple_of(parts) { + return Err(DecodeError::Invalid("partition order does not divide the block size")); + } + let per = n >> part_order; + if per < order { + return Err(DecodeError::Invalid("partition shorter than the predictor order")); + } + let mut i = order; + for p in 0..parts { + let count = if p == 0 { per - order } else { per }; + let k = r.bits(param_bits)? as u32; + if k == escape { + let width = r.bits(5)? as u32; + for v in s[i..i + count].iter_mut() { + *v = r.signed(width)?; + } + } else { + for v in s[i..i + count].iter_mut() { + let q = r.unary()?; + if q > (1 << 32) { + return Err(DecodeError::Invalid("Rice quotient too large")); + } + let low = r.bits(k)?; + let folded = q << k | low; + *v = if folded & 1 == 0 { (folded >> 1) as i64 } else { -((folded >> 1) as i64) - 1 }; + } + } + i += count; + } + Ok(()) +} + +// ---- files ---- + +/// A decoded file. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct Decoded { + /// The STREAMINFO block. + pub info: StreamInfo, + /// One `Vec` of samples per channel. + pub channels: Vec>, +} + +/// Most sample values (samples x channels) [`decode`] will produce. +pub const DEFAULT_MAX_VALUES: u64 = 1 << 28; + +/// Decode a whole `.flac` file (an ID3v2 tag in front is skipped). +pub fn decode(bytes: &[u8]) -> Result { + decode_with_limit(bytes, DEFAULT_MAX_VALUES) +} + +/// [`decode`] with a limit on the sample values (samples x channels) produced. +pub fn decode_with_limit(bytes: &[u8], max_values: u64) -> Result { + let mut pos = 0usize; + if bytes.starts_with(b"ID3") { + let size = bytes.get(6..10).ok_or(DecodeError::Truncated)?; + if size.iter().any(|&b| b & 0x80 != 0) { + return Err(DecodeError::Invalid("bad ID3v2 size")); + } + pos = 10 + size.iter().fold(0usize, |a, &b| a << 7 | usize::from(b)); + } + if bytes.get(pos..pos + 4) != Some(b"fLaC") { + return Err(DecodeError::Invalid("missing fLaC marker")); + } + pos += 4; + let mut info: Option = None; + loop { + let head = bytes.get(pos..pos + 4).ok_or(DecodeError::Truncated)?; + let last = head[0] & 0x80 != 0; + let kind = head[0] & 0x7F; + let len = usize::from(head[1]) << 16 | usize::from(head[2]) << 8 | usize::from(head[3]); + let body = bytes.get(pos + 4..pos + 4 + len).ok_or(DecodeError::Truncated)?; + match (kind, &info) { + (0, None) => info = Some(StreamInfo::parse(body)?), + (0, Some(_)) => return Err(DecodeError::Invalid("a second STREAMINFO block")), + (_, None) => return Err(DecodeError::Invalid("STREAMINFO is not the first block")), + (127, _) => return Err(DecodeError::Invalid("forbidden metadata block type")), + _ => {} + } + pos += 4 + len; + if last { + break; + } + } + let info = info.ok_or(DecodeError::Invalid("no STREAMINFO"))?; + let dec = Decoder::new(info); + let mut channels: Vec> = vec![Vec::new(); usize::from(info.channels)]; + let mut total = 0u64; + while pos + 2 <= bytes.len() && bytes[pos] == 0xFF && bytes[pos + 1] & 0xFE == 0xF8 { + let frame = dec.decode_frame(&bytes[pos..])?; + if frame.channels.len() != channels.len() { + return Err(DecodeError::Invalid("the channel count changes inside the stream")); + } + // One file has one sample rate and depth. Otherwise callers would play + // samples under incompatible STREAMINFO metadata. + if frame.sample_rate != info.sample_rate || frame.bits_per_sample != info.bits_per_sample { + return Err(DecodeError::Invalid("frame format differs from STREAMINFO")); + } + total += frame.block_size as u64; + if total.saturating_mul(channels.len() as u64) > max_values { + return Err(DecodeError::TooLarge); + } + for (dst, src) in channels.iter_mut().zip(frame.channels) { + dst.extend_from_slice(&src); + } + pos += frame.size; + } + // Unknown total_samples must not turn a corrupt/truncated next frame into + // a successfully decoded shorter recording. + if pos != bytes.len() { + return Err(DecodeError::Invalid("trailing data or missing frame sync")); + } + if info.total_samples != 0 && total != info.total_samples { + return Err(DecodeError::Truncated); + } + Ok(Decoded { info, channels }) +} + +// Validate the advertised precision, not just the storage type. Malformed +// predictors may produce an i32 that is still invalid for 4–24-bit audio. +fn checked_sample(sample: i64, bits: u32) -> Result { + if !(4..=32).contains(&bits) { + return Err(DecodeError::Invalid("invalid sample bit depth")); + } + let bound = 1i64 << (bits - 1); + if !(-bound..bound).contains(&sample) { + return Err(DecodeError::Invalid("sample outside the bit depth")); + } + i32::try_from(sample).map_err(|_| DecodeError::Invalid("sample outside the bit depth")) +} + +#[cfg(test)] +mod sample_bounds_tests { + use super::checked_sample; + + #[test] + fn samples_must_fit_declared_precision() { + for bits in 4..=32 { + let bound = 1i64 << (bits - 1); + assert_eq!(checked_sample(-bound, bits).map(i64::from), Ok(-bound)); + assert_eq!(checked_sample(bound - 1, bits).map(i64::from), Ok(bound - 1)); + assert!(checked_sample(-bound - 1, bits).is_err()); + assert!(checked_sample(bound, bits).is_err()); + } + assert!(checked_sample(0, 0).is_err()); + assert!(checked_sample(0, 255).is_err()); + } +} diff --git a/crates/audio/src/layout.rs b/crates/audio/src/layout.rs index f3cab1b7f..d0c5ae39e 100644 --- a/crates/audio/src/layout.rs +++ b/crates/audio/src/layout.rs @@ -283,6 +283,40 @@ pub(crate) fn flac_assumed_mask(channels: usize) -> u32 { } } +/// The surround speaker a layout may name in place of `speaker`: formats with one surround pair +/// do not tell side from back, so a pair of either kind feeds it. +fn surround_twin(speaker: Speaker) -> Option { + Some(match speaker { + Speaker::SideLeft => Speaker::BackLeft, + Speaker::SideRight => Speaker::BackRight, + Speaker::BackLeft => Speaker::SideLeft, + Speaker::BackRight => Speaker::SideRight, + _ => return None, + }) +} + +/// How audio with `layout` feeds a format whose layout for the channel count is `mask`: for each +/// channel the format takes, the channel of the audio that feeds it. `order` is the format's +/// channel order over `mask` (for each of its channels, the WAV-order channel of `mask`; `None` +/// when it takes WAV order). `None` when `layout` is not the format's layout: it is unordered, or +/// has a speaker the format has not. +pub(crate) fn channel_map(mask: u32, order: Option<&[usize]>, layout: &ChannelLayout) -> Option> { + let speakers = ChannelLayout::from_mask(mask).speakers(); + if speakers.len() != usize::from(layout.channels()) { + return None; + } + let mut map = Vec::with_capacity(speakers.len()); + for j in 0..speakers.len() { + let speaker = *speakers.get(order.and_then(|o| o.get(j).copied()).unwrap_or(j))?; + let at = layout.index_of(speaker).or_else(|| surround_twin(speaker).and_then(|t| layout.index_of(t)))?; + if map.contains(&at) { + return None; + } + map.push(at); + } + Some(map) +} + #[cfg(test)] mod tests { use super::*; diff --git a/crates/audio/src/lib.rs b/crates/audio/src/lib.rs index 3ed684245..c7382a973 100644 --- a/crates/audio/src/lib.rs +++ b/crates/audio/src/lib.rs @@ -13,20 +13,24 @@ //! demuxers and [`TrackDecoder`]; everything else (FLAC, Vorbis, CAF, ADPCM, A-law, µ-law) through //! `symphonia`, which is to be replaced by native decoders. Feature `decode-native` is the //! same without symphonia. +//! - [`encode_ogg_vorbis`] and [`vorbis::encoder`] (feature `vorbis-encode`): Ogg Vorbis files +//! and Vorbis packets from a pure-Rust encoder, 1 to 8 channels, quality -1 to 10, with tags. //! - [`TrackDecoder`] (any decoder feature): one track's packets from any demuxer to planar //! `f32` in WAV channel order. //! - [`encode`], [`FileEncoder`]: WAV/BWF/RF64 and AIFF/AIFF-C (feature `pcm-encode`, which also -//! gives [`wav_float_header`]) and FLAC (feature `flac-encode`), whole-buffer or incremental into -//! any `Write + Seek`. Feature `lossless-encode` turns on both; feature `encode` turns on -//! `lossless-encode` and every codec's encoder. +//! gives [`wav_float_header`]), FLAC (feature `flac-encode`), AAC-LC and ALAC in M4A +//! (`aac-encode`, `alac-encode`) and MP3, AC-3 and E-AC-3 streams (`mp3-encode`, `ac3-encode`, +//! `eac3-encode`), whole-buffer or incremental into any `Write + Seek`. Feature +//! `lossless-encode` turns on WAV, AIFF and FLAC; feature `encode` turns on every format and encoder. //! - [`wav_pcm`], [`aiff_pcm`] (feature `decode-native`) and [`pcm_wav_file`], [`pcm_wav_header`], //! [`pcm_aiff_file`], [`pcm_aiff_header`] (feature `pcm-encode`): the integer PCM of a WAV or AIFF //! file bit for bit, and minimal files around such PCM, for essence embedded in other containers. //! - [`peaks`] (feature `peaks`): min/max/RMS pyramid for waveform drawing. //! - Codec modules for packet-level use, each with its own features: [`aac`] (`aac-decode`, -//! `aac-encode`), [`alac`] (`alac-decode`, `alac-encode`), [`opus`] (`opus-decode`), [`ac3`] -//! (`ac3-decode`), [`mpa`] (`mpa-decode`), [`mp3`] (`mp3-decode`, `mp3-encode`), -//! [`flac`] (`flac-encode`: the packet encoder). +//! `aac-encode`), [`alac`] (`alac-decode`, `alac-encode`), [`opus`] (`opus-decode`, and +//! [`opus::encoder`] with `opus-encode`), [`ac3`] (`ac3-decode`), [`mpa`] (`mpa-decode`), [`mp3`] +//! (`mp3-decode`, `mp3-encode`), [`flac`] (`flac-decode`: the frame and file decoder; +//! `flac-encode`: the packet encoder), [`qoa`] (`qoa-decode`, `qoa-encode`: whole QOA files). //! //! The shared types below build with no features. @@ -42,21 +46,48 @@ mod layout; pub mod mp3; #[cfg(feature = "mpa-decode")] pub mod mpa; -#[cfg(feature = "opus-decode")] +#[cfg(any(feature = "opus-decode", feature = "opus-encode"))] pub mod opus; #[cfg(any(feature = "aac-decode", feature = "alac-decode", feature = "opus-decode", feature = "ac3-decode", feature = "mpa-decode"))] mod pcm; #[cfg(feature = "peaks")] pub mod peaks; +#[cfg(any(feature = "qoa-decode", feature = "qoa-encode"))] +pub mod qoa; #[cfg(feature = "decode-native")] mod read; #[cfg(any(feature = "aac-decode", feature = "alac-decode", feature = "opus-decode", feature = "ac3-decode", feature = "mpa-decode"))] mod track; +#[cfg(feature = "vorbis-encode")] +pub mod vorbis; #[cfg(any(feature = "decode-native", feature = "pcm-encode"))] mod wav; -#[cfg(any(feature = "pcm-encode", feature = "flac-encode"))] +/// Bounded WAV/RF64 cue, adtl and XMP marker metadata, without audio decoding. +pub mod wav_markers; +#[cfg(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode", + feature = "vorbis-encode" +))] mod write; +/// PCM WAVE and AIFF wrappers for interchange essence. +#[cfg(feature = "pcm-encode")] +pub mod essence_pcm; +/// Streaming WAVE and AIFF headers and sample packing. +pub mod pcm_export; +/// Interleaved PCM packet decoding. +pub mod pcm_interleaved; +/// Range-backed PCM WAVE access and a 16-bit PCM writer. +pub mod wav_range; +#[cfg(feature = "vorbis-encode")] +pub use write::{OGG_BITRATES, OggVorbisOptions, encode_ogg, encode_ogg_vorbis, quality_for_bitrate}; + /// The FLAC packet encoder (RFC 9639), for a caller that muxes FLAC itself (into MP4, Matroska or /// its own `.flac` writer): planar integer samples in, frames out, and the STREAMINFO to patch in /// at the end. [`FileEncoder`] and [`encode`] write whole `.flac` files from `f32` instead. @@ -76,10 +107,15 @@ mod write; /// assert!(file.len() < 48_000 * 2 * 2 / 2); // well under half of 16-bit PCM /// # Ok::<(), craft_audio::flac::Error>(()) /// ``` -#[cfg(feature = "flac-encode")] +#[cfg(any(feature = "flac-encode", feature = "flac-decode"))] pub mod flac { + #[cfg(feature = "flac-decode")] + pub use crate::flac_decode::{DEFAULT_MAX_VALUES, DecodeError, Decoded, Decoder, Frame, StreamInfo, decode, decode_with_limit}; + #[cfg(feature = "flac-encode")] pub use crate::write::flac::encoder::{Encoder, EncoderConfig, Error, STREAMINFO_OFFSET}; } +#[cfg(feature = "flac-decode")] +mod flac_decode; pub use buffer::AudioBuffer; /// The craft-media this crate is built with, for its containers: using it rather than a separate @@ -98,10 +134,29 @@ pub use read::{ pub use read::{RawPcm, aiff_pcm, wav_pcm}; #[cfg(any(feature = "aac-decode", feature = "alac-decode", feature = "opus-decode", feature = "ac3-decode", feature = "mpa-decode"))] pub use track::{LOOKAHEAD_PACKETS, TrackDecoder}; +#[cfg(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode" +))] +pub use write::FileEncoder; #[cfg(feature = "pcm-encode")] pub use write::wav_float_header; -#[cfg(any(feature = "pcm-encode", feature = "flac-encode"))] -pub use write::{ENCODE_EXTENSIONS, FileEncoder, encode, encode_format_for}; +#[cfg(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode", + feature = "vorbis-encode" +))] +pub use write::{ENCODE_EXTENSIONS, encode, encode_format_for}; #[cfg(feature = "pcm-encode")] pub use write::{pcm_aiff_file, pcm_aiff_header, pcm_wav_file, pcm_wav_header}; @@ -207,8 +262,16 @@ impl From for AudioError { } } +#[cfg(feature = "opus-encode")] +impl From for AudioError { + fn from(e: opus::encoder::Error) -> Self { + AudioError::Encode(e.to_string()) + } +} + /// The file formats [`encode`] writes. Each needs its feature: WAV and AIFF `pcm-encode`, FLAC -/// `flac-encode`; asking for one that is not built returns [`AudioError::Unsupported`]. +/// `flac-encode`, AAC `aac-encode`, ALAC `alac-encode`, MP3 `mp3-encode`, AC-3 `ac3-encode` and +/// E-AC-3 `eac3-encode`; asking for one that is not built returns [`AudioError::Unsupported`]. #[non_exhaustive] #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)] #[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))] @@ -219,20 +282,52 @@ pub enum EncodeFormat { Aiff, /// FLAC (integer depths). Flac, + /// AAC-LC in an M4A (MP4) file, 1–8 channels at one of the 13 AAC sample rates. The edit list + /// skips the encoder's priming and ends at the last frame, so the file decodes to the length + /// that was written. + Aac, + /// Apple Lossless in an M4A (MP4) file (integer depths), 1–8 channels. + Alac, + /// An MP3 stream: mono or stereo at 32, 44.1 or 48 kHz, or 16, 22.05 or 24 kHz. + Mp3, + /// An AC-3 stream: up to 5.1 at 32, 44.1 or 48 kHz. + Ac3, + /// An E-AC-3 stream: up to 7.1 at 32, 44.1 or 48 kHz. + Eac3, + /// Ogg Vorbis, 1–8 channels at any sample rate, whole-buffer encoding only. Variable bit + /// rate: a [`RateControl::Bitrate`] picks the quality whose nominal rate it is + /// ([`quality_for_bitrate`]), [`RateControl::Quality`] sets it directly. + Ogg, } impl EncodeFormat { /// Every format, in the order menus list them. Whether [`encode`] writes one depends on its - /// feature. - pub const ALL: &[EncodeFormat] = &[EncodeFormat::Wav, EncodeFormat::Aiff, EncodeFormat::Flac]; + /// feature ([`EncodeFormat::is_built`]). + pub const ALL: &[EncodeFormat] = &[ + EncodeFormat::Wav, + EncodeFormat::Aiff, + EncodeFormat::Flac, + EncodeFormat::Alac, + EncodeFormat::Aac, + EncodeFormat::Mp3, + EncodeFormat::Ac3, + EncodeFormat::Eac3, + EncodeFormat::Ogg, + ]; /// The extension a new file of this format gets, lower case and without the dot: `"wav"`, - /// `"aif"`, `"flac"`. [`encode_format_for`] maps it back. + /// `"aif"`, `"flac"`, `"m4a"` (AAC and ALAC), `"mp3"`, `"ac3"`, `"eac3"`. + /// [`encode_format_for`] maps it back, `"m4a"` to AAC. pub fn extension(self) -> &'static str { match self { EncodeFormat::Wav => "wav", EncodeFormat::Aiff => "aif", EncodeFormat::Flac => "flac", + EncodeFormat::Aac | EncodeFormat::Alac => "m4a", + EncodeFormat::Mp3 => "mp3", + EncodeFormat::Ac3 => "ac3", + EncodeFormat::Eac3 => "eac3", + EncodeFormat::Ogg => "ogg", } } @@ -242,6 +337,11 @@ impl EncodeFormat { EncodeFormat::Wav => "audio/wav", EncodeFormat::Aiff => "audio/aiff", EncodeFormat::Flac => "audio/flac", + EncodeFormat::Aac | EncodeFormat::Alac => "audio/mp4", + EncodeFormat::Mp3 => "audio/mpeg", + EncodeFormat::Ac3 => "audio/ac3", + EncodeFormat::Eac3 => "audio/eac3", + EncodeFormat::Ogg => "audio/ogg", } } @@ -251,8 +351,163 @@ impl EncodeFormat { EncodeFormat::Wav => Container::Wav, EncodeFormat::Aiff => Container::Aiff, EncodeFormat::Flac => Container::Flac, + EncodeFormat::Aac | EncodeFormat::Alac => Container::Mp4, + EncodeFormat::Mp3 => Container::MpegAudio, + EncodeFormat::Ac3 | EncodeFormat::Eac3 => Container::Ac3, + EncodeFormat::Ogg => Container::Ogg, } } + + /// The codec of the audio this format writes. + pub fn codec(self) -> CodecId { + match self { + EncodeFormat::Wav | EncodeFormat::Aiff => CodecId::Pcm, + EncodeFormat::Flac => CodecId::Flac, + EncodeFormat::Aac => CodecId::Aac, + EncodeFormat::Alac => CodecId::Alac, + EncodeFormat::Mp3 => CodecId::Mp3, + EncodeFormat::Ac3 => CodecId::Ac3, + EncodeFormat::Eac3 => CodecId::Eac3, + EncodeFormat::Ogg => CodecId::Vorbis, + } + } + + /// The feature that builds this format's writer. + pub fn feature(self) -> &'static str { + match self { + EncodeFormat::Wav | EncodeFormat::Aiff => "pcm-encode", + EncodeFormat::Flac => "flac-encode", + EncodeFormat::Aac => "aac-encode", + EncodeFormat::Alac => "alac-encode", + EncodeFormat::Mp3 => "mp3-encode", + EncodeFormat::Ac3 => "ac3-encode", + EncodeFormat::Eac3 => "eac3-encode", + EncodeFormat::Ogg => "vorbis-encode", + } + } + + /// Whether [`encode`] writes this format in this build (its + /// [`feature`](Self::feature) is on): what to filter [`EncodeFormat::ALL`] by for a menu. + pub fn is_built(self) -> bool { + match self { + EncodeFormat::Wav | EncodeFormat::Aiff => cfg!(feature = "pcm-encode"), + EncodeFormat::Flac => cfg!(feature = "flac-encode"), + EncodeFormat::Aac => cfg!(feature = "aac-encode"), + EncodeFormat::Alac => cfg!(feature = "alac-encode"), + EncodeFormat::Mp3 => cfg!(feature = "mp3-encode"), + EncodeFormat::Ac3 => cfg!(feature = "ac3-encode"), + EncodeFormat::Eac3 => cfg!(feature = "eac3-encode"), + EncodeFormat::Ogg => cfg!(feature = "vorbis-encode"), + } + } + + /// The sample rates the format takes, ascending, or `None` when it has no fixed list (WAV, + /// AIFF, FLAC and ALAC take any rate their headers can hold). + pub fn sample_rates(self) -> Option<&'static [u32]> { + match self { + EncodeFormat::Aac => Some(&[7350, 8000, 11_025, 12_000, 16_000, 22_050, 24_000, 32_000, 44_100, 48_000, 64_000, 88_200, 96_000]), + EncodeFormat::Mp3 => Some(&[16_000, 22_050, 24_000, 32_000, 44_100, 48_000]), + EncodeFormat::Ac3 | EncodeFormat::Eac3 => Some(&[32_000, 44_100, 48_000]), + EncodeFormat::Wav | EncodeFormat::Aiff | EncodeFormat::Flac | EncodeFormat::Alac | EncodeFormat::Ogg => None, + } + } + + /// The most channels a file of this format holds. A WAV's frame size is a 16-bit field, so it + /// holds 65535 / bytes per sample: 32767 at 16 bits (the count given here), 21845 at 24 and + /// 16383 at 32; [`EncodeOptions::check`] answers for the depth chosen. + pub fn max_channels(self) -> u16 { + match self { + EncodeFormat::Wav | EncodeFormat::Aiff => 32_767, + EncodeFormat::Flac | EncodeFormat::Aac | EncodeFormat::Alac | EncodeFormat::Eac3 | EncodeFormat::Ogg => 8, + EncodeFormat::Ac3 => 6, + EncodeFormat::Mp3 => 2, + } + } + + /// The speaker layouts of a format that has a fixed set of them, as WAV masks, for a channel + /// count: one each for AAC, ALAC and MP3, one per coding mode for AC-3 and E-AC-3. Empty for a + /// count the format has not, and for WAV, AIFF and FLAC, which have no such set. + pub(crate) fn layout_masks(self, channels: u16) -> &'static [u32] { + let per_count: &[&[u32]] = match self { + // Channel configurations 1–6, the 6.1 program config element, and configuration 7 + // (7.1 with a front pair left and right of centre). + EncodeFormat::Aac => &[&[0x4], &[0x3], &[0x7], &[0x107], &[0x607], &[0x60F], &[0x70F], &[0xFF]], + // Apple's layout tags for 1–8 channels, surrounds as back speakers. + EncodeFormat::Alac => &[&[0x4], &[0x3], &[0x7], &[0x107], &[0x37], &[0x3F], &[0x13F], &[0xFF]], + EncodeFormat::Mp3 => &[&[0x4], &[0x3]], + // The Vorbis I §4.3.9 layouts for 1–8 channels (6.1 with a back centre and sides, 7.1 + // with sides and backs), written in Vorbis order. + EncodeFormat::Ogg => &[&[0x4], &[0x3], &[0x7], &[0x33], &[0x37], &[0x3F], &[0x70F], &[0x63F]], + // The coding modes 1/0 to 3/2 with and without an LFE, by channel count. + EncodeFormat::Ac3 => &[&[0x4], &[0x3, 0xC], &[0x7, 0x103, 0xB], &[0x107, 0x603, 0xF, 0x10B], &[0x607, 0x10F, 0x60B], &[0x60F]], + // The same, and 7.1 with backs and sides. + EncodeFormat::Eac3 => { + &[&[0x4], &[0x3, 0xC], &[0x7, 0x103, 0xB], &[0x107, 0x603, 0xF, 0x10B], &[0x607, 0x10F, 0x60B], &[0x60F], &[], &[0x63F]] + } + EncodeFormat::Wav | EncodeFormat::Aiff | EncodeFormat::Flac => &[], + }; + usize::from(channels).checked_sub(1).and_then(|i| per_count.get(i)).copied().unwrap_or_default() + } + + /// The speaker layouts a format with a fixed set of them has for `channels` channels: one for + /// AAC, ALAC and MP3, one per coding mode for AC-3 and E-AC-3; empty when the format cannot + /// hold that many channels. `None` for WAV, AIFF and FLAC, which take any layout. Audio in + /// another layout has to be rearranged into one of these before it is written. + /// + /// ``` + /// use craft_audio::{ChannelLayout, EncodeFormat}; + /// + /// assert_eq!(EncodeFormat::Aac.fixed_layouts(6), Some(vec![ChannelLayout::SURROUND_5_1_SIDE])); + /// assert_eq!(EncodeFormat::Mp3.fixed_layouts(6), Some(vec![])); + /// assert_eq!(EncodeFormat::Wav.fixed_layouts(6), None); + /// ``` + pub fn fixed_layouts(self, channels: u16) -> Option> { + match self { + EncodeFormat::Wav | EncodeFormat::Aiff | EncodeFormat::Flac => None, + _ => Some(self.layout_masks(channels).iter().map(|&mask| ChannelLayout::from_mask(mask)).collect()), + } + } + + /// Whether the format holds audio with `layout`: for WAV, AIFF and FLAC any layout of up to + /// [`max_channels`](Self::max_channels) channels, for the others one of + /// [`fixed_layouts`](Self::fixed_layouts), where side and back surrounds stand for each other + /// in a layout with one surround pair. + pub fn supports_layout(self, layout: &ChannelLayout) -> bool { + match self { + EncodeFormat::Wav | EncodeFormat::Aiff | EncodeFormat::Flac => (1..=self.max_channels()).contains(&layout.channels()), + _ => self.layout_masks(layout.channels()).iter().any(|&mask| layout::channel_map(mask, None, layout).is_some()), + } + } + + /// Whether the format takes [`RateControl::Quality`]: AAC and Ogg Vorbis. The other lossy + /// formats are written at a constant bit rate only. + pub fn supports_quality(self) -> bool { + matches!(self, EncodeFormat::Aac | EncodeFormat::Ogg) + } + + /// Whether the format loses audio: AAC, MP3, AC-3 and E-AC-3. These take + /// [`EncodeOptions::rate_control`] and ignore the bit depth and dither. + pub fn is_lossy(self) -> bool { + matches!(self, EncodeFormat::Aac | EncodeFormat::Mp3 | EncodeFormat::Ac3 | EncodeFormat::Eac3 | EncodeFormat::Ogg) + } +} + +/// How a lossy format spends bits ([`EncodeOptions::rate_control`]). +#[non_exhaustive] +#[derive(Debug, Clone, Copy, PartialEq)] +#[cfg_attr(feature = "serde", derive(serde::Serialize, serde::Deserialize))] +pub enum RateControl { + /// A constant bit rate in bits per second, for all channels together. A format with a fixed + /// set of rates (MP3, AC-3) uses the highest one not above it; a rate outside what the + /// stream can carry is moved to the nearest one inside. AAC is written at 6 kbit/s or a + /// quarter of the sample rate per channel at least: from about half of that down its encoder + /// codes silence. Vorbis takes 64–500 kbit/s and writes the quality whose nominal rate that is + /// ([`quality_for_bitrate`]); its rate varies with the audio. + Bitrate(u32), + /// A constant quality from 0.0 (smallest) to 1.0 (best), at whatever bit rate each frame + /// needs. AAC and Vorbis (0.0 is Vorbis quality -1, 1.0 quality 10); values outside the range + /// are clamped, and NaN is an error. + Quality(f32), } /// Stored sample representation. @@ -411,9 +666,9 @@ pub enum BitDepth { pub struct EncodeOptions { /// Output file format. pub format: EncodeFormat, - /// Sample depth to write. + /// Sample depth to write (ignored by the lossy formats). pub bit_depth: BitDepth, - /// Add TPDF dither when quantizing to an integer depth. + /// Add TPDF dither when quantizing to an integer depth (ignored by the lossy formats). pub dither: bool, /// Write a BWF `bext` chunk (WAV only; ignored for other formats). pub bwf: Option, @@ -424,15 +679,27 @@ pub struct EncodeOptions { /// the count (they read 5 and 6 channels with side surrounds and 7 as 6.1 with sides, so the /// usual layouts for those counts get one; see the README's limits); AIFF has a fixed channel /// order and ignores it. It must have as many channels as the audio. + /// + /// AAC, ALAC, MP3, AC-3 and E-AC-3 store channels in their own order, so here the layout says + /// which speaker each channel of the audio (in WAV order) is, and the channels are reordered + /// to match. Each of these formats has a fixed set of layouts (AAC and ALAC one per channel + /// count, AC-3 one per coding mode); a layout outside it is [`AudioError::Unsupported`]. Side + /// and back surrounds are taken as the same pair where the format has only one. pub layout: Option, /// Write WAV in the RF64 layout even below 4 GiB (WAV only; ignored for other formats). pub force_rf64: bool, + /// The bit rate or quality of a lossy format (ignored by the others). `None` is a constant + /// 64 kbit/s per channel for AAC, 96 kbit/s per channel for MP3, and 96 kbit/s per + /// full-bandwidth channel (the LFE not counted) for AC-3 and E-AC-3, moved to the nearest rate + /// the stream can carry. + pub rate_control: Option, } impl EncodeOptions { - /// Options for `format`: 24-bit, no dither, no BWF chunk, the usual layout. + /// Options for `format`: 24-bit, no dither, no BWF chunk, the usual layout, the format's + /// default bit rate. pub fn new(format: EncodeFormat) -> Self { - Self { format, bit_depth: BitDepth::Int24, dither: false, bwf: None, layout: None, force_rf64: false } + Self { format, bit_depth: BitDepth::Int24, dither: false, bwf: None, layout: None, force_rf64: false, rate_control: None } } /// Sets `format`. @@ -470,6 +737,18 @@ impl EncodeOptions { self.force_rf64 = force_rf64; self } + + /// Sets `rate_control` to a constant bit rate in bits per second ([`RateControl::Bitrate`]). + pub fn with_bitrate(mut self, bits_per_second: u32) -> Self { + self.rate_control = Some(RateControl::Bitrate(bits_per_second)); + self + } + + /// Sets `rate_control` to a constant quality from 0.0 to 1.0 ([`RateControl::Quality`]). + pub fn with_quality(mut self, quality: f32) -> Self { + self.rate_control = Some(RateControl::Quality(quality)); + self + } } impl Default for EncodeOptions { @@ -493,3 +772,6 @@ mod tests { assert!(matches!(e, AudioError::Encode(_)), "{e}"); } } + +/// DVD and Blu-ray LPCM packet decoding. +pub mod lpcm; diff --git a/crates/audio/src/lpcm.rs b/crates/audio/src/lpcm.rs new file mode 100644 index 000000000..5ec13b35d --- /dev/null +++ b/crates/audio/src/lpcm.rs @@ -0,0 +1,118 @@ +//! DVD-Video and Blu-ray LPCM packet grammar and sample decoding. + +/// Blu-ray / AVCHD LPCM packet header (4 bytes before the samples). +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct BlurayLpcm { + /// Sample rate in hertz. + pub sample_rate: u32, + /// Number of decoded channels. + pub channels: usize, + /// Significant bits per sample. + pub bits: u32, +} + +/// Channels of each channel_assignment code (mono, stereo, 3/0, 2/1, 3/1, 2/2, 3/2, 3/2+LFE, +/// 3/4, 3/4+LFE). +const BD_LPCM_CHANNELS: [usize; 12] = [0, 1, 0, 2, 3, 3, 4, 4, 5, 6, 7, 8]; + +/// Parse and validate the four-byte Blu-ray LPCM header. +pub fn bluray_lpcm_header(p: &[u8]) -> Option { + if p.len() < 4 { + return None; + } + let channels = *BD_LPCM_CHANNELS.get((p[2] >> 4) as usize).filter(|&&c| c > 0)?; + let sample_rate = match p[2] & 15 { + 1 => 48_000, + 4 => 96_000, + 5 => 192_000, + _ => return None, + }; + let bits = match p[3] >> 6 { + 1 => 16, + 2 => 20, + 3 => 24, + _ => return None, + }; + Some(BlurayLpcm { sample_rate, channels, bits }) +} + +/// Decode one Blu-ray LPCM packet (big-endian; channels padded to an even count; 20-bit +/// samples are stored in 24 bits). +pub fn decode_bluray_lpcm(p: &[u8]) -> Option>> { + let h = bluray_lpcm_header(p)?; + let coded = (h.channels + 1) & !1; + let bps = if h.bits == 16 { 2 } else { 3 }; + let data = &p[4..]; + let n = data.len() / (coded * bps); + let mut out = vec![Vec::with_capacity(n); h.channels]; + for k in 0..n { + for (c, o) in out.iter_mut().enumerate() { + let at = (k * coded + c) * bps; + let v = if bps == 2 { + i16::from_be_bytes([data[at], data[at + 1]]) as f32 / 32768.0 + } else { + (i32::from_be_bytes([data[at], data[at + 1], data[at + 2], 0]) >> 8) as f32 / 8_388_608.0 + }; + o.push(v); + } + } + Some(out) +} + +/// Bytes per two sample frames of DVD LPCM. The caller must bound `channels` to the DVD channel range. +pub fn dvd_lpcm_group_bytes(channels: usize, bits: u32) -> usize { + match bits { + 16 => 4 * channels, + 20 => 5 * channels, + _ => 6 * channels, + } +} + +/// Decode DVD-Video LPCM (big-endian). 16-bit samples are interleaved; 20- and 24-bit samples +/// come in groups of two frames: the upper 16 bits of all the group's samples, then their low +/// 4 / 8 bits. The caller must bound `channels` to the DVD channel range. +pub fn decode_dvd_lpcm(data: &[u8], channels: usize, bits: u32) -> Vec> { + let ch = channels.max(1); + let group = dvd_lpcm_group_bytes(ch, bits); + let groups = data.len() / group; + let mut out = vec![Vec::with_capacity(groups * 2); ch]; + for g in 0..groups { + let b = &data[g * group..(g + 1) * group]; + for f in 0..2 { + for (c, o) in out.iter_mut().enumerate() { + let k = f * ch + c; + let hi = i16::from_be_bytes([b[2 * k], b[2 * k + 1]]) as i32; + let v = match bits { + 16 => hi as f32 / 32768.0, + 20 => { + let nib = b[4 * ch + k / 2]; + let lo = if k.is_multiple_of(2) { nib >> 4 } else { nib & 15 } as i32; + ((hi << 4) | lo) as f32 / 524_288.0 + } + _ => ((hi << 8) | b[4 * ch + k] as i32) as f32 / 8_388_608.0, + }; + o.push(v); + } + } + } + out +} + +#[cfg(test)] +mod tests { + use super::*; + #[test] + fn bluray_header_and_even_channel_padding() { + let packet = [0, 0, 0x31, 0x40, 0x40, 0, 0xc0, 0]; + assert_eq!(bluray_lpcm_header(&packet), Some(BlurayLpcm { sample_rate: 48_000, channels: 2, bits: 16 })); + assert_eq!(decode_bluray_lpcm(&packet), Some(vec![vec![0.5], vec![-0.5]])); + assert_eq!(decode_bluray_lpcm(&packet[..3]), None); + } + #[test] + fn dvd_lpcm_16_20_and_24_bit_groups() { + assert_eq!(decode_dvd_lpcm(&[0x40, 0, 0xc0, 0], 1, 16), vec![vec![0.5, -0.5]]); + assert_eq!(decode_dvd_lpcm(&[0x40, 0, 0xc0, 0, 0], 1, 20), vec![vec![0.5, -0.5]]); + assert_eq!(decode_dvd_lpcm(&[0x40, 0, 0xc0, 0, 0, 0], 1, 24), vec![vec![0.5, -0.5]]); + assert_eq!(decode_dvd_lpcm(&[1, 2, 3], 1, 16), vec![Vec::::new()]); + } +} diff --git a/crates/audio/src/opus/decode/celt/mdct.rs b/crates/audio/src/opus/decode/celt/mdct.rs index b3bfcf0f7..7503d9ebe 100644 --- a/crates/audio/src/opus/decode/celt/mdct.rs +++ b/crates/audio/src/opus/decode/celt/mdct.rs @@ -4,141 +4,7 @@ //! producing the N/2 "folded" time samples; the window overlap region is then unfolded against the //! folded tail of the previous block, which the caller keeps in place in the output buffer. -#[derive(Clone, Copy, Default, Debug)] -pub struct Cpx { - pub re: f32, - pub im: f32, -} - -impl Cpx { - #[inline] - fn mul(self, o: Cpx) -> Cpx { - Cpx { re: self.re * o.re - self.im * o.im, im: self.re * o.im + self.im * o.re } - } - #[inline] - fn add(self, o: Cpx) -> Cpx { - Cpx { re: self.re + o.re, im: self.im + o.im } - } - #[inline] - fn sub(self, o: Cpx) -> Cpx { - Cpx { re: self.re - o.re, im: self.im - o.im } - } -} - -/// Mixed-radix (2, 3, 4, 5) forward complex FFT, unscaled. Radix 2, 3 and 4 use dedicated -/// butterflies; radix 5 uses a generic one. -pub struct Fft { - n: usize, - factors: Vec, - twiddle: Vec, -} - -impl Fft { - pub fn new(n: usize) -> Fft { - let mut factors = Vec::new(); - let mut m = n; - for p in [4usize, 2, 3, 5] { - while m.is_multiple_of(p) { - factors.push(p); - m /= p; - } - } - assert_eq!(m, 1, "unsupported FFT size {n}"); - let twiddle = (0..n) - .map(|k| { - let ph = -2.0 * std::f64::consts::PI * k as f64 / n as f64; - Cpx { re: ph.cos() as f32, im: ph.sin() as f32 } - }) - .collect(); - Fft { n, factors, twiddle } - } - - pub fn process(&self, data: &mut [Cpx], scratch: &mut Vec) { - scratch.clear(); - scratch.extend_from_slice(&data[..self.n]); - self.rec(&mut data[..self.n], scratch, 1, self.n, 0, 1); - } - - fn rec(&self, out: &mut [Cpx], inp: &[Cpx], in_stride: usize, n: usize, fi: usize, tw_stride: usize) { - let p = self.factors[fi]; - let m = n / p; - if m == 1 { - for (i, o) in out.iter_mut().enumerate().take(p) { - *o = inp[i * in_stride]; - } - } else { - for q in 0..p { - self.rec(&mut out[q * m..(q + 1) * m], &inp[q * in_stride..], in_stride * p, m, fi + 1, tw_stride * p); - } - } - let tw = &self.twiddle; - match p { - 2 => { - for k in 0..m { - let a0 = out[k]; - let a1 = out[m + k].mul(tw[k * tw_stride]); - out[k] = a0.add(a1); - out[m + k] = a0.sub(a1); - } - } - 4 => { - for k in 0..m { - let t0 = out[k]; - let t1 = out[m + k].mul(tw[k * tw_stride]); - let t2 = out[2 * m + k].mul(tw[2 * k * tw_stride]); - let t3 = out[3 * m + k].mul(tw[3 * k * tw_stride]); - let s0 = t0.add(t2); - let d0 = t0.sub(t2); - let s1 = t1.add(t3); - let d1 = t1.sub(t3); - out[k] = s0.add(s1); - out[2 * m + k] = s0.sub(s1); - // -j * d1 = (d1.im, -d1.re) - out[m + k] = Cpx { re: d0.re + d1.im, im: d0.im - d1.re }; - out[3 * m + k] = Cpx { re: d0.re - d1.im, im: d0.im + d1.re }; - } - } - 3 => { - // sin(-2*pi/3) = -sqrt(3)/2 - let h = tw[self.n / 3].im; - for k in 0..m { - let t0 = out[k]; - let t1 = out[m + k].mul(tw[k * tw_stride]); - let t2 = out[2 * m + k].mul(tw[2 * k * tw_stride]); - let s = t1.add(t2); - let d = t1.sub(t2); - let a = Cpx { re: t0.re - 0.5 * s.re, im: t0.im - 0.5 * s.im }; - out[k] = t0.add(s); - // j * h * d = (-h * d.im, h * d.re) - out[m + k] = Cpx { re: a.re - h * d.im, im: a.im + h * d.re }; - out[2 * m + k] = Cpx { re: a.re + h * d.im, im: a.im - h * d.re }; - } - } - _ => { - // Radix 5 (at most once per transform): generic butterfly with the five roots of - // unity looked up once. - let mut w5 = [Cpx::default(); 5]; - for (q, w) in w5.iter_mut().enumerate() { - *w = tw[q * (self.n / p)]; - } - let mut t = [Cpx::default(); 5]; - for k in 0..m { - t[0] = out[k]; - for q in 1..p { - t[q] = out[q * m + k].mul(tw[q * k * tw_stride]); - } - for q2 in 0..p { - let mut acc = t[0]; - for q in 1..p { - acc = acc.add(t[q].mul(w5[(q * q2) % p])); - } - out[q2 * m + k] = acc; - } - } - } - } - } -} +use crate::opus::fft::{Cpx, Fft}; pub struct Imdct { /// Full MDCT length (2x the number of coefficients). diff --git a/crates/audio/src/opus/decode/celt/rate.rs b/crates/audio/src/opus/decode/celt/rate.rs index 782d68d5c..29e8566c1 100644 --- a/crates/audio/src/opus/decode/celt/rate.rs +++ b/crates/audio/src/opus/decode/celt/rate.rs @@ -1,272 +1,45 @@ //! Pulse cache and bit allocation (RFC 6716 §4.3.3, §4.3.4.1). -use super::tables::{ALLOC, EBANDS, LOG2_FRAC_TABLE, MAX_LM, NB_EBANDS, OVERLAP}; +use super::tables::{EBANDS, LOG2_FRAC_TABLE, MAX_LM, NB_EBANDS, OVERLAP}; use crate::opus::decode::range::{BITRES, RangeDecoder}; -const MAX_PSEUDO: i32 = 40; -const LOG_MAX_PSEUDO: i32 = 6; +pub use crate::opus::rate_common::get_pulses; +#[cfg(test)] +use crate::opus::rate_common::log2_frac; +#[cfg(test)] +pub use crate::opus::rate_common::pvq_v; pub const MAX_FINE_BITS: i32 = 8; const FINE_OFFSET: i32 = 21; pub const QTHETA_OFFSET: i32 = 4; pub const QTHETA_OFFSET_TWOPHASE: i32 = 16; const ALLOC_STEPS: i32 = 6; -/// Number of pulses represented by pseudo-pulse index `i`. -pub fn get_pulses(i: i32) -> i32 { - if i < 8 { i } else { (8 + (i & 7)) << ((i >> 3) - 1) } -} - -/// Conservative (rounded up) log2 of `val` with `frac` fractional bits. -pub fn log2_frac(mut val: u32, mut frac: u32) -> i32 { - let mut l = crate::opus::decode::range::ilog(val); - if val & val.wrapping_sub(1) != 0 { - if l > 16 { - val = ((val - 1) >> (l - 16)) + 1; - } else { - val <<= 16 - l; - } - let mut l = (l - 1) << frac; - loop { - let b = (val >> 16) as i32; - l += b << frac; - val = (val + b as u32) >> b; - val = (val * val + 0x7FFF) >> 15; - if frac == 0 { - break; - } - frac -= 1; - } - l + (val > 0x8000) as i32 - } else { - l -= 1; - l << frac - } -} - -/// PVQ codebook size V(N, K) as u128 (saturating far above 2^32). -pub fn pvq_v(n: usize, k: usize) -> u128 { - // V(N,K) = V(N-1,K) + V(N,K-1) + V(N-1,K-1); computed row by row. - if k == 0 { - return 1; - } - if n == 0 { - return 0; - } - let mut row = vec![0u128; k + 1]; // V(0, k) - row[0] = 1; - for _ in 0..n { - let mut prev_diag = row[0]; // V(n-1, 0) - // V(n, 0) = 1 - for kk in 1..=k { - let up = row[kk]; // V(n-1, kk) - let v = up.saturating_add(row[kk - 1]).saturating_add(prev_diag); - prev_diag = up; - row[kk] = v.min(u128::MAX >> 8); - } - row[0] = 1; - } - row[k] -} - -fn fits_in32(n: usize, k: i32) -> bool { - pvq_v(n, k as usize) <= u32::MAX as u128 -} - -/// Precomputed per-mode data shared by all CELT decoders. +/// Decoder mode: shared allocation tables and decoder-specific inverse transforms. pub struct Mode { - /// log2 of band width in 1/8 bits. - pub log_n: [i32; NB_EBANDS], - /// Index into `cache_bits` for `(LM+1)*NB_EBANDS + band` (-1 when unused). - pub cache_index: Vec, - /// `cache[0]` = max pseudo-pulses; `cache[k]` = bits(k) - 1 in 1/8 bits. - pub cache_bits: Vec, - /// Per-band maximum allocation: `[(2*LM + C-1)*NB_EBANDS + band]`. - pub caps: Vec, + core: crate::opus::rate_common::Mode, pub window: [f32; OVERLAP], pub mdct: [super::mdct::Imdct; MAX_LM + 1], } - impl Mode { - pub fn new() -> Mode { - let mut log_n = [0i32; NB_EBANDS]; - for (i, l) in log_n.iter_mut().enumerate() { - *l = log2_frac((EBANDS[i + 1] - EBANDS[i]) as u32, BITRES); - } - let (cache_index, cache_bits) = compute_pulse_cache(MAX_LM as i32); - let mut m = Mode { - log_n, - cache_index, - cache_bits, - caps: Vec::new(), + pub fn new() -> Self { + let mut mode = Self { + core: crate::opus::rate_common::Mode::new(), window: [0.0; OVERLAP], mdct: [super::mdct::Imdct::new(1920), super::mdct::Imdct::new(960), super::mdct::Imdct::new(480), super::mdct::Imdct::new(240)], }; - m.caps = compute_caps(&m); for i in 0..OVERLAP { let x = std::f64::consts::FRAC_PI_2 * (i as f64 + 0.5) / OVERLAP as f64; let s = x.sin(); - m.window[i] = (std::f64::consts::FRAC_PI_2 * s * s).sin() as f32; + mode.window[i] = (std::f64::consts::FRAC_PI_2 * s * s).sin() as f32; } - m - } - - #[inline] - pub fn cache(&self, lm: i32, band: usize) -> &[u8] { - let idx = self.cache_index[((lm + 1) as usize) * NB_EBANDS + band]; - &self.cache_bits[idx.max(0) as usize..] - } - - pub fn bits2pulses(&self, band: usize, lm: i32, bits: i32) -> i32 { - let cache = self.cache(lm, band); - let mut lo = 0i32; - let mut hi = cache[0] as i32; - let bits = bits - 1; - for _ in 0..LOG_MAX_PSEUDO { - let mid = (lo + hi + 1) >> 1; - if cache[mid as usize] as i32 >= bits { - hi = mid; - } else { - lo = mid; - } - } - let lo_bits = if lo == 0 { -1 } else { cache[lo as usize] as i32 }; - if bits - lo_bits <= cache[hi as usize] as i32 - bits { lo } else { hi } - } - - pub fn pulses2bits(&self, band: usize, lm: i32, pulses: i32) -> i32 { - if pulses == 0 { 0 } else { self.cache(lm, band)[pulses as usize] as i32 + 1 } - } - - /// Per-band allocation caps for this frame (`init_caps`). - pub fn init_caps(&self, lm: usize, c: usize) -> [i32; NB_EBANDS] { - let mut cap = [0i32; NB_EBANDS]; - for (i, cp) in cap.iter_mut().enumerate() { - let n = (EBANDS[i + 1] - EBANDS[i]) << lm; - *cp = ((self.caps[NB_EBANDS * (2 * lm + c - 1) + i] as i32 + 64) * c as i32 * n) >> 2; - } - cap + mode } } - -fn compute_pulse_cache(lm_max: i32) -> (Vec, Vec) { - let nb = NB_EBANDS; - let rows = (lm_max + 2) as usize; - let mut index = vec![-1i16; nb * rows]; - let mut entries: Vec<(usize, i32, usize)> = Vec::new(); // (N, K, offset) - let mut curr = 0usize; - for i in 0..rows { - for j in 0..nb { - let n = (((EBANDS[j + 1] - EBANDS[j]) << i) >> 1) as usize; - index[i * nb + j] = -1; - // Reuse the entry of an earlier band with the same size. - 'search: for k in 0..=i { - for nn in 0..nb { - if k == i && nn >= j { - break; - } - if n == (((EBANDS[nn + 1] - EBANDS[nn]) << k) >> 1) as usize { - index[i * nb + j] = index[k * nb + nn]; - break 'search; - } - } - } - if index[i * nb + j] == -1 && n != 0 { - let mut k = 0; - while k < MAX_PSEUDO && fits_in32(n, get_pulses(k + 1)) { - k += 1; - } - index[i * nb + j] = curr as i16; - entries.push((n, k, curr)); - curr += (k + 1) as usize; - } - } +impl std::ops::Deref for Mode { + type Target = crate::opus::rate_common::Mode; + fn deref(&self) -> &Self::Target { + &self.core } - let mut bits = vec![0u8; curr]; - for &(n, k, off) in &entries { - let maxk = get_pulses(k) as usize; - let tmp = required_bits(n, maxk, BITRES); - for j in 1..=k { - bits[off + j as usize] = (tmp[get_pulses(j) as usize] - 1) as u8; - } - bits[off] = k as u8; - } - (index, bits) -} - -/// Bits (1/2^frac units, rounded up) needed for each K in `0..=maxk` for dimension `n`. -fn required_bits(n: usize, maxk: usize, frac: u32) -> Vec { - let mut out = vec![0i32; maxk + 1]; - if n == 1 { - for o in out.iter_mut().skip(1) { - *o = 1 << frac; - } - } else { - for (k, o) in out.iter_mut().enumerate().skip(1) { - let v = pvq_v(n, k); - *o = log2_frac(v as u32, frac); - } - } - out -} - -fn compute_caps(m: &Mode) -> Vec { - let nb = NB_EBANDS; - let mut caps = vec![0u8; (MAX_LM + 1) * 2 * nb]; - for i in 0..=MAX_LM as i32 { - for c in 1..=2i32 { - for j in 0..nb { - let mut n0 = EBANDS[j + 1] - EBANDS[j]; - let max_bits = if n0 << i == 1 { - (c * (1 + MAX_FINE_BITS)) << BITRES - } else { - let mut lm0 = 0; - if n0 > 2 { - n0 >>= 1; - lm0 -= 1; - } else if n0 <= 1 { - lm0 = i.min(1); - n0 <<= lm0; - } - let pcache = m.cache(lm0, j); - let mut mb = pcache[pcache[0] as usize] as i32 + 1; - let mut n = n0; - for k in 0..(i - lm0) { - mb <<= 1; - let offset = ((m.log_n[j] + ((lm0 + k) << BITRES)) >> 1) - QTHETA_OFFSET; - let num = 459 * ((2 * n - 1) * offset + mb); - let den = ((2 * n - 1) << 9) - 459; - let qb = ((num + (den >> 1)) / den).min(57); - mb += qb; - n <<= 1; - } - if c == 2 { - mb <<= 1; - let offset = ((m.log_n[j] + (i << BITRES)) >> 1) - if n == 2 { QTHETA_OFFSET_TWOPHASE } else { QTHETA_OFFSET }; - let ndof = 2 * n - 1 - (n == 2) as i32; - let p = if n == 2 { 512 } else { 487 }; - let num = p * (mb + ndof * offset); - let den = (ndof << 9) - p; - let qb = ((num + (den >> 1)) / den).min(if n == 2 { 64 } else { 61 }); - mb += qb; - } - let ndof = c * n + if c == 2 && n > 2 { 1 } else { 0 }; - let mut offset = ((m.log_n[j] + (i << BITRES)) >> 1) - FINE_OFFSET; - if n == 2 { - offset += (1 << BITRES) >> 2; - } - let num = mb + ndof * offset; - let den = (ndof - 1) << BITRES; - let qb = ((num + (den >> 1)) / den).min(MAX_FINE_BITS); - mb += (c * qb) << BITRES; - mb - }; - let w = c * ((EBANDS[j + 1] - EBANDS[j]) << i); - let v = (4 * max_bits / w) - 64; - caps[((i * 2 + c - 1) as usize) * nb + j] = v.clamp(0, 255) as u8; - } - } - } - caps } /// Result of the allocation process. @@ -296,89 +69,8 @@ pub struct AllocationInput<'a, 'd> { pub fn compute_allocation(input: AllocationInput<'_, '_>) -> Allocation { let AllocationInput { mode: m, start, end, offsets, cap, trim: alloc_trim, total, channels: c, lm, decoder: dec } = input; - let ci = c as i32; - let lmi = lm as i32; - let mut total = total.max(0); - let mut skip_start = start; - let skip_rsv = if total >= 1 << BITRES { 1 << BITRES } else { 0 }; - total -= skip_rsv; - let mut intensity_rsv = 0; - let mut dual_stereo_rsv = 0; - if c == 2 { - intensity_rsv = LOG2_FRAC_TABLE[end - start]; - if intensity_rsv > total { - intensity_rsv = 0; - } else { - total -= intensity_rsv; - dual_stereo_rsv = if total >= 1 << BITRES { 1 << BITRES } else { 0 }; - total -= dual_stereo_rsv; - } - } - let mut bits1 = [0i32; NB_EBANDS]; - let mut bits2 = [0i32; NB_EBANDS]; - let mut thresh = [0i32; NB_EBANDS]; - let mut trim_offset = [0i32; NB_EBANDS]; - for j in start..end { - let w = EBANDS[j + 1] - EBANDS[j]; - thresh[j] = (ci << BITRES).max(((3 * w) << lmi << BITRES) >> 4); - trim_offset[j] = (ci * w * (alloc_trim - 5 - lmi) * (end as i32 - j as i32 - 1) * (1 << (lmi + BITRES as i32))) >> 6; - if w << lmi == 1 { - trim_offset[j] -= ci << BITRES; - } - } - let nb_alloc = ALLOC.len() as i32; - let mut lo = 1; - let mut hi = nb_alloc - 1; - loop { - let mut done = false; - let mut psum = 0; - let mid = (lo + hi) >> 1; - for j in (start..end).rev() { - let n = EBANDS[j + 1] - EBANDS[j]; - let mut bitsj = (ci * n * (ALLOC[mid as usize][j] as i32)) << lmi >> 2; - if bitsj > 0 { - bitsj = (bitsj + trim_offset[j]).max(0); - } - bitsj += offsets[j]; - if bitsj >= thresh[j] || done { - done = true; - psum += bitsj.min(cap[j]); - } else if bitsj >= ci << BITRES { - psum += ci << BITRES; - } - } - if psum > total { - hi = mid - 1; - } else { - lo = mid + 1; - } - if lo > hi { - break; - } - } - hi = lo; - lo -= 1; - for j in start..end { - let n = EBANDS[j + 1] - EBANDS[j]; - let mut bits1j = (ci * n * (ALLOC[lo as usize][j] as i32)) << lmi >> 2; - let mut bits2j = if hi >= nb_alloc { cap[j] } else { (ci * n * (ALLOC[hi as usize][j] as i32)) << lmi >> 2 }; - if bits1j > 0 { - bits1j = (bits1j + trim_offset[j]).max(0); - } - if bits2j > 0 { - bits2j = (bits2j + trim_offset[j]).max(0); - } - if lo > 0 { - bits1j += offsets[j]; - } - bits2j += offsets[j]; - if offsets[j] > 0 { - skip_start = j; - } - bits2j = (bits2j - bits1j).max(0); - bits1[j] = bits1j; - bits2[j] = bits2j; - } + let crate::opus::rate_common::Bracket { skip_start, bits1, bits2, thresh, total, skip_rsv, intensity_rsv, dual_stereo_rsv } = + crate::opus::rate_common::bracket(start, end, offsets, cap, alloc_trim, total, c, lm); interp_bits2pulses(InterpPulsesInput { mode: m, start, diff --git a/crates/audio/src/opus/decode/celt/tables.rs b/crates/audio/src/opus/decode/celt/tables.rs index f2381f714..b65d05c09 100644 --- a/crates/audio/src/opus/decode/celt/tables.rs +++ b/crates/audio/src/opus/decode/celt/tables.rs @@ -14,6 +14,7 @@ pub const MAX_LM: usize = 3; pub const PREEMPH: f32 = 0.850_006_1; /// Static allocation table (Table 57), indexed `[quality][band]`, in 1/32 bit per MDCT bin. +#[cfg(not(feature = "opus-encode"))] pub const ALLOC: [[u8; NB_EBANDS]; 11] = transpose([ [0, 90, 110, 118, 126, 134, 144, 152, 162, 172, 200], [0, 80, 100, 110, 119, 127, 137, 145, 155, 165, 200], @@ -38,6 +39,7 @@ pub const ALLOC: [[u8; NB_EBANDS]; 11] = transpose([ [0, 0, 0, 0, 0, 1, 1, 1, 1, 20, 104], ]); +#[cfg(not(feature = "opus-encode"))] const fn transpose(t: [[u8; 11]; NB_EBANDS]) -> [[u8; NB_EBANDS]; 11] { let mut out = [[0u8; NB_EBANDS]; 11]; let mut b = 0; diff --git a/crates/audio/src/opus/decode/silk/resampler.rs b/crates/audio/src/opus/decode/silk/resampler.rs index 0671778cc..7d3f803ef 100644 --- a/crates/audio/src/opus/decode/silk/resampler.rs +++ b/crates/audio/src/opus/decode/silk/resampler.rs @@ -4,25 +4,7 @@ //! RFC 6716 Table 54 (and the integer delays used when no rate conversion is needed), so that the //! SILK output lines up with the CELT layer in Hybrid mode. -/// Dot product with eight independent partial sums, so the additions pipeline and the loop -/// vectorises instead of waiting on one 4-cycle add chain. Rounds differently from a -/// sequential sum (the resampler is not normative). -#[inline] -fn dot8(a: &[f32], b: &[f32]) -> f32 { - let mut s = [0f32; 8]; - let (ca, ra) = a.as_chunks::<8>(); - let (cb, rb) = b.as_chunks::<8>(); - for (x, y) in ca.iter().zip(cb) { - for l in 0..8 { - s[l] += x[l] * y[l]; - } - } - let mut acc = ((s[0] + s[4]) + (s[1] + s[5])) + ((s[2] + s[6]) + (s[3] + s[7])); - for (x, y) in ra.iter().zip(rb) { - acc += x * y; - } - acc -} +use crate::opus::dot::dot8; #[derive(Clone)] pub struct Resampler { diff --git a/crates/audio/src/opus/dot.rs b/crates/audio/src/opus/dot.rs new file mode 100644 index 000000000..536209e0a --- /dev/null +++ b/crates/audio/src/opus/dot.rs @@ -0,0 +1,20 @@ +//! Shared fixed-order FIR dot product. + +/// Dot product with eight independent partial sums, which the compiler can vectorise (a single +/// running sum is bound by the latency of the float addition). The sums are combined pairwise. +#[inline] +pub(super) fn dot8(a: &[f32], b: &[f32]) -> f32 { + let mut s = [0f32; 8]; + let (ca, ra) = a.as_chunks::<8>(); + let (cb, rb) = b.as_chunks::<8>(); + for (x, y) in ca.iter().zip(cb) { + for l in 0..8 { + s[l] += x[l] * y[l]; + } + } + let mut acc = ((s[0] + s[4]) + (s[1] + s[5])) + ((s[2] + s[6]) + (s[3] + s[7])); + for (x, y) in ra.iter().zip(rb) { + acc += x * y; + } + acc +} diff --git a/crates/audio/src/opus/encoder/bands.rs b/crates/audio/src/opus/encoder/bands.rs new file mode 100644 index 000000000..697332eb2 --- /dev/null +++ b/crates/audio/src/opus/encoder/bands.rs @@ -0,0 +1,720 @@ +//! Band shape coding: PVQ search and codeword indexing, spreading, recursive band splitting +//! (RFC 6716 §4.3.4, encoder direction). +//! +//! The encoder codes 20 ms long-block frames only (no transients, no TF changes), and stereo +//! as dual (independent L/R) stereo, so a band is always coded by `quant_partition` with a single +//! block. Every bit-budget computation mirrors the decoder so both sides take the same +//! split/pulse decisions. + +use crate::opus::encoder::range::{BITRES, RangeEncoder, ilog}; +use crate::opus::encoder::rate::{Mode, QTHETA_OFFSET, get_pulses}; +use crate::opus::encoder::tables::EBANDS; +use crate::opus::encoder::{CeltFrame, rate::Allocation}; + +pub const SPREAD_NORMAL: usize = 2; + +#[inline] +fn frac_mul16(a: i32, b: i32) -> i32 { + (16384 + (a as i16 as i32) * (b as i16 as i32)) >> 15 +} + +fn bitexact_cos(x: i32) -> i32 { + let tmp = (4096 + x * x) >> 13; + let x2 = tmp; + let x2 = (32767 - x2) + frac_mul16(x2, -7651 + frac_mul16(x2, 8277 + frac_mul16(-626, x2))); + 1 + x2 +} + +fn bitexact_log2tan(isin: i32, icos: i32) -> i32 { + let lc = ilog(icos as u32); + let ls = ilog(isin as u32); + let icos = icos << (15 - lc); + let isin = isin << (15 - ls); + (ls - lc) * (1 << 11) + frac_mul16(isin, frac_mul16(isin, -2597) + 7932) - frac_mul16(icos, frac_mul16(icos, -2597) + 7932) +} + +fn compute_qn(n: i32, b: i32, offset: i32, pulse_cap: i32, stereo: bool) -> i32 { + const EXP2_TABLE8: [i32; 8] = [16384, 17866, 19483, 21247, 23170, 25267, 27554, 30048]; + let mut n2 = 2 * n - 1; + if stereo && n == 2 { + n2 -= 1; + } + let mut qb = (b + n2 * offset) / n2; + qb = qb.min(b - pulse_cap - (4 << BITRES)); + qb = qb.min(8 << BITRES); + if qb < (1 << BITRES >> 1) { + 1 + } else { + let qn = EXP2_TABLE8[(qb & 0x7) as usize] >> (14 - (qb >> BITRES)); + (qn + 1) >> 1 << 1 + } +} + +/// One elementary pass of the spreading rotation, inverted (the decoder applies the forward +/// pass; the encoder undoes it before the pulse search). +fn exp_rotation1_inv(x: &mut [f32], len: usize, stride: usize, c: f32, s: f32) { + if len <= stride { + return; + } + if len > 2 * stride { + for i in 0..=len - 2 * stride - 1 { + let x1 = x[i]; + let x2 = x[i + stride]; + x[i + stride] = c * x2 - s * x1; + x[i] = c * x1 + s * x2; + } + } + for i in (0..len - stride).rev() { + let x1 = x[i]; + let x2 = x[i + stride]; + x[i + stride] = c * x2 - s * x1; + x[i] = c * x1 + s * x2; + } +} + +/// Inverse of the decoder's spreading rotation (§4.3.4.3) for one block of `len` samples. +pub fn exp_rotation_inv(x: &mut [f32], len: usize, k: usize, spread: usize) { + const SPREAD_FACTOR: [usize; 3] = [15, 10, 5]; + if 2 * k >= len || spread == 0 { + return; + } + let factor = SPREAD_FACTOR[spread - 1]; + let gain = len as f32 / (len + factor * k) as f32; + let theta = 0.5 * gain * gain; + let c = (0.5 * std::f32::consts::PI * theta).cos(); + let s = (0.5 * std::f32::consts::PI * (1.0 - theta)).cos(); + let mut stride2 = 0; + if len >= 8 { + stride2 = 1; + while stride2 * stride2 + stride2 < len { + stride2 += 1; + } + } + // The decoder applies (stride2: s, c) then (1: c, s); undo in reverse order. + exp_rotation1_inv(x, len, 1, c, s); + if stride2 != 0 { + exp_rotation1_inv(x, len, stride2, s, c); + } +} + +/// Relative slack for the pulse-candidate filter in [`pvq_search`]: far above the f32 rounding of +/// the filter score (a few 1e-7), far below the spread of real scores. +const PVQ_SCORE_SLACK: f32 = 1e-5; + +/// Greedy PVQ search: the integer vector `y` with `sum |y| = k` closest in direction to `x`. +pub fn pvq_search(x: &[f32], k: usize, y: &mut [i32]) { + let n = x.len(); + y[..n].fill(0); + // `abs`, `dy` (2 y + 1, infinite in the padding) and the per-pass scores `score`, each padded to + // whole 8-lane chunks. + let np = n.next_multiple_of(8); + let mut buf = vec![0f32; 3 * np]; + let (abs, rest) = buf.split_at_mut(np); + let (dy, score) = rest.split_at_mut(np); + for (a, v) in abs.iter_mut().zip(x) { + *a = v.abs(); + } + let sum: f32 = abs[..n].iter().sum(); + if !sum.is_finite() || sum <= 1e-15 { + y[0] = k as i32; + return; + } + let mut placed = 0usize; + let mut xy = 0f32; + let mut yy = 0f32; + if k > n / 2 { + // Pre-project onto the pyramid (rounding down) and place the rest greedily. + let r = (k as f32 - 1.0) / sum; + for j in 0..n { + // Non-negative and at most k, so the truncating cast is the floor. + let p = (abs[j] * r) as i32; + y[j] = p; + placed += p as usize; + xy += abs[j] * p as f32; + yy += (p * p) as f32; + } + } + for j in 0..np { + dy[j] = if j < n { 2.0 * y[j] as f32 + 1.0 } else { f32::INFINITY }; + } + while placed < k { + // Vectorisable scores (xy + |x_j|)^2 / (yy + 2 y_j + 1) and their maximum. + for ((sc, &a), &d) in score.iter_mut().zip(abs.iter()).zip(dy.iter()) { + let t = xy + a; + *sc = t * t / (yy + d); + } + let mut lane_max = [0f32; 8]; + for c in score.as_chunks::<8>().0 { + for i in 0..8 { + lane_max[i] = if c[i] > lane_max[i] { c[i] } else { lane_max[i] }; + } + } + let top = lane_max.iter().fold(0f32, |m, &v| if v > m { v } else { m }); + // The exact cross-multiplied test below, in index order with the first maximum winning, only + // matters for entries within rounding distance of the top score: every other entry is beaten + // by the top one whenever it is reached and never beats it afterwards. Running the test on + // those alone therefore picks the same pulse as running it on all of them. + let filter = top > 1e-30 && top.is_finite(); + let lo = top * (1.0 - PVQ_SCORE_SLACK); + // Usually exactly one entry qualifies; its index falls out of a lane-wise count. + let mut count = [0u32; 8]; + let mut index = [0u32; 8]; + for (ci, c) in score.as_chunks::<8>().0.iter().enumerate() { + for i in 0..8 { + let hit = u32::from(c[i] >= lo); + count[i] += hit; + index[i] += hit * (ci * 8 + i) as u32; + } + } + let mut best = 0usize; + if filter && count.iter().sum::() == 1 { + best = index.iter().sum::() as usize; + } else { + let mut best_num = -1f32; + let mut best_den = 1f32; + for j in 0..n { + if !filter || score[j] >= lo { + let num = xy + abs[j]; + let num = num * num; + let den = yy + 2.0 * y[j] as f32 + 1.0; + if num * best_den > best_num * den { + best_num = num; + best_den = den; + best = j; + } + } + } + } + xy += abs[best]; + yy += 2.0 * y[best] as f32 + 1.0; + y[best] += 1; + dy[best] = 2.0 * y[best] as f32 + 1.0; + placed += 1; + } + for j in 0..n { + if x[j] < 0.0 { + y[j] = -y[j]; + } + } +} + +/// Largest band width and pulse count the encoder codes (176 coefficients, `get_pulses(40)` = 128). +const PVQ_MAX_N: usize = 176; +const PVQ_MAX_K: usize = 128; + +/// Row-major `V(n, k)` for `n <= max_n`, `k <= max_k` with row stride `max_k + 1`, saturated at +/// `u32::MAX` (a codebook that large is never coded, see `rate::fits_in32`). +fn build_pvq_table(max_n: usize, max_k: usize) -> Vec { + let w = max_k + 1; + let mut v = vec![0u32; (max_n + 1) * w]; + v[0] = 1; + for nn in 1..=max_n { + v[nn * w] = 1; + for kk in 1..=max_k { + let sum = v[(nn - 1) * w + kk] as u64 + v[nn * w + kk - 1] as u64 + v[(nn - 1) * w + kk - 1] as u64; + v[nn * w + kk] = sum.min(u32::MAX as u64) as u32; + } + } + v +} + +fn pvq_table() -> &'static [u32] { + static TABLE: std::sync::OnceLock> = std::sync::OnceLock::new(); + TABLE.get_or_init(|| build_pvq_table(PVQ_MAX_N, PVQ_MAX_K)) +} + +/// Encodes the PVQ codeword `y` (`sum |y| = k`) as the index the RFC 6716 §4.3.4.2 decoder maps +/// back to `y`. +pub fn encode_pulses(y: &[i32], k: usize, enc: &mut RangeEncoder) { + let n = y.len(); + // Shapes outside the coded range (never produced by the band layout) get a local table. + let local; + let (v, w) = if n <= PVQ_MAX_N && k <= PVQ_MAX_K { + (pvq_table(), PVQ_MAX_K + 1) + } else { + local = build_pvq_table(n, k); + (&local[..], k + 1) + }; + let row = |nn: usize| &v[nn * w..(nn + 1) * w]; + let total = row(n)[k] as u64; + let mut index = 0u64; + let mut kk = k; + for j in 0..n { + if kk == 0 { + break; + } + let rem = n - j; + let t = y[j].unsigned_abs() as usize; + let below = row(rem - 1); + // Codewords with |y_j| = t start at sum_{m < kk - t} V(rem-1, m) within their sign half. + let mut base: u64 = below[..kk - t].iter().map(|&x| x as u64).sum(); + if y[j] < 0 { + base += (below[kk] as u64 + row(rem)[kk] as u64) / 2; + } + index += base; + kk -= t; + } + debug_assert!(total <= u32::MAX as u64); + if total >= 2 { + enc.uint(index as u32, total as u32); + } +} + +pub struct BandEnc<'m, 'e> { + pub m: &'m Mode, + pub enc: &'e mut RangeEncoder, + pub i: usize, + pub spread: usize, + pub remaining_bits: i32, + iy: Vec, + tmp: Vec, +} + +impl BandEnc<'_, '_> { + /// `compute_theta` for a mono split (single block): measures the energy split between the + /// two halves and codes the quantised angle with the triangular pdf. + fn compute_theta_split(&mut self, xa: &[f32], xb: &[f32], b: &mut i32, lm: i32) -> (i32, i32, i32) { + let n = xa.len() as i32; + let i = self.i; + let pulse_cap = self.m.log_n[i] + lm * (1 << BITRES); + let offset = (pulse_cap >> 1) - QTHETA_OFFSET; + let qn = compute_qn(n, *b, offset, pulse_cap, false); + let tell = self.enc.tell_frac(); + let mut itheta = 0i32; + if qn != 1 { + let mid = xa.iter().map(|v| v * v).sum::().sqrt(); + let side = xb.iter().map(|v| v * v).sum::().sqrt(); + let ang = (side as f64).atan2(mid as f64) * std::f64::consts::FRAC_2_PI; + let it = ((ang * qn as f64 + 0.5).floor() as i32).clamp(0, qn) as u32; + let q = qn as u32; + let half = q >> 1; + let ft = (half + 1) * (half + 1); + let (fl, fs) = if it <= half { ((it * (it + 1)) >> 1, it + 1) } else { (ft - (((q + 1 - it) * (q + 2 - it)) >> 1), q + 1 - it) }; + self.enc.encode(fl, fl + fs, ft); + itheta = ((it * 16384) / q) as i32; + } + let qalloc = self.enc.tell_frac() - tell; + *b -= qalloc; + let delta = if itheta == 0 { + -16384 + } else if itheta == 16384 { + 16384 + } else { + let imid = bitexact_cos(itheta); + let iside = bitexact_cos(16384 - itheta); + frac_mul16((n - 1) << 7, bitexact_log2tan(iside, imid)) + }; + (itheta, delta, qalloc) + } + + /// `quant_partition` (single block, encoder side). + pub fn quant_partition(&mut self, x: &[f32], mut b: i32, mut lm: i32) { + let n = x.len(); + let i = self.i; + let cache = self.m.cache(lm, i); + let cache_max = cache[cache[0] as usize] as i32; + if lm != -1 && b > cache_max + 12 && n > 2 { + let half = n >> 1; + let (xa, xb) = x.split_at(half); + lm -= 1; + let (itheta, delta, qalloc) = self.compute_theta_split(xa, xb, &mut b, lm); + let mut mbits = 0.max(b.min((b - delta) / 2)); + let mut sbits = b - mbits; + self.remaining_bits -= qalloc; + let mut rebalance = self.remaining_bits; + if mbits >= sbits { + self.quant_partition(xa, mbits, lm); + rebalance = mbits - (rebalance - self.remaining_bits); + if rebalance > 3 << BITRES && itheta != 0 { + sbits += rebalance - (3 << BITRES); + } + self.quant_partition(xb, sbits, lm); + } else { + self.quant_partition(xb, sbits, lm); + rebalance = sbits - (rebalance - self.remaining_bits); + if rebalance > 3 << BITRES && itheta != 16384 { + mbits += rebalance - (3 << BITRES); + } + self.quant_partition(xa, mbits, lm); + } + return; + } + let m = self.m; + let mut q = m.bits2pulses(i, lm, b); + let mut curr_bits = m.pulses2bits(i, lm, q); + self.remaining_bits -= curr_bits; + while self.remaining_bits < 0 && q > 0 { + self.remaining_bits += curr_bits; + q -= 1; + curr_bits = m.pulses2bits(i, lm, q); + self.remaining_bits -= curr_bits; + } + if q != 0 { + let k = get_pulses(q) as usize; + let mut tmp = std::mem::take(&mut self.tmp); + tmp.clear(); + tmp.extend_from_slice(x); + exp_rotation_inv(&mut tmp, n, k, self.spread); + let mut iy = std::mem::take(&mut self.iy); + iy.clear(); + iy.resize(n, 0); + pvq_search(&tmp, k, &mut iy); + encode_pulses(&iy, k, self.enc); + self.iy = iy; + self.tmp = tmp; + } + } +} + +/// Shape-coding controls, independent of the selected band allocation. +pub struct BandCoding { + pub spread: usize, + pub total_bits: i32, +} + +/// Codes all band shapes (`quant_all_bands`, encoder side, long blocks). `x` holds +/// `frame.channels` channels of `n_total` normalised MDCT coefficients each. +pub fn quant_all_bands(m: &Mode, frame: CeltFrame, x: &[f32], n_total: usize, allocation: &Allocation, coding: BandCoding, enc: &mut RangeEncoder) { + let CeltFrame { start, end, channels: c, lm } = frame; + let Allocation { pulses, dual_stereo, intensity, coded_bands, balance, .. } = allocation; + let (dual_stereo, intensity, coded_bands, mut balance) = (*dual_stereo, *intensity, *coded_bands, *balance); + let BandCoding { spread, total_bits } = coding; + let mm = 1usize << lm; + debug_assert!(c == 1 || dual_stereo || intensity <= start, "stereo frames are coded as dual stereo"); + let mut ctx = BandEnc { m, enc, i: 0, spread, remaining_bits: 0, iy: Vec::new(), tmp: Vec::new() }; + for i in start..end { + ctx.i = i; + let xo = mm * EBANDS[i] as usize; + let n = mm * EBANDS[i + 1] as usize - xo; + let tell = ctx.enc.tell_frac(); + if i != start { + balance -= tell; + } + let remaining_bits = total_bits - tell - 1; + ctx.remaining_bits = remaining_bits; + let b = if i < coded_bands { + let curr_balance = balance / 3.min(coded_bands as i32 - i as i32); + 0.max(16383.min((remaining_bits + 1).min(pulses[i] + curr_balance))) + } else { + 0 + }; + // Bands at and above `intensity` (= coded_bands) carry no bits, so they emit no symbols. + if i < coded_bands || c == 1 { + if c == 2 && dual_stereo && i < intensity { + ctx.quant_partition(&x[xo..xo + n], b / 2, lm as i32); + ctx.quant_partition(&x[n_total + xo..n_total + xo + n], b / 2, lm as i32); + } else { + ctx.quant_partition(&x[xo..xo + n], b, lm as i32); + } + } + balance += pulses[i] + tell; + } +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::opus::encoder::range::dec::RangeDecoder; + + fn exp_rotation1(x: &mut [f32], len: usize, stride: usize, c: f32, s: f32) { + let ms = -s; + if len <= stride { + return; + } + for i in 0..len - stride { + let x1 = x[i]; + let x2 = x[i + stride]; + x[i + stride] = c * x2 + s * x1; + x[i] = c * x1 + ms * x2; + } + if len > 2 * stride { + for i in (0..=len - 2 * stride - 1).rev() { + let x1 = x[i]; + let x2 = x[i + stride]; + x[i + stride] = c * x2 + s * x1; + x[i] = c * x1 + ms * x2; + } + } + } + + /// The decoder's spreading rotation (single block). + fn exp_rotation(x: &mut [f32], len: usize, k: usize, spread: usize) { + const SPREAD_FACTOR: [usize; 3] = [15, 10, 5]; + if 2 * k >= len || spread == 0 { + return; + } + let factor = SPREAD_FACTOR[spread - 1]; + let gain = len as f32 / (len + factor * k) as f32; + let theta = 0.5 * gain * gain; + let c = (0.5 * std::f32::consts::PI * theta).cos(); + let s = (0.5 * std::f32::consts::PI * (1.0 - theta)).cos(); + let mut stride2 = 0; + if len >= 8 { + stride2 = 1; + while stride2 * stride2 + stride2 < len { + stride2 += 1; + } + } + if stride2 != 0 { + exp_rotation1(x, len, stride2, s, c); + } + exp_rotation1(x, len, 1, c, s); + } + + #[test] + fn spreading_inverse_is_exact() { + for &len in &[8usize, 16, 24, 48, 96, 176] { + for k in [1usize, 2, 3, 5] { + let orig: Vec = (0..len).map(|i| ((i * 31 % 17) as f32 - 8.0) / 8.0).collect(); + let mut x = orig.clone(); + exp_rotation_inv(&mut x, len, k, SPREAD_NORMAL); + exp_rotation(&mut x, len, k, SPREAD_NORMAL); + for (a, b) in x.iter().zip(&orig) { + assert!((a - b).abs() < 1e-4, "len={len} k={k}"); + } + } + } + } + + /// PVQ decoder of RFC 6716 §4.3.4.2 (as in FilmCraft's decoder). + fn decode_pulses(n: usize, k: usize, dec: &mut RangeDecoder) -> Vec { + let w = k + 1; + let mut v = vec![0u64; (n + 1) * w]; + v[0] = 1; + for nn in 1..=n { + v[nn * w] = 1; + for kk in 1..=k { + v[nn * w + kk] = v[(nn - 1) * w + kk].saturating_add(v[nn * w + kk - 1]).saturating_add(v[(nn - 1) * w + kk - 1]); + } + } + let total = v[n * w + k]; + let mut i = if total >= 2 { dec.uint(total as u32) as u64 } else { 0 }; + let mut y = vec![0i32; n]; + let mut kk = k; + for j in 0..n { + let rem = n - j; + if kk == 0 { + break; + } + let vv = |nn: usize, kx: usize| v[nn * w + kx]; + let mut p = (vv(rem - 1, kk) + vv(rem, kk)) / 2; + let neg = i >= p; + if neg { + i -= p; + } + let k0 = kk; + p -= vv(rem - 1, kk); + while p > i && kk > 0 { + kk -= 1; + p -= vv(rem - 1, kk); + } + let mag = (k0 - kk) as i32; + y[j] = if neg { -mag } else { mag }; + i -= p.min(i); + } + y + } + + fn all_codewords(n: usize, k: usize) -> Vec> { + if n == 0 { + return if k == 0 { vec![vec![]] } else { vec![] }; + } + let mut out = Vec::new(); + for t in -(k as i32)..=(k as i32) { + for mut rest in all_codewords(n - 1, k - t.unsigned_abs() as usize) { + rest.insert(0, t); + out.push(rest); + } + } + out + } + + #[test] + fn pvq_index_roundtrip_exhaustive() { + for n in 1..6usize { + for k in 1..6usize { + for y in all_codewords(n, k) { + let mut e = RangeEncoder::new(); + encode_pulses(&y, k, &mut e); + let buf = e.done(16); + let mut d = RangeDecoder::new(&buf); + assert_eq!(decode_pulses(n, k, &mut d), y, "n={n} k={k}"); + } + } + } + } + + /// The per-call table encoder this module used before the shared table. + fn encode_pulses_reference(y: &[i32], k: usize, enc: &mut RangeEncoder) { + let n = y.len(); + let w = k + 1; + let mut v = vec![0u64; (n + 1) * w]; + v[0] = 1; + for nn in 1..=n { + v[nn * w] = 1; + for kk in 1..=k { + let a = v[(nn - 1) * w + kk]; + let b = v[nn * w + kk - 1]; + let c = v[(nn - 1) * w + kk - 1]; + v[nn * w + kk] = a.saturating_add(b).saturating_add(c); + } + } + let total = v[n * w + k]; + let vv = |nn: usize, kx: usize| v[nn * w + kx]; + let mut index = 0u64; + let mut kk = k; + for j in 0..n { + if kk == 0 { + break; + } + let rem = n - j; + let t = y[j].unsigned_abs() as usize; + let mut base: u64 = (0..kk - t).map(|m| vv(rem - 1, m)).sum(); + if y[j] < 0 { + base += (vv(rem - 1, kk) + vv(rem, kk)) / 2; + } + index += base; + kk -= t; + } + if total >= 2 { + enc.uint(index as u32, total as u32); + } + } + + #[test] + fn pvq_table_encode_matches_per_call_table() { + let mut seed = 0x9E37_79B9_7F4A_7C15u64; + let mut next = move |m: u64| { + seed ^= seed << 13; + seed ^= seed >> 7; + seed ^= seed << 17; + seed % m + }; + let mut checked = 0; + for n in (1..=PVQ_MAX_N).step_by(5).chain([PVQ_MAX_N]) { + for k in (1..=PVQ_MAX_K).step_by(3).chain([PVQ_MAX_K]) { + if crate::opus::encoder::rate::pvq_v(n, k) > u32::MAX as u128 { + continue; + } + for _ in 0..4 { + let mut y = vec![0i32; n]; + for _ in 0..k { + let at = next(n as u64) as usize; + y[at] += if next(2) == 0 { 1 } else { -1 }; + } + // Opposite pulses cancel; top up so that sum |y| = k again. + let mut have: usize = y.iter().map(|v| v.unsigned_abs() as usize).sum(); + while have < k { + let at = next(n as u64) as usize; + y[at] += if y[at] < 0 { -1 } else { 1 }; + have += 1; + } + let (mut a, mut b) = (RangeEncoder::new(), RangeEncoder::new()); + encode_pulses(&y, k, &mut a); + encode_pulses_reference(&y, k, &mut b); + assert_eq!(a.done(16), b.done(16), "n={n} k={k}"); + checked += 1; + } + } + } + assert!(checked > 100, "{checked}"); + } + + /// The scalar greedy search the vectorised one replaced. + fn pvq_search_ref(x: &[f32], k: usize, y: &mut [i32]) { + let n = x.len(); + y[..n].fill(0); + let abs: Vec = x.iter().map(|v| v.abs()).collect(); + let sum: f32 = abs.iter().sum(); + if !sum.is_finite() || sum <= 1e-15 { + y[0] = k as i32; + return; + } + let mut placed = 0usize; + let mut xy = 0f32; + let mut yy = 0f32; + if k > n / 2 { + let r = (k as f32 - 1.0) / sum; + for j in 0..n { + let p = (abs[j] * r).floor() as i32; + y[j] = p; + placed += p as usize; + xy += abs[j] * p as f32; + yy += (p * p) as f32; + } + } + while placed < k { + let mut best = 0usize; + let mut best_num = -1f32; + let mut best_den = 1f32; + for j in 0..n { + let num = xy + abs[j]; + let num = num * num; + let den = yy + 2.0 * y[j] as f32 + 1.0; + if num * best_den > best_num * den { + best_num = num; + best_den = den; + best = j; + } + } + xy += abs[best]; + yy += 2.0 * y[best] as f32 + 1.0; + y[best] += 1; + placed += 1; + } + for j in 0..n { + if x[j] < 0.0 { + y[j] = -y[j]; + } + } + } + + #[test] + fn pvq_search_matches_scalar_reference() { + let mut state = 0x9e3779b97f4a7c15u64; + let mut rnd = move || { + state ^= state << 13; + state ^= state >> 7; + state ^= state << 17; + (state >> 40) as f32 / (1u64 << 24) as f32 * 2.0 - 1.0 + }; + let mut a = vec![0i32; 200]; + let mut b = vec![0i32; 200]; + let mut cases = 0; + for round in 0..6000 { + let n = 1 + (round * 37 + (rnd().abs() * 176.0) as usize) % 176; + let k = 1 + (rnd().abs() * 127.0) as usize; + let scale = [1.0f32, 1e-9, 1e6, 3e-3][round % 4]; + let x: Vec = (0..n) + .map(|_| match round % 6 { + 0 => rnd(), + 1 => if rnd() > 0.8 { rnd() } else { 0.0 }, + 2 => (rnd() * 4.0).round() / 4.0, + 3 => if rnd() > 0.0 { 1.0 } else { -1.0 }, + 4 => rnd().powi(5), + _ => if rnd() > 0.95 { 3.0 * rnd() } else { 1e-4 * rnd() }, + } * scale) + .collect(); + pvq_search(&x, k, &mut a); + pvq_search_ref(&x, k, &mut b); + assert_eq!(a[..n], b[..n], "n={n} k={k} round={round}"); + cases += 1; + } + assert_eq!(cases, 6000); + } + + #[test] + fn pvq_search_finds_good_direction() { + let x: Vec = (0..32).map(|i| ((i as f32) * 0.7).sin()).collect(); + let mut y = vec![0i32; 32]; + for k in [1usize, 4, 16, 64] { + pvq_search(&x, k, &mut y); + assert_eq!(y.iter().map(|v| v.unsigned_abs() as usize).sum::(), k); + let xy: f32 = x.iter().zip(&y).map(|(a, &b)| a * b as f32).sum(); + let yy: f32 = y.iter().map(|&b| (b * b) as f32).sum(); + let xx: f32 = x.iter().map(|a| a * a).sum(); + let corr = xy / (xx * yy).sqrt(); + assert!(corr > if k >= 16 { 0.9 } else { 0.2 }, "k={k} corr={corr}"); + } + } +} diff --git a/crates/audio/src/opus/encoder/energy.rs b/crates/audio/src/opus/encoder/energy.rs new file mode 100644 index 000000000..fb5cc5a4b --- /dev/null +++ b/crates/audio/src/opus/encoder/energy.rs @@ -0,0 +1,225 @@ +//! Band energy quantisation: Laplace-coded coarse energy with inter/intra prediction, fine +//! energy and the final refinement bits (RFC 6716 §4.3.2, encoder direction). +//! +//! The quantised energies are updated with exactly the decoder's arithmetic so the prediction +//! state never drifts between encoder and decoder. + +use crate::opus::encoder::CeltFrame; +use crate::opus::encoder::range::RangeEncoder; +use crate::opus::encoder::rate::{Allocation, MAX_FINE_BITS}; +use crate::opus::encoder::tables::{BETA_COEF, BETA_INTRA, E_PROB_MODEL, NB_EBANDS, PRED_COEF, SMALL_ENERGY_ICDF}; + +const LAPLACE_MINP: u32 = 1; +const LAPLACE_NMIN: u32 = 16; + +fn laplace_get_freq1(fs0: u32, decay: u32) -> u32 { + let ft = 32768 - LAPLACE_MINP * (2 * LAPLACE_NMIN) - fs0; + (ft * (16384 - decay)) >> 15 +} + +/// Encodes a Laplace-distributed integer (`fs` = probability of 0 in Q15, `decay` in Q14). +/// Returns the value actually coded (very large magnitudes are clamped to the representable +/// range). +pub fn laplace_encode(enc: &mut RangeEncoder, value: i32, mut fs: u32, decay: u32) -> i32 { + let mut fl = 0u32; + let mut coded = value; + if value != 0 { + let s: i32 = if value < 0 { -1 } else { 0 }; + let val = value.abs(); + fl = fs; + fs = laplace_get_freq1(fs, decay); + let mut i = 1; + while fs > 0 && i < val { + fs *= 2; + fl += fs + 2 * LAPLACE_MINP; + fs = (fs * decay) >> 15; + i += 1; + } + if fs == 0 { + let ndi_max = (32768 - fl + LAPLACE_MINP - 1) as i32; + let ndi_max = (ndi_max - s) >> 1; + let di = (val - i).min(ndi_max - 1); + fl = (fl as i32 + (2 * di + 1 + s) * LAPLACE_MINP as i32) as u32; + fs = LAPLACE_MINP.min(32768 - fl); + let mag = i + di; + coded = if s < 0 { -mag } else { mag }; + } else { + fs += LAPLACE_MINP; + if s == 0 { + fl += fs; + } + } + } + enc.encode_bin(fl, fl + fs, 15); + coded +} + +/// `quant_coarse_energy` for one prediction mode. `e` holds the target log2 energies (relative +/// to `E_MEANS`), `old_e` the quantised energies, updated in place. `budget` is the frame size in +/// bits. +pub fn quant_coarse_energy( + frame: CeltFrame, + e: &[f32; 2 * NB_EBANDS], + old_e: &mut [f32; 2 * NB_EBANDS], + intra: bool, + enc: &mut RangeEncoder, + budget: i32, +) { + let CeltFrame { start, end, channels: c, lm } = frame; + let prob = &E_PROB_MODEL[lm][intra as usize]; + let (coef, beta) = if intra { (0.0, BETA_INTRA) } else { (PRED_COEF[lm], BETA_COEF[lm]) }; + let mut prev = [0f32; 2]; + for i in start..end { + for ch in 0..c { + let idx = i + ch * NB_EBANDS; + let oldc = old_e[idx].max(-9.0); + let x = e[idx]; + let f = x - coef * oldc - prev[ch]; + let mut qi = (f + 0.5).floor().clamp(-64.0, 64.0) as i32; + let tell = enc.tell(); + // Keep a few bits per remaining band in reserve when the budget runs out. + let bits_left = budget - tell - 3 * c as i32 * (end - i) as i32; + if i != start && bits_left < 30 { + if bits_left < 24 { + qi = qi.min(1); + } + if bits_left < 16 { + qi = qi.max(-1); + } + } + if budget - tell >= 15 { + let pi = 2 * i.min(20); + qi = laplace_encode(enc, qi, (prob[pi] as u32) << 7, (prob[pi + 1] as u32) << 6); + } else if budget - tell >= 2 { + qi = qi.clamp(-1, 1); + enc.icdf(if qi < 0 { (-2 * qi - 1) as usize } else { (2 * qi) as usize }, &SMALL_ENERGY_ICDF, 2); + } else if budget - tell >= 1 { + qi = qi.clamp(-1, 0); + enc.bit_logp(qi == -1, 1); + } else { + qi = -1; + } + let q = qi as f32; + old_e[idx] = oldc; + let tmp = coef * old_e[idx] + prev[ch] + q; + old_e[idx] = tmp; + prev[ch] = prev[ch] + q - beta * q; + } + } +} + +/// `quant_fine_energy`: raw bits refining each band's quantised energy. +pub fn quant_fine_energy( + frame: CeltFrame, + e: &[f32; 2 * NB_EBANDS], + old_e: &mut [f32; 2 * NB_EBANDS], + fine_quant: &[i32; NB_EBANDS], + enc: &mut RangeEncoder, +) { + let CeltFrame { start, end, channels: c, .. } = frame; + for i in start..end { + let fq = fine_quant[i]; + if fq <= 0 { + continue; + } + for ch in 0..c { + let idx = i + ch * NB_EBANDS; + let err = e[idx] - old_e[idx]; + let levels = 1i32 << fq; + let q2 = (((err + 0.5) * levels as f32).floor() as i32).clamp(0, levels - 1); + enc.bits(q2 as u32, fq as u32); + let offset = (q2 as f32 + 0.5) * (1 << (14 - fq)) as f32 / 16384.0 - 0.5; + old_e[idx] += offset; + } + } +} + +/// `quant_energy_finalise`: spends the bits left at the end of the frame on one more bit of +/// energy resolution per band, in priority order. +pub fn quant_energy_finalise( + frame: CeltFrame, + e: &[f32; 2 * NB_EBANDS], + old_e: &mut [f32; 2 * NB_EBANDS], + allocation: &Allocation, + mut bits_left: i32, + enc: &mut RangeEncoder, +) { + let CeltFrame { start, end, channels: c, .. } = frame; + let Allocation { fine_quant, fine_priority, .. } = allocation; + for prio in 0..2 { + let mut i = start; + while i < end && bits_left >= c as i32 { + if fine_quant[i] >= MAX_FINE_BITS || fine_priority[i] != prio { + i += 1; + continue; + } + for ch in 0..c { + let idx = i + ch * NB_EBANDS; + let q2 = (e[idx] - old_e[idx] >= 0.0) as u32; + enc.bits(q2, 1); + let offset = (q2 as f32 - 0.5) * (1 << (14 - fine_quant[i] - 1)) as f32 / 16384.0; + old_e[idx] += offset; + bits_left -= 1; + } + i += 1; + } + } +} + +#[cfg(test)] +mod tests { + use super::*; + use crate::opus::encoder::range::dec::RangeDecoder; + + /// Laplace decoder of RFC 6716 §4.3.2.1 (as in FilmCraft's decoder). + fn laplace_decode(dec: &mut RangeDecoder, mut fs: u32, decay: u32) -> i32 { + let mut val = 0i32; + let fm = dec.decode(32768); + let mut fl = 0u32; + if fm >= fs { + val += 1; + fl = fs; + fs = laplace_get_freq1(fs, decay) + LAPLACE_MINP; + while fs > LAPLACE_MINP && fm >= fl + 2 * fs { + fs *= 2; + fl += fs; + fs = (((fs - 2 * LAPLACE_MINP) * decay) >> 15) + LAPLACE_MINP; + val += 1; + } + if fs <= LAPLACE_MINP { + let di = (fm - fl) >> 1; + val += di as i32; + fl += 2 * di * LAPLACE_MINP; + } + if fm < fl + fs { + val = -val; + } else { + fl += fs; + } + } + dec.update(fl, (fl + fs).min(32768), 32768); + val + } + + #[test] + fn laplace_roundtrip_all_models() { + for lm in 0..4 { + for intra in 0..2 { + for band in 0..21 { + let prob = &E_PROB_MODEL[lm][intra]; + let (fs, decay) = ((prob[2 * band] as u32) << 7, (prob[2 * band + 1] as u32) << 6); + let vals: Vec = (-64..=64).collect(); + let mut e = RangeEncoder::new(); + let coded: Vec = vals.iter().map(|&v| laplace_encode(&mut e, v, fs, decay)).collect(); + let size = (e.tell() as usize).div_ceil(8); + let buf = e.done(size); + let mut d = RangeDecoder::new(&buf); + for (&v, &c) in vals.iter().zip(&coded) { + assert!(c == v || (c.signum() == v.signum() && c.abs() < v.abs()), "clamp {v} -> {c}"); + assert_eq!(laplace_decode(&mut d, fs, decay), c, "lm={lm} intra={intra} band={band} v={v}"); + } + } + } + } + } +} diff --git a/crates/audio/src/opus/encoder/mdct.rs b/crates/audio/src/opus/encoder/mdct.rs new file mode 100644 index 000000000..7cac21d93 --- /dev/null +++ b/crates/audio/src/opus/encoder/mdct.rs @@ -0,0 +1,242 @@ +//! Forward MDCT with the CELT low-overlap window (RFC 6716 §4.3.7, run in the analysis +//! direction). +//! +//! A 20 ms frame is a 1920-point MDCT (960 coefficients) whose window is zero for 420 samples, +//! rises over the 120-sample overlap, is flat for 840 samples, falls over the next 120 samples +//! and is zero again for the last 420. The input block is therefore the 120 overlap samples kept +//! from the previous frame followed by the 960 new samples. The MDCT is folded into a DCT-IV, +//! which is computed with an N/2-point complex FFT. + +use crate::opus::encoder::tables::OVERLAP; + +use crate::opus::fft::{Cpx, Fft}; + +/// Forward MDCT producing `n` coefficients from a block of `n + OVERLAP` samples. +pub struct Mdct { + /// Number of coefficients (half the MDCT length). + n: usize, + fft: Fft, + pre: Vec, + post: Vec, + /// Rising half of the low-overlap window (RFC 6716 §4.3.7). + pub window: [f32; OVERLAP], + scale: f32, + z: Vec, + scratch: Vec, + v: Vec, +} + +impl Mdct { + pub fn new(n: usize) -> Mdct { + let n2 = n / 2; + let pi = std::f64::consts::PI; + let pre = (0..n2) + .map(|m| { + let ph = -pi * m as f64 / n as f64; + Cpx { re: ph.cos() as f32, im: ph.sin() as f32 } + }) + .collect(); + let post = (0..n2) + .map(|p| { + let ph = -pi * (p as f64 + 0.25) / n as f64; + Cpx { re: ph.cos() as f32, im: ph.sin() as f32 } + }) + .collect(); + let mut window = [0f32; OVERLAP]; + for (i, w) in window.iter_mut().enumerate() { + let x = std::f64::consts::FRAC_PI_2 * (i as f64 + 0.5) / OVERLAP as f64; + let s = x.sin(); + *w = (std::f64::consts::FRAC_PI_2 * s * s).sin() as f32; + } + Mdct { n, fft: Fft::new(n2), pre, post, window, scale: 2.0 / n as f32, z: Vec::new(), scratch: Vec::new(), v: vec![0.0; n] } + } + + /// Transforms `block` (`n + OVERLAP` samples: the previous frame's last `OVERLAP` samples, then + /// `n` new ones) into `out[..n]`, scaled so that the RFC 6716 decoder's inverse MDCT and + /// overlap-add reproduce the input. + pub fn forward(&mut self, block: &[f32], out: &mut [f32]) { + let n = self.n; + let n2 = n / 2; + debug_assert_eq!(block.len(), n + OVERLAP); + // Place the windowed block into the (virtual) 2n-sample MDCT input u at offset `pad`. + let pad = (n - OVERLAP) / 2; + let w = &self.window; + let u = |t: usize| -> f32 { + if t < pad || t >= pad + n + OVERLAP { + return 0.0; + } + let j = t - pad; + let g = if j < OVERLAP { + w[j] + } else if j >= n { + w[OVERLAP - 1 - (j - n)] + } else { + 1.0 + }; + block[j] * g + }; + // Fold to a DCT-IV input: v = (-c_r - d, a - b_r) over the four quarters of u. + for m in 0..n2 { + self.v[m] = -u(n + n2 - 1 - m) - u(n + n2 + m); + self.v[n2 + m] = u(m) - u(n2 + n2 - 1 - m); + } + // DCT-IV via an n/2-point complex FFT. + self.z.clear(); + for m in 0..n2 { + let wv = Cpx { re: self.v[2 * m], im: self.v[n - 1 - 2 * m] }; + self.z.push(wv.mul(self.pre[m])); + } + self.fft.process(&mut self.z, &mut self.scratch); + for p in 0..n2 { + let y = self.z[p].mul(self.post[p]); + out[2 * p] = y.re * self.scale; + out[n - 1 - 2 * p] = -y.im * self.scale; + } + } +} + +#[cfg(test)] +mod tests { + use super::*; + + /// Direct MDCT definition X[k] = sum u[t] cos(pi/N (t + 1/2 + N/2)(k + 1/2)). + fn direct(block: &[f32], n: usize, window: &[f32; OVERLAP]) -> Vec { + let pad = (n - OVERLAP) / 2; + let mut u = vec![0f64; 2 * n]; + for (j, &b) in block.iter().enumerate() { + let g = if j < OVERLAP { + window[j] + } else if j >= n { + window[OVERLAP - 1 - (j - n)] + } else { + 1.0 + }; + u[pad + j] = (b * g) as f64; + } + (0..n) + .map(|k| { + (0..2 * n).map(|t| u[t] * (std::f64::consts::PI / n as f64 * (t as f64 + 0.5 + n as f64 / 2.0) * (k as f64 + 0.5)).cos()).sum::() + * 2.0 + / n as f64 + }) + .collect() + } + + #[test] + fn fft_matches_dft() { + // Every radix combination, including the 480-point size of the 20 ms frame. + for &n in &[2usize, 3, 4, 5, 6, 8, 12, 15, 16, 20, 24, 30, 48, 60, 120, 240, 480] { + let fft = Fft::new(n); + let x: Vec = (0..n).map(|i| Cpx { re: ((i * 37 % 101) as f32 - 50.0) / 7.0, im: ((i * 53 % 89) as f32 - 44.0) / 9.0 }).collect(); + let mut y = x.clone(); + fft.process(&mut y, &mut Vec::new()); + for k in 0..n { + let (mut re, mut im) = (0f64, 0f64); + for (j, v) in x.iter().enumerate() { + let ph = -2.0 * std::f64::consts::PI * ((j * k) % n) as f64 / n as f64; + re += v.re as f64 * ph.cos() - v.im as f64 * ph.sin(); + im += v.re as f64 * ph.sin() + v.im as f64 * ph.cos(); + } + assert!( + (y[k].re as f64 - re).abs() < 2e-3 && (y[k].im as f64 - im).abs() < 2e-3, + "n={n} k={k}: ({}, {}) vs ({re}, {im})", + y[k].re, + y[k].im + ); + } + } + } + + #[test] + fn matches_direct_formula() { + for &n in &[240usize, 960] { + let block: Vec = (0..n + OVERLAP).map(|i| ((i * 37 % 101) as f32 - 50.0) / 7.0).collect(); + let mut m = Mdct::new(n); + let mut out = vec![0f32; n]; + m.forward(&block, &mut out); + let d = direct(&block, n, &m.window); + for k in 0..n { + assert!((out[k] as f64 - d[k]).abs() < 1e-2, "n={n} k={k}: {} vs {}", out[k], d[k]); + } + } + } + + // ---- Replica of the RFC 6716 decoder inverse MDCT (as in FilmCraft's decoder) ---- + + fn imdct_backward(n_full: usize, input: &[f32], out: &mut [f32], window: &[f32]) { + let n2 = n_full >> 1; + let n4 = n_full >> 2; + let t: Vec = (0..n2).map(|i| (2.0 * std::f64::consts::PI * (i as f64 + 0.125) / n_full as f64).cos() as f32).collect(); + let fft = Fft::new(n4); + let mut z = Vec::new(); + for i in 0..n4 { + let x1 = input[2 * i]; + let x2 = input[n2 - 1 - 2 * i]; + let yr = x2 * t[i] + x1 * t[n4 + i]; + let yi = x1 * t[i] - x2 * t[n4 + i]; + z.push(Cpx { re: yi, im: yr }); + } + fft.process(&mut z, &mut Vec::new()); + let overlap = OVERLAP; + let base = overlap >> 1; + let y = &mut out[base..base + n2]; + for i in 0..n4.div_ceil(2) { + let front = z[i]; + let back = z[n4 - 1 - i]; + let (re, im) = (front.im, front.re); + let (t0, t1) = (t[i], t[n4 + i]); + let yr0 = re * t0 + im * t1; + let yi0 = re * t1 - im * t0; + let (re, im) = (back.im, back.re); + let (t0, t1) = (t[n4 - i - 1], t[n2 - i - 1]); + let yr1 = re * t0 + im * t1; + let yi1 = re * t1 - im * t0; + y[2 * i] = yr0; + y[n2 - 1 - 2 * i] = yi0; + y[n2 - 2 - 2 * i] = yr1; + y[2 * i + 1] = yi1; + } + for i in 0..overlap / 2 { + let x1 = out[overlap - 1 - i]; + let x2 = out[i]; + let w1 = window[i]; + let w2 = window[overlap - 1 - i]; + out[i] = x2 * w2 - x1 * w1; + out[overlap - 1 - i] = x2 * w1 + x1 * w2; + } + } + + /// MDCT analysis followed by the decoder's IMDCT + TDAC overlap-add reproduces the input + /// delayed by exactly `OVERLAP` samples. + #[test] + fn perfect_reconstruction_through_decoder_imdct() { + let n = 960; + let frames = 5; + let sig: Vec = (0..n * frames).map(|i| (i as f32 * 0.013).sin() * 1000.0 + ((i * 7919) % 211) as f32 - 105.0).collect(); + let mut m = Mdct::new(n); + let mut hist = vec![0f32; OVERLAP]; + // Decoder memory: [frame output n][folded tail OVERLAP/2] laid out like decode_mem. + let mut mem = vec![0f32; n + OVERLAP]; + let mut decoded = Vec::new(); + for f in 0..frames { + let mut block = hist.clone(); + block.extend_from_slice(&sig[f * n..(f + 1) * n]); + hist.copy_from_slice(&block[n..]); + let mut coefs = vec![0f32; n]; + m.forward(&block, &mut coefs); + // Shift: the previous folded tail (at n..n+OVERLAP/2) moves to 0..OVERLAP/2. + mem.copy_within(n..n + OVERLAP / 2, 0); + imdct_backward(2 * n, &coefs, &mut mem, &m.window); + decoded.extend_from_slice(&mem[..n]); + } + let mut err = 0f64; + let mut sum = 0f64; + for t in OVERLAP..n * frames { + let d = decoded[t] as f64 - sig[t - OVERLAP] as f64; + err += d * d; + sum += (sig[t - OVERLAP] as f64).powi(2); + } + let snr = 10.0 * (sum / err).log10(); + assert!(snr > 80.0, "reconstruction SNR {snr:.1} dB"); + } +} diff --git a/crates/audio/src/opus/encoder/mod.rs b/crates/audio/src/opus/encoder/mod.rs new file mode 100644 index 000000000..8a3df07f1 --- /dev/null +++ b/crates/audio/src/opus/encoder/mod.rs @@ -0,0 +1,596 @@ +//! Clean-room pure-Rust Opus encoder (RFC 6716 SILK, hybrid and CELT modes; RFC 7845 `OpusHead`). +//! +//! Produces 20 ms Opus packets from 48 kHz float input, mono or stereo. The coding mode and +//! audio bandwidth are chosen once per stream from the per-channel bitrate and the +//! [`Application`] (see the crate README for the table): +//! +//! - **SILK** (narrowband/mediumband/wideband): linear-predictive speech coding at an internal +//! rate of 8/12/16 kHz; packet sizes vary around the target bitrate. +//! - **Hybrid** (super-wideband/fullband): SILK wideband codes 0–8 kHz and CELT codes the bands +//! above in the same range-coded frame; constant packet size. +//! - **CELT** (fullband): MDCT transform coding; constant packet size. +//! +//! From EffectCraft's `effectcraft-opusenc` (storytold/effectcraft@c618429), with its later speed-ups +//! and CELT stage refactor up to storytold/effectcraft@d6c9ad66. Public items keep their names, so +//! EffectCraft can re-export this module. +//! +//! ``` +//! # #[cfg(feature = "opus-encode")] { +//! use craft_audio::opus::encoder::{Application, Mode, Encoder}; +//! let mut enc = Encoder::new(2, 128_000).unwrap(); +//! let pcm = vec![0.0f32; Encoder::FRAME_SIZE * 2]; +//! let packet = enc.encode_float(&pcm).unwrap(); +//! assert_eq!(packet[0] >> 3, 31); // CELT-only, fullband, 20 ms +//! assert_eq!(enc.opus_head().len(), 19); +//! let voip = Encoder::with_application(1, 16_000, Application::Voip).unwrap(); +//! assert_eq!(voip.mode(), Mode::Silk); +//! # } +//! ``` + +#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] +// The bit-allocation code mirrors the normative integer arithmetic; keep C-like shift expressions. +#![allow(clippy::precedence, clippy::int_plus_one)] + +mod bands; +mod energy; +mod mdct; +mod range; +mod rate; +mod resample; +mod silk; +pub(super) mod tables; + +use std::sync::OnceLock; + +use bands::{BandCoding, SPREAD_NORMAL, quant_all_bands}; +use energy::{quant_coarse_energy, quant_energy_finalise, quant_fine_energy}; +use mdct::Mdct; +use range::{BITRES, RangeEncoder}; +use rate::{AllocationInput, Mode as CeltMode, compute_allocation}; +use resample::Downsampler; +use silk::SilkEncoder; +use tables::{E_MEANS, EBANDS, NB_EBANDS, OVERLAP, PREEMPH, SPREAD_ICDF, TF_SELECT_TABLE, TRIM_ICDF}; + +/// Band range, channel count and transform size shared by CELT coding stages. +#[derive(Clone, Copy)] +struct CeltFrame { + start: usize, + end: usize, + channels: usize, + lm: usize, +} + +/// Samples per channel in one 20 ms frame at 48 kHz. +const N: usize = 960; +/// LM of a 20 ms frame (8 short blocks). +const LM: usize = 3; +/// Smallest CELT payload the encoder produces (bytes). +const MIN_FRAME_BYTES: usize = 16; +/// Largest payload allowed in one frame (RFC 6716 §3.4, R2). +const MAX_FRAME_BYTES: usize = 1275; +/// First CELT band coded in hybrid frames (8 kHz). +const HYBRID_START_BAND: usize = 17; + +fn celt_mode() -> &'static CeltMode { + static MODE: OnceLock = OnceLock::new(); + MODE.get_or_init(CeltMode::new) +} + +/// Why an encoder could not be created or a frame could not be encoded. +#[non_exhaustive] +#[derive(Clone, Copy, Debug, PartialEq, Eq, thiserror::Error)] +pub enum Error { + /// Opus family-0 streams have 1 or 2 channels. + #[error("Opus family-0 streams have 1 or 2 channels, not {0}")] + Channels(u8), + /// The mode cannot code that bandwidth. + #[error("unsupported mode/bandwidth combination {0:?}/{1:?}")] + ModeBandwidth(Mode, Bandwidth), + /// `encode_float` takes exactly `FRAME_SIZE * channels` samples. + #[error("a frame is {expected} interleaved samples, not {got}")] + FrameLength { + /// Interleaved samples a frame takes (`FRAME_SIZE * channels`). + expected: usize, + /// Interleaved samples passed. + got: usize, + }, +} + +/// Intended use of the stream; selects the mode/bandwidth table. +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, Hash)] +pub enum Application { + /// General audio (music, mixed content): CELT unless the bitrate is low. + #[default] + Audio, + /// Speech: SILK and hybrid over a wider bitrate range, narrower bandwidths at low rates. + Voip, +} + +/// Opus coding mode (RFC 6716 §2). +#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)] +pub enum Mode { + /// MDCT transform coding only. + Celt, + /// Linear-prediction coding only. + Silk, + /// SILK below 8 kHz plus CELT above. + Hybrid, +} + +pub use crate::opus::Bandwidth; + +/// Mode and bandwidth for a per-channel bitrate (bits per second per channel). +fn choose_mode(app: Application, per_channel: u32) -> (Mode, Bandwidth) { + match app { + Application::Voip => match per_channel { + 0..10_000 => (Mode::Silk, Bandwidth::Narrowband), + 10_000..12_000 => (Mode::Silk, Bandwidth::Mediumband), + 12_000..18_000 => (Mode::Silk, Bandwidth::Wideband), + 18_000..28_000 => (Mode::Hybrid, Bandwidth::SuperWideband), + 28_000..40_000 => (Mode::Hybrid, Bandwidth::Fullband), + _ => (Mode::Celt, Bandwidth::Fullband), + }, + Application::Audio => match per_channel { + 0..16_000 => (Mode::Silk, Bandwidth::Wideband), + 16_000..36_000 => (Mode::Hybrid, Bandwidth::Fullband), + _ => (Mode::Celt, Bandwidth::Fullband), + }, + } +} + +/// SILK side of a SILK-only or hybrid stream. +struct SilkPath { + enc: SilkEncoder, + down: Vec, + /// Bits per frame aimed at (excluding the TOC byte in SILK-only mode). + target_bits: f64, + /// Bits saved (positive) or overspent by previous frames (SILK-only mode). + reservoir: f64, +} + +/// The name EffectCraft's `effectcraft-opusenc` used, kept so it can re-export this module. +#[doc(hidden)] +pub type OpusEncoder = Encoder; + +/// Opus encoder for one mono or stereo stream at 48 kHz. +pub struct Encoder { + channels: usize, + application: Application, + mode: Mode, + bandwidth: Bandwidth, + /// Requested bitrate (clamped), bits per second. + bitrate_bps: u32, + /// Payload bytes per packet in CELT and hybrid modes (the packet is one byte longer: the + /// TOC). + frame_bytes: usize, + mdct: Mdct, + /// Last input sample of each channel (pre-emphasis filter memory, Ă—32768). + preemph_mem: [f32; 2], + /// Last `OVERLAP` pre-emphasised samples of each channel. + hist: Vec, + /// Quantised band energies, exactly as the decoder holds them. + old_band_e: [f32; 2 * NB_EBANDS], + block: Vec, + freq: Vec, + silk: Option, +} + +impl Encoder { + /// Samples per channel per packet (20 ms at 48 kHz). + pub const FRAME_SIZE: usize = N; + + /// Creates a general-audio encoder ([`Application::Audio`]) for `channels` (1 or 2) + /// channels of 48 kHz audio at `bitrate_bps` (total for all channels). + pub fn new(channels: u8, bitrate_bps: u32) -> Result { + Self::with_application(channels, bitrate_bps, Application::Audio) + } + + /// Creates an encoder for `channels` (1 or 2) channels of 48 kHz audio at `bitrate_bps` + /// (total for all channels; clamped to 6 kb/s per channel – 510 kb/s), choosing the mode + /// and bandwidth for `application`. + pub fn with_application(channels: u8, bitrate_bps: u32, application: Application) -> Result { + // Checked before the clamp below, whose lower bound passes its upper one from 86 channels. + if channels != 1 && channels != 2 { + return Err(Error::Channels(channels)); + } + let c = channels as u32; + let (mode, bandwidth) = choose_mode(application, bitrate_bps.clamp(6_000 * c, 510_000) / c); + Self::build(channels, bitrate_bps, application, mode, bandwidth) + } + + /// Creates an encoder with an explicit mode and bandwidth instead of the application's + /// table. Valid combinations: SILK with narrowband, mediumband or wideband; hybrid with + /// super-wideband or fullband; CELT with fullband. + /// + /// Any other combination is [`Error::ModeBandwidth`]; `channels` other than 1 or 2 is + /// [`Error::Channels`]. + pub fn with_mode(channels: u8, bitrate_bps: u32, mode: Mode, bandwidth: Bandwidth) -> Result { + let ok = match mode { + Mode::Silk => bandwidth <= Bandwidth::Wideband, + Mode::Hybrid => bandwidth >= Bandwidth::SuperWideband, + Mode::Celt => bandwidth == Bandwidth::Fullband, + }; + if !ok { + return Err(Error::ModeBandwidth(mode, bandwidth)); + } + Self::build(channels, bitrate_bps, Application::Audio, mode, bandwidth) + } + + fn build(channels: u8, bitrate_bps: u32, application: Application, mode: Mode, bandwidth: Bandwidth) -> Result { + if channels != 1 && channels != 2 { + return Err(Error::Channels(channels)); + } + let c = channels as usize; + let bitrate_bps = bitrate_bps.clamp(6_000 * c as u32, 510_000); + let packet_bytes = (bitrate_bps as usize).div_ceil(400); + let frame_bytes = packet_bytes.saturating_sub(1).clamp(MIN_FRAME_BYTES, MAX_FRAME_BYTES); + let silk = match mode { + Mode::Celt => None, + Mode::Silk | Mode::Hybrid => { + let fs_khz = match bandwidth { + Bandwidth::Narrowband => 8, + Bandwidth::Mediumband => 12, + _ => 16, + }; + let target_bits = if mode == Mode::Silk { + bitrate_bps as f64 / 50.0 - 8.0 + } else { + // SILK gets most of a hybrid packet; CELT codes 8 kHz and up with the rest. + let per = (bitrate_bps / c as u32) as f64; + let share = if bandwidth == Bandwidth::SuperWideband { 0.72 } else { 0.62 }; + c as f64 * (per * share).min(per - 6_000.0).max(per * 0.5) / 50.0 + }; + Some(SilkPath { + enc: SilkEncoder::new(fs_khz, c), + down: (0..c).map(|_| Downsampler::new(fs_khz as u32 * 1000)).collect(), + target_bits, + reservoir: 0.0, + }) + } + }; + Ok(Encoder { + channels: c, + application, + mode, + bandwidth, + bitrate_bps, + frame_bytes, + mdct: Mdct::new(N), + preemph_mem: [0.0; 2], + hist: vec![0.0; c * OVERLAP], + old_band_e: [0.0; 2 * NB_EBANDS], + block: vec![0.0; N + OVERLAP], + freq: vec![0.0; c * N], + silk, + }) + } + + /// Number of channels. + pub fn channels(&self) -> u8 { + self.channels as u8 + } + + /// The application the encoder was created for. + pub fn application(&self) -> Application { + self.application + } + + /// Coding mode of every packet of the stream. + pub fn mode(&self) -> Mode { + self.mode + } + + /// Coded audio bandwidth of every packet of the stream. + pub fn bandwidth(&self) -> Bandwidth { + self.bandwidth + } + + /// Bitrate in bits per second: the constant packet size Ă— 50 packets per second in CELT + /// and hybrid modes, the (average) target in SILK mode. + pub fn bitrate(&self) -> u32 { + match self.mode { + Mode::Silk => self.bitrate_bps, + _ => ((self.frame_bytes + 1) * 8 * 50) as u32, + } + } + + /// Samples at 48 kHz the decoder must discard from the start of the stream (the `OpusHead` + /// pre-skip): the encoder's total delay. CELT: the 120-sample MDCT overlap. SILK/hybrid: + /// the decimator (83 samples, 88 for narrowband) plus the decoder's SILK delay (one + /// internal-rate sample and the resampler delay of RFC 6716 Table 54), which also rounds + /// to 120 samples. + pub fn pre_skip(&self) -> u16 { + let (fs, table54_ms) = match (self.mode, self.bandwidth) { + (Mode::Celt, _) => return OVERLAP as u16, + (_, Bandwidth::Narrowband) => (8_000, 0.538), + (_, Bandwidth::Mediumband) => (12_000, 0.692), + _ => (16_000, 0.706), + }; + (resample::delay(fs) as f64 + 48_000.0 / fs as f64 + table54_ms * 48.0).round() as u16 + } + + /// The 19-byte `OpusHead` identification header (RFC 7845 §5.1): version 1, channel count, + /// pre-skip, input sample rate 48000, output gain 0, channel mapping family 0. + pub fn opus_head(&self) -> Vec { + let mut h = Vec::with_capacity(19); + h.extend_from_slice(b"OpusHead"); + h.push(1); + h.push(self.channels as u8); + h.extend_from_slice(&self.pre_skip().to_le_bytes()); + h.extend_from_slice(&48_000u32.to_le_bytes()); + h.extend_from_slice(&0i16.to_le_bytes()); + h.push(0); + h + } + + /// TOC byte (RFC 6716 §3.1, Table 2) of a 20 ms single-frame packet. + fn toc(&self) -> u8 { + let config: u8 = match (self.mode, self.bandwidth) { + (Mode::Silk, Bandwidth::Narrowband) => 1, + (Mode::Silk, Bandwidth::Mediumband) => 5, + (Mode::Silk, _) => 9, + (Mode::Hybrid, Bandwidth::SuperWideband) => 13, + (Mode::Hybrid, _) => 15, + (Mode::Celt, _) => 31, + }; + (config << 3) | if self.channels == 2 { 4 } else { 0 } + } + + /// Encodes one 20 ms frame of interleaved f32 samples in [-1, 1] (exactly + /// `FRAME_SIZE * channels` values) and returns one Opus packet, or + /// [`Error::FrameLength`] for any other length. + pub fn encode_float(&mut self, pcm: &[f32]) -> Result, Error> { + let c = self.channels; + if pcm.len() != N * c { + return Err(Error::FrameLength { expected: N * c, got: pcm.len() }); + } + let mut packet = vec![self.toc()]; + match self.mode { + Mode::Celt => { + let (log_e, silence) = self.celt_analysis(pcm); + let len = self.frame_bytes; + packet.extend_from_slice(&self.encode_celt(RangeEncoder::new(), len, 0..NB_EBANDS, &log_e, silence)); + } + Mode::Silk => { + let x = self.silk_input(pcm); + // The SILK state exists whenever the mode uses SILK (set with the mode). + if let Some(sp) = self.silk.as_mut() { + let target = sp.target_bits + 0.25 * sp.reservoir; + let mut enc = RangeEncoder::new(); + sp.enc.encode(&x, target.max(24.0), (MAX_FRAME_BYTES * 8) as i32, &mut enc); + let bytes = (enc.tell() as usize).div_ceil(8).max(2); + let lim = 4.0 * sp.target_bits; + sp.reservoir = (sp.reservoir + sp.target_bits - (bytes * 8) as f64).clamp(-lim, lim); + packet.extend_from_slice(&enc.done(bytes)); + } + } + Mode::Hybrid => { + let x = self.silk_input(pcm); + let len = self.frame_bytes; + let total = (len * 8) as i32; + let celt_min = (total / 5).max(80); + let mut enc = RangeEncoder::new(); + if let Some(sp) = self.silk.as_mut() { + sp.enc.encode(&x, sp.target_bits, total - celt_min, &mut enc); + } + // No redundancy (RFC 6716 §4.5.1); the flag is present when there is room. + if enc.tell() + 17 + 20 <= total { + enc.bit_logp(false, 12); + } + let (log_e, _) = self.celt_analysis(pcm); + let end = if self.bandwidth == Bandwidth::SuperWideband { 19 } else { NB_EBANDS }; + packet.extend_from_slice(&self.encode_celt(enc, len, HYBRID_START_BAND..end, &log_e, false)); + } + } + Ok(packet) + } + + /// Decimates the input to the SILK internal rate (16-bit scale), one vector per channel. + fn silk_input(&mut self, pcm: &[f32]) -> Vec> { + let c = self.channels; + let Some(sp) = self.silk.as_mut() else { return vec![vec![]; c] }; + let mut out = Vec::with_capacity(c); + let mut tmp = vec![0f32; N]; + for ch in 0..c { + for (j, t) in tmp.iter_mut().enumerate() { + let s = pcm[j * c + ch]; + *t = if s.is_finite() { s.clamp(-1.0, 1.0) * 32768.0 } else { 0.0 }; + } + let mut y = Vec::with_capacity(N / 3); + sp.down[ch].process(&tmp, &mut y); + out.push(y); + } + out + } + + /// Pre-emphasis, the MDCT, band energies (log2, relative to `E_MEANS`) and normalised + /// bands; returns the energies and whether the frame is digital silence. + fn celt_analysis(&mut self, pcm: &[f32]) -> ([f32; 2 * NB_EBANDS], bool) { + let c = self.channels; + let mut log_e = [0f32; 2 * NB_EBANDS]; + let mut peak = 0f32; + for ch in 0..c { + self.block[..OVERLAP].copy_from_slice(&self.hist[ch * OVERLAP..(ch + 1) * OVERLAP]); + let mut mem = self.preemph_mem[ch]; + for j in 0..N { + let s = pcm[j * c + ch]; + let s = if s.is_finite() { s.clamp(-2.0, 2.0) * 32768.0 } else { 0.0 }; + let v = s - PREEMPH * mem; + mem = s; + self.block[OVERLAP + j] = v; + } + self.preemph_mem[ch] = mem; + self.hist[ch * OVERLAP..(ch + 1) * OVERLAP].copy_from_slice(&self.block[N..]); + peak = self.block.iter().fold(peak, |p, v| p.max(v.abs())); + let freq = &mut self.freq[ch * N..(ch + 1) * N]; + self.mdct.forward(&self.block, freq); + for i in 0..NB_EBANDS { + let lo = (EBANDS[i] as usize) << LM; + let hi = (EBANDS[i + 1] as usize) << LM; + let e = (1e-27f32 + freq[lo..hi].iter().map(|v| v * v).sum::()).sqrt(); + log_e[ch * NB_EBANDS + i] = e.log2() - E_MEANS[i]; + let g = 1.0 / e; + for v in freq[lo..hi].iter_mut() { + *v *= g; + } + } + } + (log_e, peak < 1e-4) + } + + /// Codes one CELT frame (RFC 6716 §4.3, in bitstream order) of `len` bytes into `enc` + /// (which already holds the SILK frame in hybrid mode), coding bands `start..end`, and + /// finalises the frame. + fn encode_celt( + &mut self, + mut enc: RangeEncoder, + len: usize, + bands: std::ops::Range, + log_e: &[f32; 2 * NB_EBANDS], + silence: bool, + ) -> Vec { + let m = celt_mode(); + let c = self.channels; + let frame = CeltFrame { start: bands.start, end: bands.end, channels: c, lm: LM }; + let CeltFrame { start, end, .. } = frame; + if c == 1 { + for i in 0..NB_EBANDS { + self.old_band_e[i] = self.old_band_e[i].max(self.old_band_e[NB_EBANDS + i]); + } + } + if silence && start == 0 { + // The silence flag; the decoder treats every remaining bit as consumed. + enc.bit_logp(true, 15); + self.old_band_e = [-28.0; 2 * NB_EBANDS]; + return enc.done(2); + } + let total_bits_raw = len as i32 * 8; + if enc.tell() == 1 { + enc.bit_logp(false, 15); + } + // No pitch post-filter (only signalled in CELT-only frames). + if start == 0 && enc.tell() + 16 <= total_bits_raw { + enc.bit_logp(false, 1); + } + // Long blocks only. + if enc.tell() + 3 <= total_bits_raw { + enc.bit_logp(false, 3); + } + // Coarse energy, choosing intra or inter prediction by trial (whichever is cheaper). + if enc.tell() + 3 <= total_bits_raw { + let mut enc_intra = enc.clone(); + let mut old_intra = self.old_band_e; + enc_intra.bit_logp(true, 3); + quant_coarse_energy(frame, log_e, &mut old_intra, true, &mut enc_intra, total_bits_raw); + enc.bit_logp(false, 3); + let mut old_inter = self.old_band_e; + quant_coarse_energy(frame, log_e, &mut old_inter, false, &mut enc, total_bits_raw); + if enc_intra.tell_frac() < enc.tell_frac() { + enc = enc_intra; + self.old_band_e = old_intra; + } else { + self.old_band_e = old_inter; + } + } else { + quant_coarse_energy(frame, log_e, &mut self.old_band_e, false, &mut enc, total_bits_raw); + } + // TF resolution: no changes. + { + let mut budget = total_bits_raw; + let mut tell = enc.tell(); + let mut logp = 4; + let tf_select_rsv = LM > 0 && tell + logp < budget; + budget -= tf_select_rsv as i32; + for _ in start..end { + if tell + logp <= budget { + enc.bit_logp(false, logp as u32); + tell = enc.tell(); + } + logp = 5; + } + if tf_select_rsv && TF_SELECT_TABLE[LM][0] != TF_SELECT_TABLE[LM][2] { + enc.bit_logp(false, 1); + } + } + if enc.tell() + 4 <= total_bits_raw { + enc.icdf(SPREAD_NORMAL, &SPREAD_ICDF, 5); + } + // Dynamic allocation: no boosts. + let cap = m.init_caps(LM, c); + let offsets = [0i32; NB_EBANDS]; + let total_bits = total_bits_raw << BITRES; + let mut tellf = enc.tell_frac(); + for &cp in cap.iter().take(end).skip(start) { + if tellf + (6 << BITRES) < total_bits && 0 < cp { + enc.bit_logp(false, 6); + tellf = enc.tell_frac(); + } + } + let alloc_trim = 5; + if tellf + (6 << BITRES) <= total_bits { + enc.icdf(alloc_trim as usize, &TRIM_ICDF, 7); + } + let bits = (total_bits_raw << BITRES) - enc.tell_frac() - 1; + let alloc = compute_allocation(m, frame, AllocationInput { offsets: &offsets, cap: &cap, trim: alloc_trim, bits }, &mut enc); + quant_fine_energy(frame, log_e, &mut self.old_band_e, &alloc.fine_quant, &mut enc); + quant_all_bands(m, frame, &self.freq, N, &alloc, BandCoding { spread: SPREAD_NORMAL, total_bits: total_bits_raw << BITRES }, &mut enc); + let bits_left = total_bits_raw - enc.tell(); + quant_energy_finalise(frame, log_e, &mut self.old_band_e, &alloc, bits_left, &mut enc); + debug_assert!(enc.tell() <= total_bits_raw, "frame overflow: {} > {}", enc.tell(), total_bits_raw); + if c == 1 { + let (a, b) = self.old_band_e.split_at_mut(NB_EBANDS); + b.copy_from_slice(a); + } + // The decoder forgets the energies of bands it did not code. + for ch in 0..2 { + for i in (0..start).chain(end..NB_EBANDS) { + self.old_band_e[ch * NB_EBANDS + i] = 0.0; + } + } + enc.done(len) + } +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn opus_head_layout() { + let e = Encoder::new(2, 96_000).unwrap(); + let h = e.opus_head(); + assert_eq!(&h[..8], b"OpusHead"); + assert_eq!(h[8], 1); + assert_eq!(h[9], 2); + assert_eq!(u16::from_le_bytes([h[10], h[11]]), e.pre_skip()); + assert_eq!(u32::from_le_bytes([h[12], h[13], h[14], h[15]]), 48_000); + assert_eq!(&h[16..], &[0, 0, 0]); + } + + #[test] + fn packet_sizes_follow_bitrate() { + for (ch, rate) in [(1u8, 64_000u32), (2, 128_000), (2, 256_000)] { + let mut e = Encoder::new(ch, rate).unwrap(); + let pcm: Vec = (0..N * ch as usize).map(|i| (i as f32 * 0.05).sin() * 0.5).collect(); + let p = e.encode_float(&pcm).unwrap(); + assert_eq!(p.len(), rate as usize / 400); + assert_eq!(p[0], 0xF8 | if ch == 2 { 4 } else { 0 }); + assert_eq!(e.bitrate(), rate); + } + } + + #[test] + fn silence_is_a_tiny_packet() { + let mut e = Encoder::new(1, 64_000).unwrap(); + let p = e.encode_float(&vec![0.0; N]).unwrap(); + assert_eq!(p.len(), 3); + } + + #[test] + fn wrong_frame_length_is_an_error() { + let mut e = Encoder::new(2, 128_000).unwrap(); + assert_eq!(e.encode_float(&vec![0.0; N]), Err(Error::FrameLength { expected: 2 * N, got: N })); + assert_eq!(e.encode_float(&[]), Err(Error::FrameLength { expected: 2 * N, got: 0 })); + assert!(e.encode_float(&vec![0.0; 2 * N]).is_ok()); + } +} diff --git a/crates/audio/src/opus/encoder/range.rs b/crates/audio/src/opus/encoder/range.rs new file mode 100644 index 000000000..daedd58ae --- /dev/null +++ b/crates/audio/src/opus/encoder/range.rs @@ -0,0 +1,472 @@ +//! Range encoder (RFC 6716 §5.1), the exact inverse of the §4.1 range decoder. +//! +//! Range-coded symbols are written from the front of the buffer and raw bits from the back; the +//! bit accounting ([`RangeEncoder::tell`], [`RangeEncoder::tell_frac`]) matches the decoder's, so +//! every budget decision the decoder takes can be mirrored exactly. + +const SYM_BITS: u32 = 8; +const CODE_BITS: u32 = 32; +const SYM_MAX: u32 = (1 << SYM_BITS) - 1; +const CODE_SHIFT: u32 = CODE_BITS - SYM_BITS - 1; +const CODE_TOP: u32 = 1 << (CODE_BITS - 1); +const CODE_BOT: u32 = CODE_TOP >> SYM_BITS; +/// Bit resolution of [`RangeEncoder::tell_frac`] (1/8 bit). +pub const BITRES: u32 = 3; + +/// Number of bits needed to represent `x` (0 for 0). +#[inline] +pub fn ilog(x: u32) -> i32 { + (32 - x.leading_zeros()) as i32 +} + +#[derive(Clone, Debug)] +pub struct RangeEncoder { + out: Vec, + end: Vec, + end_window: u32, + nend_bits: i32, + nbits_total: i32, + rng: u32, + low: u32, + rem: i32, + ext: u32, +} + +impl Default for RangeEncoder { + fn default() -> Self { + Self::new() + } +} + +impl RangeEncoder { + pub fn new() -> Self { + RangeEncoder { + out: Vec::new(), + end: Vec::new(), + end_window: 0, + nend_bits: 0, + nbits_total: (CODE_BITS + 1) as i32, + rng: CODE_TOP, + low: 0, + rem: -1, + ext: 0, + } + } + + /// Current range (equals the decoder's range after the same symbols). + #[cfg(test)] + pub fn rng(&self) -> u32 { + self.rng + } + + fn carry_out(&mut self, c: u32) { + if c != SYM_MAX { + let carry = c >> SYM_BITS; + if self.rem >= 0 { + self.out.push((self.rem as u32 + carry) as u8); + } + if self.ext > 0 { + let sym = (SYM_MAX + carry) & SYM_MAX; + for _ in 0..self.ext { + self.out.push(sym as u8); + } + self.ext = 0; + } + self.rem = (c & SYM_MAX) as i32; + } else { + self.ext += 1; + } + } + + #[inline] + fn normalize(&mut self) { + while self.rng <= CODE_BOT { + self.carry_out(self.low >> CODE_SHIFT); + self.low = (self.low << SYM_BITS) & (CODE_TOP - 1); + self.rng <<= SYM_BITS; + self.nbits_total += SYM_BITS as i32; + } + } + + /// Encodes the symbol with cumulative range `[fl, fh)` of total `ft`. + pub fn encode(&mut self, fl: u32, fh: u32, ft: u32) { + let r = self.rng / ft; + if fl > 0 { + self.low += self.rng - r * (ft - fl); + self.rng = r * (fh - fl); + } else { + self.rng -= r * (ft - fh); + } + self.normalize(); + } + + /// Like [`Self::encode`] with `ft = 1 << bits`. + pub fn encode_bin(&mut self, fl: u32, fh: u32, bits: u32) { + let r = self.rng >> bits; + let ft = 1u32 << bits; + if fl > 0 { + self.low += self.rng - r * (ft - fl); + self.rng = r * (fh - fl); + } else { + self.rng -= r * (ft - fh); + } + self.normalize(); + } + + /// Encodes a binary symbol whose probability of being 1 is `1/2^logp`. + pub fn bit_logp(&mut self, val: bool, logp: u32) { + let r = self.rng; + let s = r >> logp; + let r2 = r - s; + if val { + self.low += r2; + self.rng = s; + } else { + self.rng = r2; + } + self.normalize(); + } + + /// Encodes symbol `s` with an "inverse CDF" table (`ft = 1 << ftb`). + pub fn icdf(&mut self, s: usize, icdf: &[u8], ftb: u32) { + let r = self.rng >> ftb; + if s > 0 { + self.low += self.rng - r * icdf[s - 1] as u32; + self.rng = r * (icdf[s - 1] as u32 - icdf[s] as u32); + } else { + self.rng -= r * icdf[s] as u32; + } + self.normalize(); + } + + /// Encodes a uniformly distributed integer `fl` in `[0, ft)` (`ft >= 2`). + pub fn uint(&mut self, fl: u32, ft: u32) { + debug_assert!(ft > 1 && fl < ft); + let ft1 = ft - 1; + let ftb = ilog(ft1); + if ftb > 8 { + let ftb = (ftb - 8) as u32; + let top = (ft1 >> ftb) + 1; + let v = fl >> ftb; + self.encode(v, v + 1, top); + self.bits(fl & ((1 << ftb) - 1), ftb); + } else { + self.encode(fl, fl + 1, ft); + } + } + + /// Writes `bits` raw bits (0..=25) to the end of the buffer. + pub fn bits(&mut self, val: u32, bits: u32) { + if bits == 0 { + return; + } + let mut window = self.end_window; + let mut used = self.nend_bits; + if used as u32 + bits > CODE_BITS { + while used >= SYM_BITS as i32 { + self.end.push((window & SYM_MAX) as u8); + window >>= SYM_BITS; + used -= SYM_BITS as i32; + } + } + window |= val << used; + used += bits as i32; + self.end_window = window; + self.nend_bits = used; + self.nbits_total += bits as i32; + } + + /// Number of bits written so far (rounded up), identical to the decoder's `tell`. + #[inline] + pub fn tell(&self) -> i32 { + self.nbits_total - ilog(self.rng) + } + + /// Number of 1/8 bits written so far (rounded up), identical to the decoder's `tell_frac`. + pub fn tell_frac(&self) -> i32 { + let nbits = (self.nbits_total as u32) << BITRES; + let mut l = ilog(self.rng) as u32; + let mut r = self.rng >> (l - 16); + for _ in 0..BITRES { + r = (r * r) >> 15; + let b = r >> 16; + l = (l << 1) | b; + r >>= b; + } + nbits.wrapping_sub(l) as i32 + } + + /// Finalises the stream into a buffer of exactly `size` bytes (range data at the front, raw + /// bits at the back, zero padding in between; a partial raw-bit byte is merged into the byte + /// just before the full raw-bit bytes). + pub fn done(mut self, size: usize) -> Vec { + let mut l = CODE_BITS as i32 - ilog(self.rng); + let mut msk = (CODE_TOP - 1) >> l; + let mut end = (self.low.wrapping_add(msk)) & !msk; + if (end | msk) >= self.low.wrapping_add(self.rng) { + l += 1; + msk >>= 1; + end = (self.low.wrapping_add(msk)) & !msk; + } + while l > 0 { + self.carry_out(end >> CODE_SHIFT); + end = (end << SYM_BITS) & (CODE_TOP - 1); + l -= SYM_BITS as i32; + } + if self.rem >= 0 || self.ext > 0 { + self.carry_out(0); + } + let mut window = self.end_window; + let mut used = self.nend_bits; + while used >= SYM_BITS as i32 { + self.end.push((window & SYM_MAX) as u8); + window >>= SYM_BITS; + used -= SYM_BITS as i32; + } + let mut buf = vec![0u8; size]; + let full = self.end.len().min(size); + let room = size - full; + let n = self.out.len().min(room); + buf[..n].copy_from_slice(&self.out[..n]); + for (k, &b) in self.end.iter().take(full).enumerate() { + buf[size - 1 - k] = b; + } + if used > 0 && room > 0 { + buf[room - 1] |= window as u8; + } + buf + } +} + +#[cfg(test)] +pub(crate) mod dec { + //! Minimal range decoder (RFC 6716 §4.1) used to verify the encoder. + use super::*; + const CODE_EXTRA: u32 = (CODE_BITS - 2) % SYM_BITS + 1; + + pub struct RangeDecoder<'a> { + buf: &'a [u8], + offs: usize, + end_offs: usize, + end_window: u32, + nend_bits: i32, + nbits_total: i32, + rng: u32, + val: u32, + rem: u32, + } + + impl<'a> RangeDecoder<'a> { + pub fn new(buf: &'a [u8]) -> Self { + let mut d = RangeDecoder { + buf, + offs: 0, + end_offs: 0, + end_window: 0, + nend_bits: 0, + nbits_total: (CODE_BITS + 1 - ((CODE_BITS - CODE_EXTRA) / SYM_BITS) * SYM_BITS) as i32, + rng: 1 << CODE_EXTRA, + val: 0, + rem: 0, + }; + d.rem = d.read_byte(); + d.val = d.rng - 1 - (d.rem >> (SYM_BITS - CODE_EXTRA)); + d.normalize(); + d + } + fn read_byte(&mut self) -> u32 { + if self.offs < self.buf.len() { + self.offs += 1; + self.buf[self.offs - 1] as u32 + } else { + 0 + } + } + fn read_byte_from_end(&mut self) -> u32 { + if self.end_offs < self.buf.len() { + self.end_offs += 1; + self.buf[self.buf.len() - self.end_offs] as u32 + } else { + 0 + } + } + fn normalize(&mut self) { + while self.rng <= CODE_BOT { + self.nbits_total += SYM_BITS as i32; + self.rng <<= SYM_BITS; + let sym = self.rem; + self.rem = self.read_byte(); + let sym = ((sym << SYM_BITS) | self.rem) >> (SYM_BITS - CODE_EXTRA); + self.val = ((self.val << SYM_BITS).wrapping_add(SYM_MAX & !sym)) & (CODE_TOP - 1); + } + } + pub fn rng(&self) -> u32 { + self.rng + } + pub fn decode(&mut self, ft: u32) -> u32 { + let ext = self.rng / ft; + let s = self.val / ext; + ft - (s + 1).min(ft) + } + pub fn update(&mut self, fl: u32, fh: u32, ft: u32) { + let ext = self.rng / ft; + let s = ext * (ft - fh); + self.val = self.val.wrapping_sub(s); + self.rng = if fl > 0 { ext * (fh - fl) } else { self.rng - s }; + self.normalize(); + } + pub fn bit_logp(&mut self, logp: u32) -> bool { + let r = self.rng; + let d = self.val; + let s = r >> logp; + let ret = d < s; + if !ret { + self.val = d - s; + } + self.rng = if ret { s } else { r - s }; + self.normalize(); + ret + } + pub fn icdf(&mut self, icdf: &[u8], ftb: u32) -> usize { + let mut s = self.rng; + let d = self.val; + let r = s >> ftb; + let mut ret = 0usize; + let mut t; + loop { + t = s; + s = r * icdf[ret] as u32; + if d >= s { + break; + } + ret += 1; + } + self.val = d - s; + self.rng = t - s; + self.normalize(); + ret + } + pub fn uint(&mut self, ft: u32) -> u32 { + let ft1 = ft - 1; + let ftb = ilog(ft1); + if ftb > 8 { + let ftb = (ftb - 8) as u32; + let top = (ft1 >> ftb) + 1; + let s = self.decode(top); + self.update(s, s + 1, top); + (s << ftb) | self.bits(ftb) + } else { + let s = self.decode(ft); + self.update(s, s + 1, ft); + s + } + } + pub fn bits(&mut self, bits: u32) -> u32 { + let mut window = self.end_window; + let mut available = self.nend_bits; + if (available as u32) < bits { + loop { + window |= self.read_byte_from_end() << available; + available += SYM_BITS as i32; + if available > (CODE_BITS - SYM_BITS) as i32 { + break; + } + } + } + let ret = window & ((1u32 << bits) - 1); + self.end_window = window >> bits; + self.nend_bits = available - bits as i32; + self.nbits_total += bits as i32; + ret + } + pub fn tell(&self) -> i32 { + self.nbits_total - ilog(self.rng) + } + } +} + +#[cfg(test)] +mod tests { + use super::dec::RangeDecoder; + use super::*; + + struct Lcg(u32); + impl Lcg { + fn next(&mut self) -> u32 { + self.0 = self.0.wrapping_mul(1664525).wrapping_add(1013904223); + self.0 >> 8 + } + } + + #[test] + fn starts_at_one_bit() { + let e = RangeEncoder::new(); + assert_eq!(e.tell(), 1); + assert_eq!(e.tell_frac(), 8); + } + + #[test] + fn roundtrip_mixed_symbols_and_tell_matches() { + let mut rng = Lcg(7); + let icdf: [u8; 4] = [200, 120, 30, 0]; + for trial in 0..300 { + let n = 1 + (rng.next() % 300) as usize; + let mut ops = Vec::new(); + for _ in 0..n { + match rng.next() % 5 { + 0 => { + let ft = 2 + rng.next() % 1000; + ops.push((0u8, rng.next() % ft, ft)); + } + 1 => { + let logp = 1 + rng.next() % 15; + ops.push((1, rng.next().is_multiple_of(5) as u32, logp)); + } + 2 => { + let bits = 1 + rng.next() % 24; + ops.push((2, rng.next() & ((1 << bits) - 1), bits)); + } + 3 => { + let ft = 2 + rng.next() % 100_000_000; + ops.push((3, rng.next() % ft, ft)); + } + _ => ops.push((4, rng.next() % 4, 0)), + } + } + let mut e = RangeEncoder::new(); + let mut tells = Vec::new(); + for &(k, v, p) in &ops { + match k { + 0 => e.encode(v, v + 1, p), + 1 => e.bit_logp(v != 0, p), + 2 => e.bits(v, p), + 3 => e.uint(v, p), + _ => e.icdf(v as usize, &icdf, 8), + } + tells.push(e.tell()); + } + // Exactly the number of bytes `tell` says were used must suffice. + let size = (e.tell() as usize).div_ceil(8); + let final_rng = e.rng(); + let buf = e.done(size); + assert_eq!(buf.len(), size); + let mut d = RangeDecoder::new(&buf); + for (i, &(k, v, p)) in ops.iter().enumerate() { + let got = match k { + 0 => { + let s = d.decode(p); + d.update(s, s + 1, p); + s + } + 1 => d.bit_logp(p) as u32, + 2 => d.bits(p), + 3 => d.uint(p), + _ => d.icdf(&icdf, 8) as u32, + }; + assert_eq!(got, v, "trial {trial} op {i} kind {k}"); + assert_eq!(d.tell(), tells[i], "tell mismatch trial {trial} op {i}"); + } + assert_eq!(d.rng(), final_rng); + } + } +} diff --git a/crates/audio/src/opus/encoder/rate.rs b/crates/audio/src/opus/encoder/rate.rs new file mode 100644 index 000000000..3eb9575d4 --- /dev/null +++ b/crates/audio/src/opus/encoder/rate.rs @@ -0,0 +1,243 @@ +//! Pulse cache and bit allocation (RFC 6716 §4.3.3, §4.3.4.1). +//! +//! The allocation must be bit-identical to the decoder's, so this mirrors FilmCraft's +//! spec-derived `filmcraft-opus` decoder (first-party, MIT OR Apache-2.0) step by step; the only +//! difference is that the encoder *writes* the band-skip, intensity and dual-stereo symbols. + +use crate::opus::encoder::CeltFrame; +use crate::opus::encoder::range::{BITRES, RangeEncoder}; +use crate::opus::encoder::tables::{EBANDS, LOG2_FRAC_TABLE, NB_EBANDS}; + +#[cfg(test)] +pub use crate::opus::rate_common::pvq_v; +pub use crate::opus::rate_common::{Mode, get_pulses}; +pub const MAX_FINE_BITS: i32 = 8; +const FINE_OFFSET: i32 = 21; +pub const QTHETA_OFFSET: i32 = 4; +const ALLOC_STEPS: i32 = 6; + +/// Result of the allocation process. +pub struct Allocation { + pub coded_bands: usize, + pub intensity: usize, + pub dual_stereo: bool, + pub balance: i32, + pub pulses: [i32; NB_EBANDS], + pub fine_quant: [i32; NB_EBANDS], + pub fine_priority: [i32; NB_EBANDS], +} + +/// Per-band limits, boosts and available budget for the allocation search. +pub struct AllocationInput<'a> { + pub offsets: &'a [i32; NB_EBANDS], + pub cap: &'a [i32; NB_EBANDS], + pub trim: i32, + pub bits: i32, +} + +/// `clt_compute_allocation`, encoder side. Stereo frames are always coded as dual (L/R) stereo +/// with no intensity bands, and no band is skipped voluntarily. +pub fn compute_allocation(m: &Mode, frame: CeltFrame, input: AllocationInput<'_>, enc: &mut RangeEncoder) -> Allocation { + let CeltFrame { start, end, channels: c, lm } = frame; + let AllocationInput { offsets, cap, trim: alloc_trim, bits: total } = input; + let crate::opus::rate_common::Bracket { skip_start, bits1, bits2, thresh, total, skip_rsv, intensity_rsv, dual_stereo_rsv } = + crate::opus::rate_common::bracket(start, end, offsets, cap, alloc_trim, total, c, lm); + let interpolation = + AllocationInterpolation { skip_start, bits1: &bits1, bits2: &bits2, thresh: &thresh, cap, total, skip_rsv, intensity_rsv, dual_stereo_rsv }; + interp_bits2pulses(m, frame, interpolation, enc) +} + +/// Bracketing allocation vectors and signalling reserves for the interpolation stage. +struct AllocationInterpolation<'a> { + skip_start: usize, + bits1: &'a [i32; NB_EBANDS], + bits2: &'a [i32; NB_EBANDS], + thresh: &'a [i32; NB_EBANDS], + cap: &'a [i32; NB_EBANDS], + total: i32, + skip_rsv: i32, + intensity_rsv: i32, + dual_stereo_rsv: i32, +} + +fn interp_bits2pulses(m: &Mode, frame: CeltFrame, interpolation: AllocationInterpolation<'_>, enc: &mut RangeEncoder) -> Allocation { + let CeltFrame { start, end, channels: c, lm } = frame; + let AllocationInterpolation { skip_start, bits1, bits2, thresh, cap, mut total, skip_rsv, mut intensity_rsv, mut dual_stereo_rsv } = + interpolation; + let ci = c as i32; + let alloc_floor = ci << BITRES; + let stereo = (c > 1) as i32; + let log_m = (lm as i32) << BITRES; + let mut lo = 0; + let mut hi = 1 << ALLOC_STEPS; + for _ in 0..ALLOC_STEPS { + let mid = (lo + hi) >> 1; + let mut psum = 0; + let mut done = false; + for j in (start..end).rev() { + let tmp = bits1[j] + ((mid * bits2[j]) >> ALLOC_STEPS); + if tmp >= thresh[j] || done { + done = true; + psum += tmp.min(cap[j]); + } else if tmp >= alloc_floor { + psum += alloc_floor; + } + } + if psum > total { + hi = mid; + } else { + lo = mid; + } + } + let mut bits = [0i32; NB_EBANDS]; + let mut psum = 0; + let mut done = false; + for j in (start..end).rev() { + let mut tmp = bits1[j] + ((lo * bits2[j]) >> ALLOC_STEPS); + if tmp < thresh[j] && !done { + tmp = if tmp >= alloc_floor { alloc_floor } else { 0 }; + } else { + done = true; + } + tmp = tmp.min(cap[j]); + bits[j] = tmp; + psum += tmp; + } + let s = start; + let mut coded_bands = end; + loop { + let j = coded_bands - 1; + if j <= skip_start { + total += skip_rsv; + break; + } + let mut left = total - psum; + let span = EBANDS[coded_bands] - EBANDS[s]; + let percoeff = left / span; + left -= span * percoeff; + let rem = (left - (EBANDS[j] - EBANDS[s])).max(0); + let band_width = EBANDS[coded_bands] - EBANDS[j]; + let band_bits = bits[j] + percoeff * band_width + rem; + if band_bits >= thresh[j].max(alloc_floor + (1 << BITRES)) { + // Keep this band (and every band below it). + enc.bit_logp(true, 1); + break; + } + psum -= bits[j] + intensity_rsv; + if intensity_rsv > 0 { + intensity_rsv = LOG2_FRAC_TABLE[j - s]; + } + psum += intensity_rsv; + if band_bits >= alloc_floor { + psum += alloc_floor; + bits[j] = alloc_floor; + } else { + bits[j] = 0; + } + coded_bands -= 1; + } + let intensity = if intensity_rsv > 0 { + // No intensity stereo: every coded band is coded in full. + enc.uint((coded_bands - s) as u32, (coded_bands + 1 - s) as u32); + coded_bands + } else { + 0 + }; + if intensity <= s { + total += dual_stereo_rsv; + dual_stereo_rsv = 0; + } + let dual_stereo = if dual_stereo_rsv > 0 { + enc.bit_logp(true, 1); + true + } else { + false + }; + let mut left = total - psum; + let span = EBANDS[coded_bands] - EBANDS[s]; + let percoeff = left / span; + left -= span * percoeff; + for j in s..coded_bands { + bits[j] += percoeff * (EBANDS[j + 1] - EBANDS[j]); + } + for j in s..coded_bands { + let tmp = left.min(EBANDS[j + 1] - EBANDS[j]); + bits[j] += tmp; + left -= tmp; + } + let mut ebits = [0i32; NB_EBANDS]; + let mut fine_priority = [0i32; NB_EBANDS]; + let mut balance = 0; + for j in s..coded_bands { + let n0 = EBANDS[j + 1] - EBANDS[j]; + let n = n0 << lm; + let bit = bits[j] + balance; + let mut excess; + if n > 1 { + excess = (bit - cap[j]).max(0); + bits[j] = bit - excess; + let den = ci * n + if c == 2 && n > 2 && !dual_stereo && j < intensity { 1 } else { 0 }; + let nclogn = den * (m.log_n[j] + log_m); + let mut offset = (nclogn >> 1) - den * FINE_OFFSET; + if n == 2 { + offset += (den << BITRES) >> 2; + } + if bits[j] + offset < (den * 2) << BITRES { + offset += nclogn >> 2; + } else if bits[j] + offset < (den * 3) << BITRES { + offset += nclogn >> 3; + } + let mut eb = (bits[j] + offset + (den << (BITRES - 1))).max(0); + eb = (eb / den) >> BITRES; + if ci * eb > (bits[j] >> BITRES) { + eb = bits[j] >> stereo >> BITRES; + } + eb = eb.min(MAX_FINE_BITS); + ebits[j] = eb; + fine_priority[j] = (eb * (den << BITRES) >= bits[j] + offset) as i32; + bits[j] -= (ci * eb) << BITRES; + } else { + excess = (bit - (ci << BITRES)).max(0); + bits[j] = bit - excess; + ebits[j] = 0; + fine_priority[j] = 1; + } + if excess > 0 { + let extra_fine = (excess >> (stereo + BITRES as i32)).min(MAX_FINE_BITS - ebits[j]); + ebits[j] += extra_fine; + let extra_bits = (extra_fine * ci) << BITRES; + fine_priority[j] = (extra_bits >= excess - balance) as i32; + excess -= extra_bits; + } + balance = excess; + } + for j in coded_bands..end { + ebits[j] = bits[j] >> stereo >> BITRES; + bits[j] = 0; + fine_priority[j] = (ebits[j] < 1) as i32; + } + Allocation { coded_bands, intensity, dual_stereo, balance, pulses: bits, fine_quant: ebits, fine_priority } +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn caps_match_decoder_reference() { + let m = Mode::new(); + let first_rows: [u8; 42] = [ + 224, 224, 224, 224, 224, 224, 224, 224, 160, 160, 160, 160, 185, 185, 185, 178, 178, 168, 134, 61, 37, 224, 224, 224, 224, 224, 224, 224, + 224, 240, 240, 240, 240, 207, 207, 207, 198, 198, 183, 144, 66, 40, + ]; + assert_eq!(&m.caps[..42], &first_rows[..]); + } + + #[test] + fn pvq_counts() { + assert_eq!(pvq_v(5, 1), 10); + assert_eq!(pvq_v(1, 7), 2); + assert_eq!(pvq_v(2, 9), 36); + assert_eq!(pvq_v(3, 2), 18); + } +} diff --git a/crates/audio/src/opus/encoder/resample.rs b/crates/audio/src/opus/encoder/resample.rs new file mode 100644 index 000000000..a72589919 --- /dev/null +++ b/crates/audio/src/opus/encoder/resample.rs @@ -0,0 +1,143 @@ +//! 48 kHz → 8/12/16 kHz decimator feeding the SILK layer (own design). +//! +//! A linear-phase low-pass FIR (Kaiser-windowed sinc at 48 kHz) evaluated only at the retained +//! output phases. Its group delay ([`delay`]) is a whole number of 48 kHz samples, chosen so +//! that the SILK path (this filter + the decoder's internal-rate delay and resampler, RFC 6716 +//! Table 54) lands within a fraction of a sample of the CELT path's 120-sample MDCT overlap. + +/// Group delay of the decimator to `fs_out` in 48 kHz samples (the filter has `2 * delay + 1` +/// taps). +pub fn delay(fs_out: u32) -> usize { + if fs_out == 8_000 { 88 } else { 83 } +} + +/// Zeroth-order modified Bessel function of the first kind (power series). +fn bessel_i0(x: f64) -> f64 { + let mut sum = 1.0; + let mut term = 1.0; + let q = x * x / 4.0; + for k in 1..64 { + term *= q / (k * k) as f64; + sum += term; + if term < sum * 1e-17 { + break; + } + } + sum +} + +/// Windowed-sinc low-pass with cutoff `fc` (Hz at 48 kHz), unit DC gain. +fn design(fc: f64, beta: f64, delay: usize) -> Vec { + let wc = 2.0 * std::f64::consts::PI * fc / 48_000.0; + let i0b = bessel_i0(beta); + let mut h: Vec = (0..2 * delay + 1) + .map(|k| { + let m = k as f64 - delay as f64; + let sinc = if m == 0.0 { wc / std::f64::consts::PI } else { (wc * m).sin() / (std::f64::consts::PI * m) }; + let r = m / delay as f64; + sinc * bessel_i0(beta * (1.0 - r * r).max(0.0).sqrt()) / i0b + }) + .collect(); + let s: f64 = h.iter().sum(); + for v in h.iter_mut() { + *v /= s; + } + h.iter().map(|&v| v as f32).collect() +} + +use crate::opus::dot::dot8; + +/// Decimates one channel of 48 kHz audio by 3, 4 or 6. +#[derive(Clone)] +pub struct Downsampler { + factor: usize, + h: Vec, + taps: usize, + /// The last `taps - 1` input samples followed by the current frame. + buf: Vec, +} + +impl Downsampler { + /// `fs_out` is 8000, 12000 or 16000. + pub fn new(fs_out: u32) -> Self { + let factor = (48_000 / fs_out) as usize; + // Cutoff at 95 % of the output Nyquist frequency (pass band to ~88 %, stop band from ~110 %). + let fc = 0.95 * fs_out as f64 / 2.0; + let d = delay(fs_out); + let taps = 2 * d + 1; + Downsampler { factor, h: design(fc, 6.5, d), taps, buf: vec![0.0; taps - 1] } + } + + /// Decimates `input` (a multiple of the factor long), appending to `out`. Output sample `j` + /// is the filtered input at frame position `factor * j`, delayed by [`delay`]. + pub fn process(&mut self, input: &[f32], out: &mut Vec) { + debug_assert!(input.len().is_multiple_of(self.factor)); + let taps = self.taps; + self.buf.truncate(taps - 1); + self.buf.extend_from_slice(input); + for j in 0..input.len() / self.factor { + let t = taps - 1 + self.factor * j; + let win = &self.buf[t + 1 - taps..=t]; + // h is symmetric, so the window can be taken in forward order. + out.push(dot8(win, &self.h)); + } + let n = self.buf.len(); + self.buf.copy_within(n - (taps - 1).., 0); + self.buf.truncate(taps - 1); + } +} + +#[cfg(test)] +mod tests { + use super::*; + + fn tone(f: f64, n: usize) -> Vec { + (0..n).map(|i| (2.0 * std::f64::consts::PI * f * i as f64 / 48_000.0).sin() as f32).collect() + } + + fn rms(x: &[f32]) -> f64 { + (x.iter().map(|&v| (v as f64).powi(2)).sum::() / x.len() as f64).sqrt() + } + + #[test] + fn dot8_matches_sequential_sum() { + for len in [0usize, 1, 7, 8, 9, 63, 167, 251] { + let a: Vec = (0..len).map(|i| ((i * 37 % 101) as f32 - 50.0) / 50.0).collect(); + let b: Vec = (0..len).map(|i| ((i * 53 % 89) as f32 - 44.0) / 44.0).collect(); + let exact: f64 = a.iter().zip(&b).map(|(&x, &y)| x as f64 * y as f64).sum(); + assert!((dot8(&a, &b) as f64 - exact).abs() < 1e-4, "len {len}"); + } + } + + #[test] + fn passes_band_and_rejects_aliases() { + for &(fs, pass, stop) in &[(16_000u32, 6_000.0, 8_600.0), (12_000, 4_500.0, 6_500.0), (8_000, 3_000.0, 4_500.0)] { + for &(f, want_pass) in &[(pass, true), (stop, false)] { + let mut d = Downsampler::new(fs); + let x = tone(f, 9600); + let mut y = Vec::new(); + for c in x.chunks(960) { + d.process(c, &mut y); + } + let r = rms(&y[y.len() / 2..]) * std::f64::consts::SQRT_2; + if want_pass { + assert!((r - 1.0).abs() < 0.02, "{fs}: {f} Hz gain {r}"); + } else { + assert!(r < 0.01, "{fs}: {f} Hz leaks {r}"); + } + } + } + } + + #[test] + fn delay_is_exact() { + let mut d = Downsampler::new(16_000); + let mut x = vec![0f32; 960]; + x[300] = 1.0; + let mut y = Vec::new(); + d.process(&x, &mut y); + // The impulse at 300 shows up centred at 48 kHz time 300 + 83 = 383 (output 127.67). + let peak = y.iter().enumerate().max_by(|a, b| a.1.total_cmp(b.1)).map(|(i, _)| i).unwrap_or(0); + assert!((peak as f64 - (300.0 + delay(16_000) as f64) / 3.0).abs() < 1.0, "peak at {peak}"); + } +} diff --git a/crates/audio/src/opus/encoder/silk/analysis.rs b/crates/audio/src/opus/encoder/silk/analysis.rs new file mode 100644 index 000000000..d80052046 --- /dev/null +++ b/crates/audio/src/opus/encoder/silk/analysis.rs @@ -0,0 +1,418 @@ +//! Encoder-side signal analysis for SILK (own design): windowed autocorrelation LPC, LPC → NLSF +//! conversion, open-loop pitch search and LTP codebook search. + +use super::tables::{CB_NB_20, CB_WB_20, LTP_FILTER, LTP_PERIODICITY, LTP_TAPS_0, LTP_TAPS_1, LTP_TAPS_2}; + +/// Short-term prediction coefficients `a` (prediction `x[n] ≈ Σ a[j] x[n-1-j]`) from the +/// windowed segment `x`, with a Gaussian lag window and white-noise correction. +pub fn lpc(x: &[f32], order: usize, fs_khz: usize) -> Vec { + let n = x.len(); + // Asymmetric window: sin² rise over the first half, flat, cos fall over the last 1/8. + let rise = n / 2; + let fall = n / 8; + let w = |i: usize| -> f64 { + if i < rise { + let s = (std::f64::consts::FRAC_PI_2 * (i as f64 + 0.5) / rise as f64).sin(); + s * s + } else if i >= n - fall { + (std::f64::consts::FRAC_PI_2 * (i + fall - n) as f64 / fall as f64 + 0.0).cos() + } else { + 1.0 + } + }; + let xw: Vec = x.iter().enumerate().map(|(i, &v)| v as f64 * w(i)).collect(); + let mut r = vec![0f64; order + 1]; + for (k, rk) in r.iter_mut().enumerate() { + *rk = xw[k..].iter().zip(&xw).map(|(a, b)| a * b).sum(); + } + if r[0] <= 1e-6 { + return vec![0.0; order]; + } + // White-noise correction (-40 dB) and a 50 Hz Gaussian lag window. + r[0] *= 1.0 + 1e-4; + let f0 = 2.0 * std::f64::consts::PI * 50.0 / (fs_khz as f64 * 1000.0); + for (k, rk) in r.iter_mut().enumerate().skip(1) { + *rk *= (-0.5 * (f0 * k as f64).powi(2)).exp(); + } + let mut a = levinson(&r, order); + // Mild bandwidth expansion keeps the NLSFs apart. + let mut g = 1.0; + for v in a.iter_mut() { + g *= 0.996; + *v *= g; + } + a +} + +/// Levinson-Durbin recursion on autocorrelation `r` (predictor convention as in [`lpc`]). +pub fn levinson(r: &[f64], order: usize) -> Vec { + let mut a = vec![0f64; order]; + let mut err = r[0]; + let mut tmp = vec![0f64; order]; + for i in 0..order { + let mut acc = r[i + 1]; + for j in 0..i { + acc -= a[j] * r[i - j]; + } + let k = (acc / err).clamp(-0.9999, 0.9999); + tmp[..i].copy_from_slice(&a[..i]); + for j in 0..i { + a[j] = tmp[j] - k * tmp[i - 1 - j]; + } + a[i] = k; + err *= 1.0 - k * k; + if err <= 0.0 { + break; + } + } + a +} + +/// Evaluates the symmetric polynomial `p` (length `d + 1`, `d` even) as the cosine series +/// `e^{jωd/2} P(e^{jω})`, with `cos_m(m)` giving `cos(m ω)`. +#[inline] +fn cos_series_with(p: &[f64], cos_m: impl Fn(usize) -> f64) -> f64 { + let h = (p.len() - 1) / 2; + let mut s = p[h]; + for k in 0..h { + s += 2.0 * p[k] * cos_m(h - k); + } + s +} + +fn cos_series(p: &[f64], w: f64) -> f64 { + cos_series_with(p, |m| (m as f64 * w).cos()) +} + +/// Number of grid intervals the root search scans over `[0, Ï€]`. +const GRID: usize = 1024; +/// Largest multiple of ω in the series (half the LPC order, at most 16 / 2). +const MAX_H: usize = 8; +/// Bisection steps per root: the bracket shrinks to 3e-9 rad, far below one Q15 NLSF step (9.6e-5 rad). +const BISECT_STEPS: usize = 20; + +/// `cos(m · Ï€ g / GRID)` for every grid point `g` and `m` in `0..=MAX_H`, computed with the same +/// expression as [`cos_series`], so the table values are bit-identical to calling `cos` there. +fn grid_cos() -> &'static [[f64; MAX_H + 1]] { + static TABLE: std::sync::OnceLock> = std::sync::OnceLock::new(); + TABLE.get_or_init(|| { + (0..=GRID) + .map(|g| { + let w = std::f64::consts::PI * g as f64 / GRID as f64; + std::array::from_fn(|m| (m as f64 * w).cos()) + }) + .collect() + }) +} + +fn roots(p: &[f64], out: &mut Vec) { + let grid = grid_cos(); + let at_grid = |g: usize| cos_series_with(p, |m| grid[g][m]); + let mut w0 = 0.0; + let mut f0 = at_grid(0); + for g in 1..=GRID { + let w1 = std::f64::consts::PI * g as f64 / GRID as f64; + let f1 = at_grid(g); + if f0 == 0.0 { + out.push(w0); + } else if f0 * f1 < 0.0 { + let (mut lo, mut hi, mut flo) = (w0, w1, f0); + for _ in 0..BISECT_STEPS { + let mid = 0.5 * (lo + hi); + let fm = cos_series(p, mid); + if fm * flo <= 0.0 { + hi = mid; + } else { + lo = mid; + flo = fm; + } + } + out.push(0.5 * (lo + hi)); + } + w0 = w1; + f0 = f1; + } +} + +/// LPC → normalised line spectral frequencies in radians (ascending, `order` values), or `None` +/// when the root search fails (unstable or degenerate filter). +pub fn a2nlsf(a: &[f64]) -> Option> { + let d = a.len(); + if d > 2 * MAX_H { + return None; + } + let mut c = vec![0f64; d + 2]; + c[0] = 1.0; + for k in 0..d { + c[k + 1] = -a[k]; + } + let mut p = vec![0f64; d + 2]; + let mut q = vec![0f64; d + 2]; + for k in 0..d + 2 { + p[k] = c[k] + c[d + 1 - k]; + q[k] = c[k] - c[d + 1 - k]; + } + // Remove the trivial roots at z = -1 (P) and z = +1 (Q). + let mut p1 = vec![0f64; d + 1]; + let mut q1 = vec![0f64; d + 1]; + p1[0] = p[0]; + q1[0] = q[0]; + for k in 1..=d { + p1[k] = p[k] - p1[k - 1]; + q1[k] = q[k] + q1[k - 1]; + } + let mut r = Vec::with_capacity(d); + roots(&p1, &mut r); + roots(&q1, &mut r); + if r.len() != d { + return None; + } + r.sort_by(f64::total_cmp); + Some(r) +} + +/// Normalised cross-correlation of `x[n]` and `x[n - lag]` over `n` in `range`. +fn ncorr(x: &[f32], range: std::ops::Range, lag: usize) -> f64 { + let (mut xy, mut xx, mut yy) = (0f64, 0f64, 0f64); + for n in range { + let a = x[n] as f64; + let b = x[n - lag] as f64; + xy += a * b; + xx += a * a; + yy += b * b; + } + if xy <= 0.0 { 0.0 } else { xy / (xx * yy + 1e-9).sqrt() } +} + +pub struct Pitch { + pub lag_index: i32, + pub contour: usize, + pub lags: [i32; 4], + pub score: f64, +} + +/// Open-loop pitch search on `x` (history followed by the frame starting at `start`, `len` +/// samples, four subframes) at `fs_khz`. +pub fn pitch_search(x: &[f32], start: usize, len: usize, fs_khz: usize) -> Pitch { + let fs = fs_khz as i32; + let min_lag = 2 * fs; + let max_lag = 18 * fs - 1; + let sub = len / 4; + let nl = (max_lag - min_lag + 1) as usize; + // Whole-frame score per lag. + let full: Vec = (0..nl).map(|i| ncorr(x, start..start + len, min_lag as usize + i)).collect(); + let mut best = 0usize; + for i in 0..nl { + // Slight preference for shorter lags. + if full[i] * (1.0 - 0.0004 * i as f64) > full[best] * (1.0 - 0.0004 * best as f64) { + best = i; + } + } + let mut lag = min_lag + best as i32; + // Prefer a sub-multiple when it explains the signal almost as well (octave errors). + for div in [4, 3, 2] { + let cand = (lag as f64 / div as f64).round() as i32; + let mut cbest = (0.0, 0); + for c in cand - 1..=cand + 1 { + if c >= min_lag && c <= max_lag { + let s = full[(c - min_lag) as usize]; + if s > cbest.0 { + cbest = (s, c); + } + } + } + if cbest.1 != 0 && cbest.0 > 0.9 * full[(lag - min_lag) as usize] { + lag = cbest.1; + break; + } + } + let score = full[(lag - min_lag) as usize]; + // Contour: per-subframe lag offsets around the chosen lag. + let cb: &[[i32; 4]] = if fs_khz == 8 { &CB_NB_20 } else { &CB_WB_20 }; + // The contour search revisits the same (subframe, lag) pairs for neighbouring base lags and + // codebook entries, so each correlation is computed once. + let span = (18 * fs - min_lag + 1) as usize; + let memo = std::cell::RefCell::new(vec![f64::NAN; 4 * span]); + let sub_score = |k: usize, l: i32| -> f64 { + let l = l.clamp(min_lag, 18 * fs); + let slot = k * span + (l - min_lag) as usize; + let cached = memo.borrow()[slot]; + if !cached.is_nan() { + return cached; + } + let v = ncorr(x, start + k * sub..start + (k + 1) * sub, l as usize); + memo.borrow_mut()[slot] = v; + v + }; + let mut best = (f64::MIN, lag, 0usize); + for l0 in (lag - 2).max(min_lag)..=(lag + 2).min(max_lag) { + for (ci, offs) in cb.iter().enumerate() { + let s: f64 = (0..4).map(|k| sub_score(k, l0 + offs[k])).sum::() - 0.002 * ci as f64; + if s > best.0 { + best = (s, l0, ci); + } + } + } + let (_, l0, ci) = best; + let mut lags = [0i32; 4]; + for k in 0..4 { + lags[k] = (l0 + cb[ci][k]).clamp(min_lag, 18 * fs); + } + Pitch { lag_index: l0 - min_lag, contour: ci, lags, score } +} + +/// Result of the LTP codebook search. +pub struct Ltp { + pub per_index: usize, + pub ltp_index: [usize; 4], + /// Residual energy after LTP prediction / before, per subframe. + pub err: [f64; 4], + pub energy: [f64; 4], +} + +fn icdf_bits(icdf: &[u8], s: usize) -> f64 { + let hi = if s == 0 { 256 } else { icdf[s - 1] as u32 }; + let p = (hi - icdf[s] as u32).max(1) as f64 / 256.0; + -p.log2() +} + +/// Searches the three LTP codebooks (RFC 6716 Tables 39–42) for the taps that best predict the +/// residual `r` (history + frame starting at `start`) from its past at the per-subframe `lags`. +pub fn ltp_search(r: &[f64], start: usize, sub: usize, lags: &[i32; 4]) -> Ltp { + let mut rr = [[0f64; 25]; 4]; + let mut cv = [[0f64; 5]; 4]; + let mut e = [0f64; 4]; + for k in 0..4 { + let s0 = start + k * sub; + let lag = lags[k] as usize; + for n in s0..s0 + sub { + let t = r[n]; + e[k] += t * t; + let base = n + 2 - lag; + for i in 0..5 { + let vi = r[base - i]; + cv[k][i] += t * vi; + for j in 0..5 { + rr[k][i * 5 + j] += vi * r[base - j]; + } + } + } + } + let books: [&[[i32; 5]]; 3] = [<P_TAPS_0, <P_TAPS_1, <P_TAPS_2]; + let mut best = (f64::MAX, 0usize, [0usize; 4], [0f64; 4]); + for (p, book) in books.iter().enumerate() { + let mut total = icdf_bits(<P_PERIODICITY, p) * 1e-3 * e.iter().sum::() / 4.0; + let mut idx = [0usize; 4]; + let mut errs = [0f64; 4]; + for k in 0..4 { + let mut bk = (f64::MAX, 0usize, 0f64); + for (i, taps) in book.iter().enumerate() { + let b: [f64; 5] = std::array::from_fn(|j| taps[j] as f64 / 128.0); + let mut err = e[k]; + for a in 0..5 { + err -= 2.0 * b[a] * cv[k][a]; + for c in 0..5 { + err += b[a] * b[c] * rr[k][a * 5 + c]; + } + } + let cost = err + icdf_bits(LTP_FILTER[p], i) * 1e-3 * e[k]; + if cost < bk.0 { + bk = (cost, i, err.max(0.0)); + } + } + total += bk.0; + idx[k] = bk.1; + errs[k] = bk.2; + } + if total < best.0 { + best = (total, p, idx, errs); + } + } + Ltp { per_index: best.1, ltp_index: best.2, err: best.3, energy: e } +} + +#[cfg(test)] +mod tests { + use super::super::lpc::nlsf2a; + use super::*; + + #[test] + fn nlsf_roundtrip_through_decoder_conversion() { + // A resonant 10th/16th-order predictor from a synthetic signal. + for &(order, fs) in &[(10usize, 8usize), (16, 16)] { + let mut st = [0f64; 4]; + let mut seed = 1u32; + let x: Vec = (0..400) + .map(|_| { + seed = seed.wrapping_mul(1_664_525).wrapping_add(1_013_904_223); + let w = (seed >> 9) as f64 / (1u32 << 23) as f64 - 1.0; + let y = w + 1.6 * st[0] - 0.9 * st[1] + 0.3 * st[2] - 0.2 * st[3]; + st = [y, st[0], st[1], st[2]]; + (y * 1000.0) as f32 + }) + .collect(); + let a = lpc(&x, order, fs); + let nlsf = a2nlsf(&a).expect("roots"); + let q15: Vec = nlsf.iter().map(|w| (w / std::f64::consts::PI * 32768.0).round() as i32).collect(); + let a_q12 = nlsf2a(&q15); + for j in 0..order { + assert!((a_q12[j] as f64 / 4096.0 - a[j]).abs() < 0.02, "order {order} coef {j}: {} vs {}", a_q12[j] as f64 / 4096.0, a[j]); + } + } + } + + /// The root search as it was before the grid table: `cos` called for every evaluation. + fn roots_reference(p: &[f64], out: &mut Vec) { + let mut w0 = 0.0; + let mut f0 = cos_series(p, w0); + for g in 1..=GRID { + let w1 = std::f64::consts::PI * g as f64 / GRID as f64; + let f1 = cos_series(p, w1); + if f0 == 0.0 { + out.push(w0); + } else if f0 * f1 < 0.0 { + let (mut lo, mut hi, mut flo) = (w0, w1, f0); + for _ in 0..BISECT_STEPS { + let mid = 0.5 * (lo + hi); + let fm = cos_series(p, mid); + if fm * flo <= 0.0 { + hi = mid; + } else { + lo = mid; + flo = fm; + } + } + out.push(0.5 * (lo + hi)); + } + w0 = w1; + f0 = f1; + } + } + + #[test] + fn grid_table_roots_are_bit_identical_to_direct_cos() { + let mut seed = 7u64; + let mut rnd = move || { + seed ^= seed << 13; + seed ^= seed >> 7; + seed ^= seed << 17; + (seed >> 11) as f64 / (1u64 << 53) as f64 * 2.0 - 1.0 + }; + let mut found = 0; + for len in [11usize, 17] { + for _ in 0..300 { + let mut p = vec![0f64; len]; + for k in 0..len / 2 + 1 { + p[k] = rnd(); + p[len - 1 - k] = p[k]; + } + let (mut a, mut b) = (Vec::new(), Vec::new()); + roots(&p, &mut a); + roots_reference(&p, &mut b); + assert_eq!(a.iter().map(|v| v.to_bits()).collect::>(), b.iter().map(|v| v.to_bits()).collect::>()); + found += a.len(); + } + } + assert!(found > 1000, "{found}"); + } +} diff --git a/crates/audio/src/opus/encoder/silk/lpc.rs b/crates/audio/src/opus/encoder/silk/lpc.rs new file mode 100644 index 000000000..e4e9ae624 --- /dev/null +++ b/crates/audio/src/opus/encoder/silk/lpc.rs @@ -0,0 +1,370 @@ +//! Fixed-point NLSF decoding, NLSF→LPC conversion and filter stabilisation +//! (RFC 6716 §4.2.7.5, with the RFC 8251 §6/§7 overflow fixes). +//! +//! This is the decoder's normative arithmetic, mirrored from FilmCraft's first-party, +//! spec-derived `filmcraft-opus` decoder (MIT OR Apache-2.0): the encoder must reconstruct the +//! quantised NLSFs and LPC coefficients bit-exactly as the decoder does. + +use super::tables::{ + COS_Q12, LSF_ORDER_NBMB, LSF_ORDER_WB, NLSF_CB_NBMB, NLSF_CB_WB, NLSF_MIN_SPACING_NBMB, NLSF_MIN_SPACING_WB, NLSF_PRED_NBMB, NLSF_PRED_WB, + NLSF_PSEL_NBMB, NLSF_PSEL_WB, +}; + +pub const MAX_LPC_ORDER: usize = 16; + +#[inline] +pub fn smulwb(a: i32, b: i32) -> i32 { + ((a as i64 * (b as i16) as i64) >> 16) as i32 +} +#[inline] +pub fn smlawb(a: i32, b: i32, c: i32) -> i32 { + a.wrapping_add(smulwb(b, c)) +} +#[inline] +pub fn smulww(a: i32, b: i32) -> i32 { + ((a as i64 * b as i64) >> 16) as i32 +} +#[inline] +pub fn smmul(a: i32, b: i32) -> i32 { + ((a as i64 * b as i64) >> 32) as i32 +} +#[inline] +pub fn rshift_round(a: i32, s: u32) -> i32 { + if s == 1 { (a >> 1) + (a & 1) } else { ((a >> (s - 1)) + 1) >> 1 } +} +#[inline] +pub fn rshift_round64(a: i64, s: u32) -> i64 { + if s == 1 { (a >> 1) + (a & 1) } else { ((a >> (s - 1)) + 1) >> 1 } +} +#[inline] +pub fn sat16(a: i32) -> i32 { + a.clamp(-32768, 32767) +} + +/// `silk_INVERSE32_varQ`: approximation of `(1 << q_res) / b32`. +pub fn inverse32_varq(b32: i32, q_res: i32) -> i32 { + let b_headrm = b32.unsigned_abs().leading_zeros() as i32 - 1; + let b32_nrm = b32.wrapping_shl(b_headrm as u32); + let b32_inv = (i32::MAX >> 2) / (b32_nrm >> 16); + let mut result = b32_inv << 16; + let err_q32 = ((1i32 << 29).wrapping_sub(smulwb(b32_nrm, b32_inv))).wrapping_shl(3); + result = result.wrapping_add(((err_q32 as i64 * b32_inv as i64) >> 16) as i32); + let lshift = 61 - b_headrm - q_res; + if lshift <= 0 { + let s = (-lshift) as u32; + let lim_hi = i32::MAX >> s; + let lim_lo = i32::MIN >> s; + result.clamp(lim_lo, lim_hi) << s + } else if lshift < 32 { + result >> lshift + } else { + 0 + } +} + +/// `silk_DIV32_varQ`: approximation of `(a32 << q_res) / b32`. +pub fn div32_varq(a32: i32, b32: i32, q_res: i32) -> i32 { + let a_headrm = a32.unsigned_abs().leading_zeros() as i32 - 1; + let a32_nrm = a32.wrapping_shl(a_headrm as u32); + let b_headrm = b32.unsigned_abs().leading_zeros() as i32 - 1; + let b32_nrm = b32.wrapping_shl(b_headrm as u32); + let b32_inv = (i32::MAX >> 2) / (b32_nrm >> 16); + let mut result = smulwb(a32_nrm, b32_inv); + let a32_nrm = a32_nrm.wrapping_sub(((smmul(b32_nrm, result) as i64) << 3) as i32); + result = smlawb(result, a32_nrm, b32_inv); + let lshift = 29 + a_headrm - b_headrm - q_res; + if lshift < 0 { + let s = (-lshift) as u32; + let lim_hi = i32::MAX >> s; + let lim_lo = i32::MIN >> s; + result.clamp(lim_lo, lim_hi) << s + } else if lshift < 32 { + result >> lshift + } else { + 0 + } +} + +pub fn sqrt_approx(x: i32) -> i32 { + if x <= 0 { + return 0; + } + let lz = x.leading_zeros() as i32; + let frac = (x.rotate_right((24 - lz).rem_euclid(32) as u32)) & 0x7f; + let mut y = if lz & 1 != 0 { 32768 } else { 46214 }; + y >>= lz >> 1; + smlawb(y, y, 213 * frac) +} + +/// Decodes the stage-1/stage-2 indices into stabilised NLSFs (Q15). +pub fn nlsf_decode(indices: &[i32], wb: bool) -> [i32; MAX_LPC_ORDER] { + let order = if wb { 16 } else { 10 }; + let i1 = indices[0] as usize; + let cb1: &[i32] = if wb { &NLSF_CB_WB[i1] } else { &NLSF_CB_NBMB[i1] }; + let qstep = if wb { 9830 } else { 11796 }; + // Residual dequantisation with backwards prediction. + let mut res_q10 = [0i32; MAX_LPC_ORDER]; + let mut out_q10 = 0i32; + for i in (0..order).rev() { + let pred = if i + 1 < order { + let sel = if wb { NLSF_PSEL_WB[i1][i] } else { NLSF_PSEL_NBMB[i1][i] } as usize; + let w = if wb { NLSF_PRED_WB[sel][i] } else { NLSF_PRED_NBMB[sel][i] }; + (out_q10 * w) >> 8 + } else { + 0 + }; + let mut v = indices[i + 1] << 10; + if v > 0 { + v -= 102; + } else if v < 0 { + v += 102; + } + out_q10 = smlawb(pred, v, qstep); + res_q10[i] = out_q10; + } + // Laroia weights. + let mut nlsf = [0i32; MAX_LPC_ORDER]; + for k in 0..order { + let prev = if k == 0 { 0 } else { cb1[k - 1] }; + let next = if k + 1 == order { 256 } else { cb1[k + 1] }; + let w = (1024 / (cb1[k] - prev).max(1) + 1024 / (next - cb1[k]).max(1)).min(32767); + let w_q9 = sqrt_approx(w << 16); + let v = (cb1[k] << 7) + (res_q10[k] << 14) / w_q9.max(1); + nlsf[k] = v.clamp(0, 32767); + } + let dmin: &[i32] = if wb { &NLSF_MIN_SPACING_WB } else { &NLSF_MIN_SPACING_NBMB }; + nlsf_stabilize(&mut nlsf[..order], dmin); + nlsf +} + +pub fn nlsf_stabilize(nlsf: &mut [i32], dmin: &[i32]) { + let l = nlsf.len(); + for _ in 0..20 { + let mut min_diff = nlsf[0] - dmin[0]; + let mut idx = 0; + for i in 1..l { + let d = nlsf[i] - (nlsf[i - 1] + dmin[i]); + if d < min_diff { + min_diff = d; + idx = i; + } + } + let d = 32768 - (nlsf[l - 1] + dmin[l]); + if d < min_diff { + min_diff = d; + idx = l; + } + if min_diff >= 0 { + return; + } + if idx == 0 { + nlsf[0] = dmin[0]; + } else if idx == l { + nlsf[l - 1] = 32768 - dmin[l]; + } else { + let mut min_center = dmin[..idx].iter().sum::(); + min_center += dmin[idx] >> 1; + let mut max_center = 32768 - dmin[idx + 1..=l].iter().sum::(); + max_center -= dmin[idx] >> 1; + let c = rshift_round(nlsf[idx - 1] + nlsf[idx], 1); + let c = if min_center > max_center { c.clamp(max_center, min_center) } else { c.clamp(min_center, max_center) }; + nlsf[idx - 1] = c - (dmin[idx] >> 1); + nlsf[idx] = nlsf[idx - 1] + dmin[idx]; + } + } + // Fallback. + nlsf.sort_unstable(); + nlsf[0] = nlsf[0].max(dmin[0]); + for i in 1..l { + nlsf[i] = nlsf[i].max(sat16(nlsf[i - 1] + dmin[i])); + } + nlsf[l - 1] = nlsf[l - 1].min(32768 - dmin[l]); + for i in (0..l - 1).rev() { + nlsf[i] = nlsf[i].min(nlsf[i + 1] - dmin[i + 1]); + } +} + +fn bwexpander_32(ar: &mut [i32], mut chirp_q16: i32) { + let d = ar.len(); + let chirp_minus_one = chirp_q16 - 65536; + for v in ar.iter_mut().take(d - 1) { + *v = smulww(chirp_q16, *v); + chirp_q16 += rshift_round(chirp_q16.wrapping_mul(chirp_minus_one), 16); + } + ar[d - 1] = smulww(chirp_q16, ar[d - 1]); +} + +fn lpc_inverse_pred_gain(a_q12: &[i32]) -> i32 { + const QA: u32 = 24; + const A_LIMIT: i32 = 16_773_022; + let order = a_q12.len(); + let mut a = [0i32; MAX_LPC_ORDER]; + let mut dc = 0; + for k in 0..order { + dc += a_q12[k]; + a[k] = a_q12[k] << (QA - 12); + } + if dc >= 4096 { + return 0; + } + let mut inv_gain: i32 = 1 << 30; + let mul32_frac_q31 = |x: i32, y: i32| rshift_round64(x as i64 * y as i64, 31) as i32; + for k in (1..order).rev() { + if a[k] > A_LIMIT || a[k] < -A_LIMIT { + return 0; + } + let rc_q31 = -(a[k] << (31 - QA)); + let rc_mult1 = (1i32 << 30) - smmul(rc_q31, rc_q31); + inv_gain = smmul(inv_gain, rc_mult1) << 2; + if inv_gain < 107_374 { + return 0; + } + let mult2q = 32 - rc_mult1.unsigned_abs().leading_zeros() as i32; + let rc_mult2 = inverse32_varq(rc_mult1, mult2q + 30); + for n in 0..(k + 1) >> 1 { + let t1 = a[n]; + let t2 = a[k - n - 1]; + let v1 = rshift_round64(t1.saturating_sub(mul32_frac_q31(t2, rc_q31)) as i64 * rc_mult2 as i64, mult2q as u32); + if v1 > i32::MAX as i64 || v1 < i32::MIN as i64 { + return 0; + } + let v2 = rshift_round64(t2.saturating_sub(mul32_frac_q31(t1, rc_q31)) as i64 * rc_mult2 as i64, mult2q as u32); + if v2 > i32::MAX as i64 || v2 < i32::MIN as i64 { + return 0; + } + a[n] = v1 as i32; + a[k - n - 1] = v2 as i32; + } + } + if a[0] > A_LIMIT || a[0] < -A_LIMIT { + return 0; + } + let rc_q31 = -(a[0] << (31 - QA)); + let rc_mult1 = (1i32 << 30) - smmul(rc_q31, rc_q31); + inv_gain = smmul(inv_gain, rc_mult1) << 2; + if inv_gain < 107_374 { + return 0; + } + inv_gain +} + +/// Converts NLSFs (Q15) to stable Q12 LPC coefficients (`silk_NLSF2A`). +pub fn nlsf2a(nlsf: &[i32]) -> [i32; MAX_LPC_ORDER] { + let d = nlsf.len(); + let ordering: &[usize] = if d == 16 { &LSF_ORDER_WB } else { &LSF_ORDER_NBMB }; + let mut c_q17 = [0i32; MAX_LPC_ORDER]; + for k in 0..d { + let i = (nlsf[k] >> 8) as usize; + let f = nlsf[k] & 255; + let c0 = COS_Q12[i]; + let c1 = COS_Q12[(i + 1).min(128)]; + c_q17[ordering[k]] = (c0 * 256 + (c1 - c0) * f + 4) >> 3; + } + let d2 = d / 2; + let find_poly = |off: usize| -> [i64; 10] { + let mut out = [0i64; 10]; + out[0] = 1 << 16; + out[1] = -(c_q17[off] as i64); + for k in 1..d2 { + let ftmp = c_q17[2 * k + off] as i64; + out[k + 1] = 2 * out[k - 1] - ((ftmp * out[k] + 32768) >> 16); + for n in (2..=k).rev() { + out[n] += out[n - 2] - ((ftmp * out[n - 1] + 32768) >> 16); + } + out[1] -= ftmp; + } + out + }; + let p = find_poly(0); + let q = find_poly(1); + let mut a32 = [0i32; MAX_LPC_ORDER]; + for k in 0..d2 { + let ptmp = p[k + 1] + p[k]; + let qtmp = q[k + 1] - q[k]; + a32[k] = (-qtmp - ptmp) as i32; + a32[d - k - 1] = (qtmp - ptmp) as i32; + } + let a32 = &mut a32[..d]; + // Limit the dynamic range (silk_LPC_fit with QIN = 17, QOUT = 12). + let mut a_q12 = [0i32; MAX_LPC_ORDER]; + let mut i = 0; + while i < 10 { + let mut maxabs = 0i32; + let mut idx = 0; + for (k, &v) in a32.iter().enumerate() { + let av = v.saturating_abs(); + if av > maxabs { + maxabs = av; + idx = k; + } + } + let maxabs = rshift_round(maxabs, 5); + if maxabs > 32767 { + let maxabs = maxabs.min(163_838); + let chirp = 65470 - ((maxabs - 32767) << 14) / ((maxabs * (idx as i32 + 1)) >> 2); + bwexpander_32(a32, chirp); + } else { + break; + } + i += 1; + } + if i == 10 { + for k in 0..d { + a_q12[k] = sat16(rshift_round(a32[k], 5)); + a32[k] = a_q12[k] << 5; + } + } else { + for k in 0..d { + a_q12[k] = rshift_round(a32[k], 5); + } + } + for i in 0..16 { + if lpc_inverse_pred_gain(&a_q12[..d]) == 0 { + bwexpander_32(a32, 65536 - (2 << i)); + for k in 0..d { + a_q12[k] = rshift_round(a32[k], 5); + } + } else { + break; + } + } + a_q12 +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn flat_nlsf_gives_small_lpc() { + // Uniformly spaced NLSFs correspond to an (almost) flat spectrum. + let nlsf: Vec = (1..=16).map(|k| k * 32768 / 17).collect(); + let a = nlsf2a(&nlsf); + assert!(a.iter().all(|v| v.abs() < 600), "{a:?}"); + let nlsf: Vec = (1..=10).map(|k| k * 32768 / 11).collect(); + let a = nlsf2a(&nlsf); + assert!(a[..10].iter().all(|v| v.abs() < 600), "{a:?}"); + } + + #[test] + fn stabilize_enforces_spacing() { + let mut v = [100, 90, 80, 30000, 30001, 30002, 31000, 32000, 32700, 32760]; + nlsf_stabilize(&mut v, &NLSF_MIN_SPACING_NBMB); + assert!(v[0] >= NLSF_MIN_SPACING_NBMB[0]); + for i in 1..10 { + assert!(v[i] - v[i - 1] >= NLSF_MIN_SPACING_NBMB[i], "{v:?}"); + } + assert!(32768 - v[9] >= NLSF_MIN_SPACING_NBMB[10]); + } + + #[test] + fn inverse_approximations() { + for &b in &[65536, 100_000, 1 << 20, 81920, 1_686_110_208] { + let inv = inverse32_varq(b, 47) as f64; + let exact = 2f64.powi(47) / b as f64; + assert!((inv - exact).abs() / exact < 1e-3, "{b}"); + } + let q = div32_varq(65536, 131072, 16) as f64; + assert!((q - 32768.0).abs() < 2.0); + } +} diff --git a/crates/audio/src/opus/encoder/silk/mod.rs b/crates/audio/src/opus/encoder/silk/mod.rs new file mode 100644 index 000000000..0fda4f59a --- /dev/null +++ b/crates/audio/src/opus/encoder/silk/mod.rs @@ -0,0 +1,862 @@ +//! SILK layer encoder (RFC 6716 §4.2, encoder direction), 20 ms frames at 8, 12 or 16 kHz. +//! +//! Each frame is analysed (VAD, open-loop pitch, windowed-autocorrelation LPC, LTP codebook +//! search), its NLSFs are quantised, and the excitation is chosen sample by sample by running +//! the decoder's exact fixed-point reconstruction (LTP and LPC synthesis with the dequantised +//! Q12/Q14 coefficients, quantisation offsets and the pseudo-random sign sequence), so the +//! encoder always holds the decoder's state and there is no drift. A noise-feedback loop +//! shapes the coding noise with a bandwidth-expanded LPC envelope, and a per-frame search over +//! a gain offset hits the bit budget. Stereo input is coded as mid/side with the decoder's +//! stereo prediction (§4.2.8). No LBRR, no NLSF interpolation. + +pub(crate) mod analysis; +pub(crate) mod lpc; +mod nlsf; +mod pulses; +#[allow(dead_code)] +pub(crate) mod tables; + +use lpc::{MAX_LPC_ORDER, div32_varq, inverse32_varq, nlsf2a, rshift_round, sat16, smlawb, smulwb, smulww}; +use tables::{ + DELTA_GAIN, GAIN_MSB, LTP_FILTER, LTP_PERIODICITY, LTP_SCALE, LTP_TAPS_0, LTP_TAPS_1, LTP_TAPS_2, MID_ONLY, NLSF_EXT, NLSF_INTERP, NLSF_SEL_NBMB, + NLSF_SEL_WB, NLSF_STAGE1, NLSF_STAGE2, PITCH_CONTOUR, PITCH_HIGH, PITCH_LOW, STEREO_PRED_JOINT, STEREO_WEIGHTS_Q13, TYPE_ACTIVE, TYPE_INACTIVE, + UNIFORM3, UNIFORM4, UNIFORM5, UNIFORM8, +}; + +use crate::opus::encoder::range::RangeEncoder; + +const MAX_FRAME: usize = 320; +const LTP_ORDER: usize = 5; +/// Frame RMS (16-bit scale) below which a frame is coded as inactive (VAD off). +const VAD_RMS: f32 = 30.0; +/// Normalised pitch correlation above which a frame is a voicing candidate. +const VOICING_THRESHOLD: f64 = 0.45; +/// Minimum LTP prediction gain (energy ratio) for a frame to be coded as voiced. +const MIN_LTP_GAIN: f64 = 1.3; +/// Rate penalty (squared pulse units per unit magnitude) of the excitation quantiser. +const RD_LAMBDA: f64 = 0.15; +/// Extra gain offset (log2) for inactive frames. +const INACTIVE_OFFSET: f64 = 1.0; +/// Noise feedback: the coding noise is shaped by `1 / A(z/γ)` (A the short-term predictor). +/// γ = 0 leaves it white (closed-loop DPCM, best waveform SNR at high rates), γ = 1 shapes it +/// like the spectral envelope (open-loop residual coding, needed at low rates). γ grows with the +/// gain offset, i.e. as the quantisation gets coarser. +fn nf_gamma(offset: f64) -> f64 { + (0.75 + 0.05 * offset).clamp(0.75, 0.9) +} + +/// `silk_log2lin` (decoder arithmetic). +fn log2lin(in_log_q7: i32) -> i32 { + if in_log_q7 < 0 { + return 0; + } + if in_log_q7 >= 3967 { + return i32::MAX; + } + let out = 1i32 << (in_log_q7 >> 7); + let frac = in_log_q7 & 0x7F; + if in_log_q7 < 2048 { + out + ((out * smlawb(frac, frac * (128 - frac), -174)) >> 7) + } else { + out + (out >> 7) * smlawb(frac, frac * (128 - frac), -174) + } +} + +fn gain_q16(index: i32) -> i32 { + log2lin((smulwb(0x1D1C71, index) + 2090).min(3967)) +} + +/// The decoder's LPC analysis filter on 16-bit samples. +fn lpc_analysis_filter(out: &mut [i16], input: &[i16], b: &[i32], d: usize) { + let len = out.len(); + for ix in d..len { + let mut acc: i32 = 0; + for j in 0..d { + acc = acc.wrapping_add((input[ix - 1 - j] as i32).wrapping_mul(b[j])); + } + let v = ((input[ix] as i32) << 12).wrapping_sub(acc); + out[ix] = sat16(rshift_round(v, 12)) as i16; + } + for v in out.iter_mut().take(d.min(len)) { + *v = 0; + } +} + +/// The part of the decoder's channel state that the reconstruction depends on. +#[derive(Clone)] +struct DecState { + prev_gain_q16: i32, + s_lpc_q14: [i32; MAX_LPC_ORDER], + out_buf: [i16; MAX_FRAME], + last_gain_index: i32, +} + +impl DecState { + fn new() -> Self { + DecState { prev_gain_q16: 65536, s_lpc_q14: [0; MAX_LPC_ORDER], out_buf: [0; MAX_FRAME], last_gain_index: 10 } + } + + /// The decoder's reset of the side channel when it is coded again after mid-only frames. + fn reset_side(&mut self) { + self.out_buf = [0; MAX_FRAME]; + self.s_lpc_q14 = [0; MAX_LPC_ORDER]; + self.last_gain_index = 10; + } +} + +/// Coded parameters of one frame (everything except gains and pulses). +#[derive(Clone, Default)] +struct Params { + vad: bool, + signal_type: usize, + quant_offset_type: usize, + nlsf_ix: [i32; 17], + a_q12: [i32; MAX_LPC_ORDER], + lag_index: i32, + contour: usize, + per_index: usize, + ltp_index: [usize; 4], + pitch_l: [i32; 4], + /// RMS (16-bit scale) of the open-loop prediction residual per subframe. + res_rms: [f32; 4], +} + +/// Gains, seed and pulses chosen by the quantiser. +struct Coded { + gain_ix: [i32; 4], + gains_q16: [i32; 4], + seed: usize, + pulses: Vec, +} + +/// Noise-feedback history (decoded − input), most recent first, as a double-length ring: each +/// sample is stored at `pos` and `pos + MAX_LPC_ORDER`, so `buf[pos..pos + MAX_LPC_ORDER]` is +/// always the contiguous window and a push costs two stores instead of a shift. +struct NfRing { + buf: [f64; 2 * MAX_LPC_ORDER], + /// Index of the most recent sample, always below `MAX_LPC_ORDER`. + pos: usize, +} + +impl NfRing { + fn new(hist: &[f64; MAX_LPC_ORDER]) -> Self { + let mut buf = [0.0; 2 * MAX_LPC_ORDER]; + buf[..MAX_LPC_ORDER].copy_from_slice(hist); + buf[MAX_LPC_ORDER..].copy_from_slice(hist); + NfRing { buf, pos: 0 } + } + + /// The history, most recent first. + fn window(&self) -> &[f64] { + self.buf.get(self.pos..self.pos + MAX_LPC_ORDER).unwrap_or(&[]) + } + + /// Makes `v` the most recent sample and drops the oldest. + fn push(&mut self, v: f64) { + self.pos = self.pos.checked_sub(1).unwrap_or(MAX_LPC_ORDER - 1); + for i in [self.pos, self.pos + MAX_LPC_ORDER] { + if let Some(slot) = self.buf.get_mut(i) { + *slot = v; + } + } + } + + /// The history as a plain array, most recent first. + fn to_array(&self) -> [f64; MAX_LPC_ORDER] { + let mut out = [0.0; MAX_LPC_ORDER]; + for (o, v) in out.iter_mut().zip(self.window()) { + *o = *v; + } + out + } +} + +#[derive(Clone)] +struct Channel { + fs_khz: usize, + len: usize, + sub: usize, + order: usize, + dec: DecState, + /// The last two frames of input (internal rate, 16-bit scale). + hist: Vec, + /// Recent output errors (decoded − input), most recent first, for noise feedback. + nf: [f64; MAX_LPC_ORDER], +} + +impl Channel { + fn new(fs_khz: usize) -> Self { + let len = 20 * fs_khz; + Channel { + fs_khz, + len, + sub: 5 * fs_khz, + order: if fs_khz == 16 { 16 } else { 10 }, + dec: DecState::new(), + hist: vec![0.0; 2 * len], + nf: [0.0; MAX_LPC_ORDER], + } + } + + fn push_history(&mut self, x: &[f32]) { + let l = self.len; + self.hist.copy_within(l.., 0); + self.hist[l..].copy_from_slice(x); + } + + /// Frame analysis: VAD, signal type, pitch, LPC/NLSF quantisation and LTP taps. + fn analyze(&self, x: &[f32]) -> Params { + let l = self.len; + let sub = self.sub; + let order = self.order; + let wb = order == 16; + let mut buf = self.hist.clone(); + buf.extend_from_slice(x); + let start = 2 * l; + let rms = (x.iter().map(|v| v * v).sum::() / l as f32).sqrt(); + let vad = rms > VAD_RMS; + let pitch = if vad { Some(analysis::pitch_search(&buf, start, l, self.fs_khz)) } else { None }; + let mut voiced = pitch.as_ref().is_some_and(|p| p.score > VOICING_THRESHOLD); + + let a = analysis::lpc(&buf[start - l / 4..], order, self.fs_khz); + let target: Vec = match analysis::a2nlsf(&a) { + Some(w) => w.iter().map(|v| v / std::f64::consts::PI * 32768.0).collect(), + None => (1..=order).map(|k| k as f64 * 32768.0 / (order + 1) as f64).collect(), + }; + let quant = |signal_type: usize| { + let (ix, nlsf) = nlsf::quantize(&target, signal_type, wb); + (ix, nlsf2a(&nlsf[..order])) + }; + let mut p = Params { vad, ..Default::default() }; + let (mut ix, mut a_q12) = quant(if voiced { 2 } else { 1 }); + // Open-loop residual with the quantised predictor. + let mut r = vec![0f64; buf.len()]; + for n in order..buf.len() { + let mut acc = buf[n] as f64; + for j in 0..order { + acc -= a_q12[j] as f64 / 4096.0 * buf[n - 1 - j] as f64; + } + r[n] = acc; + } + if voiced && let Some(pitch) = &pitch { + let ltp = analysis::ltp_search(&r, start, sub, &pitch.lags); + let gain = ltp.energy.iter().sum::() / ltp.err.iter().sum::().max(1e-9); + if gain >= MIN_LTP_GAIN { + p.lag_index = pitch.lag_index; + p.contour = pitch.contour; + p.pitch_l = pitch.lags; + p.per_index = ltp.per_index; + p.ltp_index = ltp.ltp_index; + } else { + voiced = false; + (ix, a_q12) = quant(1); + } + } + // Gains follow the short-term residual level (the long-term predictor then only + // reduces the number of pulses), which keeps the noise-to-signal ratio steady. + for k in 0..4 { + let e: f64 = r[start + k * sub..start + (k + 1) * sub].iter().map(|v| v * v).sum(); + p.res_rms[k] = (e / sub as f64).sqrt() as f32; + } + p.signal_type = if !vad { + 0 + } else if voiced { + 2 + } else { + 1 + }; + p.quant_offset_type = 0; + p.nlsf_ix = ix; + p.a_q12 = a_q12; + p + } + + /// Chooses gains and pulses for target `x` by running the decoder's reconstruction, and + /// advances the mirrored decoder state. `offset` scales the gains (log2), `floor` gives + /// minimum gains (Q16) per subframe. Returns the coded values and the decoded frame. + fn quantize(&mut self, p: &Params, x: &[f32], offset: f64, floor: Option<&[i32; 4]>, seed: usize) -> (Coded, Vec) { + let l = self.len; + let sub = self.sub; + let order = self.order; + let st = &mut self.dec; + let mut ring = NfRing::new(&self.nf); + let off = if p.signal_type == 0 { offset + INACTIVE_OFFSET } else { offset }; + let gamma = nf_gamma(off); + let mut nf_c = [0f64; MAX_LPC_ORDER]; + let mut gpow = 1.0; + for j in 0..order { + gpow *= gamma; + nf_c[j] = p.a_q12[j] as f64 / 4096.0 * gpow; + } + // Gains (independent coding of the first subframe, deltas for the rest). + let mut gain_ix = [0i32; 4]; + let mut gains_q16 = [0i32; 4]; + let mut prev = st.last_gain_index; + for k in 0..4 { + let mut g = p.res_rms[k] as f64 * off.exp2(); + if let Some(f) = floor { + g = g.max(f[k] as f64 / 65536.0); + } + let want = (((g * 65536.0).max(1.0).log2() * 128.0 - 2090.0) / (1_907_825.0 / 65536.0)).round().clamp(0.0, 63.0) as i32; + if k == 0 { + gain_ix[0] = want; + prev = want.max(prev - 16).clamp(0, 63); + } else { + let mut best = (i32::MAX, 0i32, prev); + for ind in 0..41 { + let tmp = ind - 4; + let thr = 2 * 36 - 64 + prev; + let d = if tmp > thr { prev + (tmp << 1) - thr } else { prev + tmp }.clamp(0, 63); + let e = (d - want).abs(); + if e < best.0 { + best = (e, ind, d); + } + } + gain_ix[k] = best.1; + prev = best.2; + } + gains_q16[k] = gain_q16(prev); + } + st.last_gain_index = prev; + + // Analysis-by-synthesis excitation quantisation through the decoder's exact synthesis. + let voiced = p.signal_type == 2; + let offset_q10 = [[100, 240], [32, 100]][p.signal_type >> 1][p.quant_offset_type]; + let off16 = offset_q10 << 4; + let mut seed_v = seed as i32; + let mut pulses = vec![0i32; l]; + let mut xq = vec![0i16; l]; + let mut s_lpc = [0i32; MAX_FRAME / 4 + MAX_LPC_ORDER]; + s_lpc[..MAX_LPC_ORDER].copy_from_slice(&st.s_lpc_q14); + let mut s_ltp = [0i16; MAX_FRAME]; + let mut s_ltp_q15 = [0i32; 2 * MAX_FRAME]; + let ltp_mem = l; + let mut ltp_buf_idx = ltp_mem; + let a_q12 = &p.a_q12; + let ltp_scale_q14 = 15565; + for k in 0..4 { + let mut b_q14 = [0i32; LTP_ORDER]; + if voiced { + let ix = p.ltp_index[k]; + let taps = match p.per_index { + 0 => LTP_TAPS_0[ix], + 1 => LTP_TAPS_1[ix], + _ => LTP_TAPS_2[ix], + }; + for i in 0..LTP_ORDER { + b_q14[i] = taps[i] << 7; + } + } + let gain_q10 = gains_q16[k] >> 6; + let mut inv_gain_q31 = inverse32_varq(gains_q16[k], 47); + let gain_adj_q16 = if gains_q16[k] != st.prev_gain_q16 { + let g = div32_varq(st.prev_gain_q16, gains_q16[k], 16); + for v in s_lpc[..MAX_LPC_ORDER].iter_mut() { + *v = smulww(g, *v); + } + g + } else { + 1 << 16 + }; + st.prev_gain_q16 = gains_q16[k]; + let lag = p.pitch_l[k]; + if voiced { + let lag_u = lag as usize; + if k == 0 { + let start_idx = (ltp_mem as isize - lag as isize - order as isize - (LTP_ORDER / 2) as isize).max(0) as usize; + lpc_analysis_filter(&mut s_ltp[start_idx..ltp_mem], &st.out_buf[start_idx..ltp_mem], a_q12, order); + inv_gain_q31 = smulwb(inv_gain_q31, ltp_scale_q14) << 2; + for i in 0..(lag_u + LTP_ORDER / 2).min(ltp_mem) { + s_ltp_q15[ltp_buf_idx - i - 1] = smulwb(inv_gain_q31, s_ltp[ltp_mem - i - 1] as i32); + } + } else if gain_adj_q16 != 1 << 16 { + for i in 0..(lag_u + LTP_ORDER / 2).min(ltp_buf_idx) { + let q = ltp_buf_idx - i - 1; + s_ltp_q15[q] = smulww(gain_adj_q16, s_ltp_q15[q]); + } + } + } + let inv_g = 16_777_216.0 / gain_q10.max(1) as f64; + for i in 0..sub { + let n = k * sub + i; + seed_v = seed_v.wrapping_mul(196_314_165).wrapping_add(907_633_515); + let ltp_q14 = if voiced { + let base = ltp_buf_idx as isize - lag as isize + (LTP_ORDER / 2) as isize; + let g = |o: isize| -> i32 { if base - o >= 0 { s_ltp_q15[(base - o) as usize] } else { 0 } }; + let mut pred = 2i32; + for o in 0..LTP_ORDER { + pred = smlawb(pred, g(o as isize), b_q14[o]); + } + pred << 1 + } else { + 0 + }; + let mut lpc_pred = (order >> 1) as i32; + for j in 0..order { + lpc_pred = smlawb(lpc_pred, s_lpc[MAX_LPC_ORDER + i - 1 - j], a_q12[j]); + } + let lpc_q14 = lpc_pred.saturating_mul(16); + // Target excitation (Q14) for this sample. + let fb: f64 = nf_c.iter().zip(ring.window()).take(order).map(|(c, e)| c * e).sum(); + let tgt = (x[n] as f64 + fb).clamp(-32768.0, 32767.0) * inv_g - lpc_q14 as f64 - ltp_q14 as f64; + let sgn = if seed_v < 0 { -1.0 } else { 1.0 }; + let u = sgn * tgt; + let pf = (u - off16 as f64) / 16384.0; + let p0 = if pf > 0.0 { (pf + 0.078_125).round() } else { (pf - 0.078_125).round() } as i32; + let level = |q: i32| -> i32 { + let mut e = q << 14; + if e > 0 { + e -= 80 << 4; + } else if e < 0 { + e += 80 << 4; + } + e + off16 + }; + let mut best = (f64::MAX, 0i32); + for q in [p0 - 1, p0, p0 + 1, 0] { + let q = q.clamp(-pulses::MAX_PULSE, pulses::MAX_PULSE); + let d = (level(q) as f64 - u) / 16384.0; + let cost = d * d + RD_LAMBDA * q.abs() as f64; + if cost < best.0 { + best = (cost, q); + } + } + let q = best.1; + pulses[n] = q; + let mut e = level(q); + if seed_v < 0 { + e = -e; + } + seed_v = seed_v.wrapping_add(q); + let res = e.wrapping_add(ltp_q14); + if voiced { + s_ltp_q15[ltp_buf_idx] = res << 1; + ltp_buf_idx += 1; + } + let v = res.saturating_add(lpc_q14); + s_lpc[MAX_LPC_ORDER + i] = v; + xq[n] = sat16(rshift_round(smulww(v, gain_q10), 8)) as i16; + ring.push(xq[n] as f64 - x[n] as f64); + } + s_lpc.copy_within(sub..sub + MAX_LPC_ORDER, 0); + } + self.nf = ring.to_array(); + st.s_lpc_q14.copy_from_slice(&s_lpc[..MAX_LPC_ORDER]); + st.out_buf[..l].copy_from_slice(&xq); + (Coded { gain_ix, gains_q16, seed, pulses }, xq) + } + + /// Writes the frame's indices and pulses (`decode_indices` + `decode_pulses` order, coded + /// independently). + fn write(&self, enc: &mut RangeEncoder, p: &Params, c: &Coded) { + let ix = p.signal_type * 2 + p.quant_offset_type; + if p.vad { + enc.icdf(ix - 2, &TYPE_ACTIVE, 8); + } else { + enc.icdf(ix, &TYPE_INACTIVE, 8); + } + enc.icdf((c.gain_ix[0] >> 3) as usize, GAIN_MSB[p.signal_type], 8); + enc.icdf((c.gain_ix[0] & 7) as usize, &UNIFORM8, 8); + for k in 1..4 { + enc.icdf(c.gain_ix[k] as usize, &DELTA_GAIN, 8); + } + let wb = self.order == 16; + let i1 = p.nlsf_ix[0] as usize; + enc.icdf(i1, NLSF_STAGE1[(p.signal_type >> 1) + if wb { 2 } else { 0 }], 8); + for i in 0..self.order { + let cb = if wb { 8 + NLSF_SEL_WB[i1][i] as usize } else { NLSF_SEL_NBMB[i1][i] as usize }; + let v = p.nlsf_ix[i + 1] + 4; + if v <= 0 { + enc.icdf(0, NLSF_STAGE2[cb], 8); + enc.icdf((-v) as usize, &NLSF_EXT, 8); + } else if v >= 8 { + enc.icdf(8, NLSF_STAGE2[cb], 8); + enc.icdf((v - 8) as usize, &NLSF_EXT, 8); + } else { + enc.icdf(v as usize, NLSF_STAGE2[cb], 8); + } + } + // No NLSF interpolation. + enc.icdf(4, &NLSF_INTERP, 8); + if p.signal_type == 2 { + let half = (self.fs_khz >> 1) as i32; + enc.icdf((p.lag_index / half) as usize, &PITCH_HIGH, 8); + let low = match self.fs_khz { + 8 => PITCH_LOW[0], + 12 => PITCH_LOW[1], + _ => PITCH_LOW[2], + }; + enc.icdf((p.lag_index % half) as usize, low, 8); + enc.icdf(p.contour, if self.fs_khz == 8 { PITCH_CONTOUR[1] } else { PITCH_CONTOUR[3] }, 8); + enc.icdf(p.per_index, <P_PERIODICITY, 8); + for k in 0..4 { + enc.icdf(p.ltp_index[k], LTP_FILTER[p.per_index], 8); + } + // LTP scaling index 0 (15565 in Q14), matching `quantize`. + enc.icdf(0, <P_SCALE, 8); + } + enc.icdf(c.seed, &UNIFORM4, 8); + pulses::encode(enc, &c.pulses, p.signal_type, p.quant_offset_type); + } +} + +/// Quantises a stereo prediction weight (Q13) to the decoder's grid (§4.2.7.1): returns the +/// (UNIFORM3, UNIFORM5, joint) indices and the reconstructed value. +fn quant_pred(w: f64) -> ([usize; 3], i32) { + let mut best = (f64::MAX, [0usize; 3], 0i32); + for i in 0..15 { + let low = STEREO_WEIGHTS_Q13[i]; + let step = smulwb(STEREO_WEIGHTS_Q13[i + 1] - low, 6554); + for j in 0..5 { + let v = low + step * (2 * j as i32 + 1); + let e = (v as f64 - w).abs(); + if e < best.0 { + best = (e, [i % 3, j, i / 3], v); + } + } + } + (best.1, best.2) +} + +/// Per-frame preparation shared by every rate-loop trial. +struct Prep { + x: [Vec; 2], + params: [Params; 2], + mid_only: bool, + pred_ix: [[usize; 3]; 2], + pred_q13: [i32; 2], +} + +/// One rate-loop trial: the frame coded at gain offset `offset`. +struct Trial { + offset: f64, + bits: i32, + state: SilkEncoder, + coder: RangeEncoder, +} + +/// SILK encoder for one mono or stereo (mid/side) stream. +#[derive(Clone)] +pub struct SilkEncoder { + fs_khz: usize, + channels: usize, + ch: [Channel; 2], + pred_prev_q13: [i32; 2], + prev_mid_only: bool, + /// Last mid sample of the previous frame (for the stereo predictor's low-pass). + prev_mid: f32, + frame: usize, + /// Gain offset chosen for the previous frame (search starting point). + pub last_offset: f64, +} + +impl SilkEncoder { + pub fn new(fs_khz: usize, channels: usize) -> Self { + SilkEncoder { + fs_khz, + channels, + ch: [Channel::new(fs_khz), Channel::new(fs_khz)], + pred_prev_q13: [0; 2], + prev_mid_only: false, + prev_mid: 0.0, + frame: 0, + last_offset: 1.0, + } + } + + /// Samples per channel of one 20 ms frame at the internal rate. + pub fn frame_len(&self) -> usize { + 20 * self.fs_khz + } + + fn prepare(&self, input: &[Vec]) -> Prep { + let l = self.frame_len(); + if self.channels == 1 { + let x = input[0].clone(); + let p = self.ch[0].analyze(&x); + return Prep { x: [x, vec![0.0; l]], params: [p, Params::default()], mid_only: true, pred_ix: [[0; 3]; 2], pred_q13: [0; 2] }; + } + let mid: Vec = (0..l).map(|i| 0.5 * (input[0][i] + input[1][i])).collect(); + let side: Vec = (0..l).map(|i| 0.5 * (input[0][i] - input[1][i])).collect(); + let lp = |j: usize| -> f64 { + let a = if j == 0 { self.prev_mid } else { mid[j - 1] }; + let c = if j + 1 < l { mid[j + 1] } else { mid[l - 1] }; + (a as f64 + 2.0 * mid[j] as f64 + c as f64) / 4.0 + }; + // Least-squares prediction of the side from the low-passed mid and the mid. + let (mut r00, mut r01, mut r11, mut c0, mut c1, mut ss, mut mm) = (0f64, 0f64, 0f64, 0f64, 0f64, 0f64, 0f64); + for j in 0..l { + let (a, b, s) = (lp(j), mid[j] as f64, side[j] as f64); + r00 += a * a; + r01 += a * b; + r11 += b * b; + c0 += a * s; + c1 += b * s; + ss += s * s; + mm += b * b; + } + // Ridge regularisation: the low-passed mid and the mid are nearly collinear. + let lam = 0.02 * 0.5 * (r00 + r11) + 1e-9; + let (a00, a11) = (r00 + lam, r11 + lam); + let det = a00 * a11 - r01 * r01; + let (p0, p1) = ((c0 * a11 - c1 * r01) / det, (c1 * a00 - c0 * r01) / det); + let lim = 13732.0; + let quant = |p0: f64, p1: f64| { + let (ix0, q0) = quant_pred(((p0 + p1) * 8192.0).clamp(-lim, lim)); + let (ix1, q1) = quant_pred((p1 * 8192.0).clamp(-lim, lim)); + ([ix0, ix1], [q0 - q1, q1]) + }; + // Side residual with the decoder's interpolated predictor. + let fs = self.fs_khz as i32; + let denom_q16 = (1 << 16) / (8 * fs); + let interp = 8 * self.fs_khz; + let residual = |pred_q13: [i32; 2]| -> Vec { + let d0 = rshift_round((pred_q13[0] - self.pred_prev_q13[0]) as i16 as i32 * denom_q16 as i16 as i32, 16); + let d1 = rshift_round((pred_q13[1] - self.pred_prev_q13[1]) as i16 as i32 * denom_q16 as i16 as i32, 16); + let (mut pp0, mut pp1) = (self.pred_prev_q13[0] + d0, self.pred_prev_q13[1] + d1); + (0..l) + .map(|j| { + if j + 1 < interp { + pp0 += d0; + pp1 += d1; + } else { + pp0 = pred_q13[0]; + pp1 = pred_q13[1]; + } + (side[j] as f64 - (pp0 as f64 * lp(j) + pp1 as f64 * mid[j] as f64) / 8192.0) as f32 + }) + .collect() + }; + let energy = |v: &[f32]| -> f64 { v.iter().map(|&x| (x as f64).powi(2)).sum() }; + let (mut pred_ix, mut pred_q13) = quant(p0, p1); + let mut res = residual(pred_q13); + let mut res_e = energy(&res); + // Never predict worse than not predicting at all. + if res_e > ss { + (pred_ix, pred_q13) = quant(0.0, 0.0); + res = residual(pred_q13); + res_e = energy(&res); + } + let res_rms = (res_e / l as f64).sqrt(); + let mid_rms = (mm / l as f64).sqrt(); + let mid_only = res_rms < 0.01 * mid_rms + 1.0; + let pm = self.ch[0].analyze(&mid); + let ps = if mid_only { Params::default() } else { self.ch[1].analyze(&res) }; + Prep { x: [mid, res], params: [pm, ps], mid_only, pred_ix, pred_q13 } + } + + /// Codes one frame into `enc` with gain offset `offset`. + fn code(&mut self, prep: &Prep, offset: f64, enc: &mut RangeEncoder) { + let stereo = self.channels == 2; + let side_vad = stereo && !prep.mid_only && prep.params[1].vad; + enc.bit_logp(prep.params[0].vad, 1); + enc.bit_logp(false, 1); + if stereo { + enc.bit_logp(side_vad, 1); + enc.bit_logp(false, 1); + let [a, b] = prep.pred_ix; + enc.icdf(5 * a[2] + b[2], &STEREO_PRED_JOINT, 8); + enc.icdf(a[0], &UNIFORM3, 8); + enc.icdf(a[1], &UNIFORM5, 8); + enc.icdf(b[0], &UNIFORM3, 8); + enc.icdf(b[1], &UNIFORM5, 8); + if !side_vad { + enc.icdf(prep.mid_only as usize, &MID_ONLY, 8); + } + } + let seed = self.frame & 3; + let (cm, _) = self.ch[0].quantize(&prep.params[0], &prep.x[0], offset, None, seed); + self.ch[0].write(enc, &prep.params[0], &cm); + if stereo { + if !prep.mid_only { + if self.prev_mid_only { + self.ch[1].dec.reset_side(); + } + let (cs, _) = self.ch[1].quantize(&prep.params[1], &prep.x[1], offset, Some(&cm.gains_q16), seed); + self.ch[1].write(enc, &prep.params[1], &cs); + } + self.pred_prev_q13 = prep.pred_q13; + self.prev_mid_only = prep.mid_only; + } + } + + /// Encodes one 20 ms frame (`input`: one internal-rate, 16-bit-scale vector per channel) + /// into `enc`, aiming at `target_bits` total and never exceeding `max_bits`. + pub fn encode(&mut self, input: &[Vec], target_bits: f64, max_bits: i32, enc: &mut RangeEncoder) { + let prep = self.prepare(input); + let start_bits = enc.tell(); + // Every trial codes the frame into a clone of the encoder and the range coder. The result + // is kept per offset: the hard-limit check below hits the memo for the chosen offset, and + // the final step adopts that trial instead of coding the frame once more. + let mut trials: Vec = Vec::new(); + let mut trial = |s: &Self, o: f64| -> i32 { + if let Some(t) = trials.iter().find(|t| t.offset == o) { + return t.bits; + } + let mut state = s.clone(); + let mut coder = enc.clone(); + state.code(&prep, o, &mut coder); + let bits = coder.tell() - start_bits; + trials.push(Trial { offset: o, bits, state, coder }); + bits + }; + // Bisection on the gain offset: the smallest offset (finest quantisation) within budget. + let target = target_bits.min(max_bits as f64); + // Bits are not strictly monotonic in the offset (very coarse quantisation feeds large + // errors back), so walk an integer grid from the previous frame's offset to find a + // bracket [infeasible, feasible], then bisect inside it. + const MIN_O: f64 = -3.0; + const MAX_O: f64 = 6.0; + let mut best_any = (i32::MAX, MAX_O); + let mut probe = |o: f64| -> bool { + let b = trial(self, o); + if b < best_any.0 { + best_any = (b, o); + } + b as f64 <= target + }; + let g0 = self.last_offset.round().clamp(MIN_O, MAX_O); + let mut bracket = None; + if probe(g0) { + let mut hi = g0; + loop { + if hi - 1.0 < MIN_O { + bracket = Some((None, hi)); + break; + } + if probe(hi - 1.0) { + hi -= 1.0; + } else { + bracket = Some((Some(hi - 1.0), hi)); + break; + } + } + } else { + let mut lo = g0; + while lo + 1.0 <= MAX_O { + if probe(lo + 1.0) { + bracket = Some((Some(lo), lo + 1.0)); + break; + } + lo += 1.0; + } + } + let mut o = match bracket { + Some((Some(mut lo), mut hi)) => { + for _ in 0..4 { + let m = 0.5 * (lo + hi); + if probe(m) { + hi = m; + } else { + lo = m; + } + } + hi + } + Some((None, hi)) => hi, + None => best_any.1, + }; + // Hard limit (hybrid packets must leave room for CELT). + while trial(self, o) > max_bits && o < 12.0 { + o += 1.0; + } + match trials.iter().position(|t| t.offset == o) { + Some(i) => { + let t = trials.swap_remove(i); + *self = t.state; + *enc = t.coder; + } + None => self.code(&prep, o, enc), + } + self.end_frame(&prep, o); + } + + /// Advances the input history and the frame counter after frame `prep` was coded at `o`. + fn end_frame(&mut self, prep: &Prep, o: f64) { + self.last_offset = o; + for ch in 0..self.channels { + self.ch[ch].push_history(&prep.x[ch]); + } + if self.channels == 2 { + self.prev_mid = prep.x[0][self.frame_len() - 1]; + } + self.frame += 1; + } +} + +#[cfg(test)] +mod tests { + use super::*; + + /// Deterministic speech-like test input: two harmonics with a moving pitch, a syllable + /// envelope, noise and short silences, at the internal rate and 16-bit scale. + fn signal(len: usize, fs_khz: usize, seed: u32) -> Vec { + let mut s = seed; + let fs = (fs_khz * 1000) as f64; + (0..len) + .map(|i| { + s = s.wrapping_mul(1_664_525).wrapping_add(1_013_904_223); + let noise = (s >> 8) as f64 / (1u32 << 24) as f64 - 0.5; + let t = i as f64 / fs; + let f0 = 140.0 + 40.0 * (t * 1.3).sin(); + let env = ((t * 4.0).sin().max(0.0)).powf(0.7); + let voiced = (std::f64::consts::TAU * f0 * t).sin() + 0.5 * (std::f64::consts::TAU * 2.0 * f0 * t).sin(); + (8000.0 * env * voiced + 600.0 * noise) as f32 + }) + .collect() + } + + #[test] + fn nf_ring_matches_a_shifted_history() { + let mut hist = [0f64; MAX_LPC_ORDER]; + for (i, h) in hist.iter_mut().enumerate() { + *h = i as f64 * 0.5 - 3.0; + } + let mut ring = NfRing::new(&hist); + assert_eq!(ring.window(), &hist[..]); + for n in 0..100 { + let v = (n as f64 * 0.37).sin() * 100.0; + hist.copy_within(0..MAX_LPC_ORDER - 1, 1); + hist[0] = v; + ring.push(v); + assert_eq!(ring.window(), &hist[..], "after push {n}"); + } + assert_eq!(ring.to_array(), hist); + } + + /// The rate loop adopts the memoised trial for the chosen offset. A second encoder that codes + /// every frame afresh at that offset (what the loop did before the memo) must produce the + /// same bytes, frame after frame, so the adopted encoder and range-coder states are exact. + #[test] + fn adopting_the_memoised_trial_equals_coding_the_frame_again() { + for (fs_khz, channels, target_bits, max_bits) in + [(8, 1, 200.0, 1000), (12, 1, 320.0, 1000), (16, 1, 480.0, 1000), (16, 2, 640.0, 1000), (16, 1, 900.0, 400), (8, 2, 120.0, 160)] + { + let mut memo = SilkEncoder::new(fs_khz, channels); + let mut fresh = memo.clone(); + let l = memo.frame_len(); + let frames = 40; + let left = signal(l * frames, fs_khz, 7); + let right: Vec = signal(l * frames + 37, fs_khz, 11)[37..].iter().map(|v| v * 0.6).collect(); + for f in 0..frames { + let mut input = vec![left[f * l..(f + 1) * l].to_vec()]; + if channels == 2 { + input.push(right[f * l..(f + 1) * l].to_vec()); + } + let mut enc = RangeEncoder::new(); + memo.encode(&input, target_bits, max_bits, &mut enc); + let o = memo.last_offset; + let prep = fresh.prepare(&input); + let mut enc_fresh = RangeEncoder::new(); + fresh.code(&prep, o, &mut enc_fresh); + fresh.end_frame(&prep, o); + assert_eq!(enc.tell(), enc_fresh.tell(), "{fs_khz} kHz {channels} ch frame {f}"); + assert_eq!(enc.done(1275), enc_fresh.done(1275), "{fs_khz} kHz {channels} ch frame {f}"); + for c in 0..channels { + assert_eq!(memo.ch[c].nf, fresh.ch[c].nf, "noise feedback, channel {c}, frame {f}"); + assert_eq!(memo.ch[c].dec.s_lpc_q14, fresh.ch[c].dec.s_lpc_q14, "LPC state, channel {c}, frame {f}"); + } + } + } + } +} diff --git a/crates/audio/src/opus/encoder/silk/nlsf.rs b/crates/audio/src/opus/encoder/silk/nlsf.rs new file mode 100644 index 000000000..59fca68cb --- /dev/null +++ b/crates/audio/src/opus/encoder/silk/nlsf.rs @@ -0,0 +1,131 @@ +//! NLSF quantisation (encoder side of RFC 6716 §4.2.7.5): stage-1 vector codebook preselection +//! and a backwards-predictive stage-2 scalar quantiser that reproduces the decoder's residual +//! dequantisation exactly. The search strategy is our own. + +use super::lpc::{MAX_LPC_ORDER, nlsf_decode, smlawb, sqrt_approx}; +use super::tables::{ + NLSF_CB_NBMB, NLSF_CB_WB, NLSF_EXT, NLSF_PRED_NBMB, NLSF_PRED_WB, NLSF_PSEL_NBMB, NLSF_PSEL_WB, NLSF_SEL_NBMB, NLSF_SEL_WB, NLSF_STAGE1, + NLSF_STAGE2, +}; + +/// Number of stage-1 candidates refined with stage 2. +const SURVIVORS: usize = 8; +/// Rate weight (weighted squared Q15 error per bit). +const LAMBDA: f64 = 60.0; + +fn icdf_bits(icdf: &[u8], s: usize) -> f64 { + let hi = if s == 0 { 256 } else { icdf[s - 1] as u32 }; + let p = (hi - icdf[s] as u32).max(1) as f64 / 256.0; + -p.log2() +} + +/// Bits of a stage-2 index `q` (-10..=10) with stage-2 table `tab`. +fn stage2_bits(tab: &[u8], q: i32) -> f64 { + let v = q + 4; + if v <= 0 { + icdf_bits(tab, 0) + icdf_bits(&NLSF_EXT, (-v) as usize) + } else if v >= 8 { + icdf_bits(tab, 8) + icdf_bits(&NLSF_EXT, (v - 8) as usize) + } else { + icdf_bits(tab, v as usize) + } +} + +/// Quantises target NLSFs (`target`, Q15 as f64, ascending) and returns the 1 + order indices +/// in the decoder's layout (`[stage1, stage2...]`) together with the reconstructed NLSFs. +pub fn quantize(target: &[f64], signal_type: usize, wb: bool) -> ([i32; 17], [i32; MAX_LPC_ORDER]) { + let order = target.len(); + // Laroia-style weights. + let w: Vec = (0..order) + .map(|k| { + let prev = if k == 0 { 0.0 } else { target[k - 1] }; + let next = if k + 1 == order { 32768.0 } else { target[k + 1] }; + 1.0 / (target[k] - prev).max(50.0) + 1.0 / (next - target[k]).max(50.0) + }) + .collect(); + let s1_tab = NLSF_STAGE1[(signal_type >> 1) + if wb { 2 } else { 0 }]; + let n_cb = 32; + let mut cands: Vec<(f64, usize)> = (0..n_cb) + .map(|i1| { + let cb: &[i32] = if wb { &NLSF_CB_WB[i1] } else { &NLSF_CB_NBMB[i1] }; + let d: f64 = (0..order).map(|k| w[k] * ((cb[k] << 7) as f64 - target[k]).powi(2)).sum(); + (d + LAMBDA * icdf_bits(s1_tab, i1), i1) + }) + .collect(); + cands.sort_by(|a, b| a.0.total_cmp(&b.0)); + let qstep = if wb { 9830 } else { 11796 }; + let mut best = (f64::MAX, [0i32; 17], [0i32; MAX_LPC_ORDER]); + for &(_, i1) in cands.iter().take(SURVIVORS) { + let cb1: &[i32] = if wb { &NLSF_CB_WB[i1] } else { &NLSF_CB_NBMB[i1] }; + let mut ix = [0i32; 17]; + ix[0] = i1 as i32; + let mut bits = icdf_bits(s1_tab, i1); + // Exact decoder weights per coefficient. + let w_q9: Vec = (0..order) + .map(|k| { + let prev = if k == 0 { 0 } else { cb1[k - 1] }; + let next = if k + 1 == order { 256 } else { cb1[k + 1] }; + let ww = (1024 / (cb1[k] - prev).max(1) + 1024 / (next - cb1[k]).max(1)).min(32767); + sqrt_approx(ww << 16).max(1) + }) + .collect(); + let mut out_q10 = 0i32; + for i in (0..order).rev() { + let pred = if i + 1 < order { + let sel = if wb { NLSF_PSEL_WB[i1][i] } else { NLSF_PSEL_NBMB[i1][i] } as usize; + let wp = if wb { NLSF_PRED_WB[sel][i] } else { NLSF_PRED_NBMB[sel][i] }; + (out_q10 * wp) >> 8 + } else { + 0 + }; + let cb = if wb { 8 + NLSF_SEL_WB[i1][i] as usize } else { NLSF_SEL_NBMB[i1][i] as usize }; + let tab = NLSF_STAGE2[cb]; + let want_q10 = (target[i] - (cb1[i] << 7) as f64) * w_q9[i] as f64 / 16384.0; + let step = qstep as f64 * 1024.0 / 65536.0; + let q0 = ((want_q10 - pred as f64) / step).round() as i32; + let mut bq = (f64::MAX, 0i32, 0i32); + for q in (q0 - 1).max(-10)..=(q0 + 1).min(10) { + let mut v = q << 10; + if v > 0 { + v -= 102; + } else if v < 0 { + v += 102; + } + let o = smlawb(pred, v, qstep); + let nl = (cb1[i] << 7) as f64 + ((o << 14) / w_q9[i]) as f64; + let cost = w[i] * (nl - target[i]).powi(2) + LAMBDA * stage2_bits(tab, q); + if cost < bq.0 { + bq = (cost, q, o); + } + } + ix[i + 1] = bq.1; + out_q10 = bq.2; + bits += stage2_bits(tab, bq.1); + } + let nlsf = nlsf_decode(&ix, wb); + let d: f64 = (0..order).map(|k| w[k] * (nlsf[k] as f64 - target[k]).powi(2)).sum(); + let cost = d + LAMBDA * bits; + if cost < best.0 { + best = (cost, ix, nlsf); + } + } + (best.1, best.2) +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn quantised_nlsfs_track_the_target() { + for wb in [false, true] { + let order = if wb { 16 } else { 10 }; + // A smooth, plausible NLSF vector. + let target: Vec = (0..order).map(|k| 32768.0 * (k as f64 + 0.8 + 0.3 * (k as f64).sin()) / (order as f64 + 1.0)).collect(); + let (ix, nlsf) = quantize(&target, 1, wb); + assert!(ix[1..=order].iter().all(|v| (-10..=10).contains(v))); + let max_err = (0..order).map(|k| (nlsf[k] as f64 - target[k]).abs()).fold(0.0, f64::max); + assert!(max_err < 1200.0, "wb={wb} max err {max_err}: {:?} vs {target:?}", &nlsf[..order]); + } + } +} diff --git a/crates/audio/src/opus/encoder/silk/pulses.rs b/crates/audio/src/opus/encoder/silk/pulses.rs new file mode 100644 index 000000000..7ec4fa6bf --- /dev/null +++ b/crates/audio/src/opus/encoder/silk/pulses.rs @@ -0,0 +1,113 @@ +//! Excitation coding (encoder side of RFC 6716 §4.2.7.8): rate level, per-block pulse counts +//! with LSB escapes, the recursive shell code, LSBs and signs, written in exactly the order the +//! decoder reads them. + +use super::tables::{EXC_LSB, EXC_SIGN, PULSE_COUNT, RATE_LEVEL, SHELL_2, SHELL_4, SHELL_8, SHELL_16}; +use crate::opus::encoder::range::RangeEncoder; + +/// Largest pulse magnitude the quantiser produces (keeps every 16-sample block codable). +pub const MAX_PULSE: i32 = 500; + +pub fn icdf_bits(icdf: &[u8], s: usize) -> f32 { + let hi = if s == 0 { 256 } else { icdf[s - 1] as u32 }; + let p = (hi - icdf[s] as u32).max(1) as f32 / 256.0; + -p.log2() +} + +fn count_table(nls: usize) -> &'static [u8] { + if nls == 10 { PULSE_COUNT[10] } else { PULSE_COUNT[9] } +} + +fn shell(enc: &mut RangeEncoder, v: &[i32], count: i32) { + let n = v.len(); + if n == 1 || count == 0 { + return; + } + let tab = match n { + 16 => SHELL_16[(count - 1) as usize], + 8 => SHELL_8[(count - 1) as usize], + 4 => SHELL_4[(count - 1) as usize], + _ => SHELL_2[(count - 1) as usize], + }; + let (a, b) = v.split_at(n / 2); + let left: i32 = a.iter().sum(); + enc.icdf(left as usize, tab, 8); + shell(enc, a, left); + shell(enc, b, count - left); +} + +/// Codes the frame's pulses (`len` a multiple of 16). +pub fn encode(enc: &mut RangeEncoder, pulses: &[i32], signal_type: usize, quant_offset_type: usize) { + let nblk = pulses.len() / 16; + let mut nls = vec![0usize; nblk]; + let mut sums = vec![0i32; nblk]; + let mut hi = vec![0i32; pulses.len()]; + for b in 0..nblk { + let blk = &pulses[b * 16..(b + 1) * 16]; + let mut s = 0usize; + loop { + let sum: i32 = blk.iter().map(|p| p.abs() >> s).sum(); + if sum <= 16 || s == 10 { + sums[b] = sum.min(16); + break; + } + s += 1; + } + nls[b] = s; + for (h, p) in hi[b * 16..(b + 1) * 16].iter_mut().zip(blk) { + *h = p.abs() >> s; + } + } + // Rate level with the cheapest pulse-count symbols. + let rl_tab = RATE_LEVEL[signal_type >> 1]; + let mut best = (f32::MAX, 0usize); + for rl in 0..9 { + let mut bits = icdf_bits(rl_tab, rl); + for b in 0..nblk { + bits += if nls[b] == 0 { icdf_bits(PULSE_COUNT[rl], sums[b] as usize) } else { icdf_bits(PULSE_COUNT[rl], 17) }; + } + if bits < best.0 { + best = (bits, rl); + } + } + let rl = best.1; + enc.icdf(rl, rl_tab, 8); + for b in 0..nblk { + if nls[b] == 0 { + enc.icdf(sums[b] as usize, PULSE_COUNT[rl], 8); + } else { + enc.icdf(17, PULSE_COUNT[rl], 8); + for j in 1..nls[b] { + enc.icdf(17, count_table(j), 8); + } + enc.icdf(sums[b] as usize, count_table(nls[b]), 8); + } + } + for b in 0..nblk { + if sums[b] > 0 { + shell(enc, &hi[b * 16..(b + 1) * 16], sums[b]); + } + } + for b in 0..nblk { + if nls[b] > 0 { + for p in &pulses[b * 16..(b + 1) * 16] { + let a = p.abs(); + for j in (0..nls[b]).rev() { + enc.icdf(((a >> j) & 1) as usize, &EXC_LSB, 8); + } + } + } + } + let base = 7 * (quant_offset_type + (signal_type << 1)); + for b in 0..nblk { + let p = sums[b] | ((nls[b] as i32) << 5); + if p > 0 { + let tab = EXC_SIGN[base + ((p & 0x1F) as usize).min(6)]; + for &q in &pulses[b * 16..(b + 1) * 16] { + if q != 0 { + enc.icdf((q > 0) as usize, tab, 8); + } + } + } + } +} diff --git a/crates/audio/src/opus/encoder/silk/tables.rs b/crates/audio/src/opus/encoder/silk/tables.rs new file mode 100644 index 000000000..a4789a414 --- /dev/null +++ b/crates/audio/src/opus/encoder/silk/tables.rs @@ -0,0 +1,543 @@ +//! SILK probability tables and codebooks, transcribed from RFC 6716 §4.2 (Tables 4–53). +//! +//! PDFs are written as in the RFC (`{...}/256`) and converted at compile time to the inverse-CDF +//! form consumed by the range coder. +//! +//! Copied from FilmCraft's first-party, spec-derived `filmcraft-opus` decoder (MIT OR +//! Apache-2.0) so encoder and decoder share exactly the same values. + +/// Converts a PDF (summing to 256) into an inverse CDF (`256 - cdf[i+1]`). +pub const fn icdf(pdf: &[u8; N]) -> [u8; N] { + let mut out = [0u8; N]; + let mut acc: u32 = 0; + let mut i = 0; + while i < N { + acc += pdf[i] as u32; + out[i] = (256 - acc) as u8; + i += 1; + } + out +} + +pub const STEREO_PRED_JOINT: [u8; 25] = icdf(&[7, 2, 1, 1, 1, 10, 24, 8, 1, 1, 3, 23, 92, 23, 3, 1, 1, 8, 24, 10, 1, 1, 1, 2, 7]); +pub const UNIFORM3: [u8; 3] = icdf(&[85, 86, 85]); +pub const UNIFORM5: [u8; 5] = icdf(&[51, 51, 52, 51, 51]); +pub const STEREO_WEIGHTS_Q13: [i32; 16] = [-13732, -10050, -8266, -7526, -6500, -5000, -2950, -820, 820, 2950, 5000, 6500, 7526, 8266, 10050, 13732]; +pub const MID_ONLY: [u8; 2] = icdf(&[192, 64]); +pub const TYPE_INACTIVE: [u8; 2] = icdf(&[26, 230]); +pub const TYPE_ACTIVE: [u8; 4] = icdf(&[24, 74, 148, 10]); +pub const GAIN_MSB: [&[u8]; 3] = + [&icdf::<8>(&[32, 112, 68, 29, 12, 1, 1, 1]), &icdf::<8>(&[2, 17, 45, 60, 62, 47, 19, 4]), &icdf::<8>(&[1, 3, 26, 71, 94, 50, 9, 2])]; +pub const UNIFORM8: [u8; 8] = icdf(&[32, 32, 32, 32, 32, 32, 32, 32]); +pub const DELTA_GAIN: [u8; 41] = + icdf(&[6, 5, 11, 31, 132, 21, 8, 4, 3, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]); +pub const NLSF_STAGE1: [&[u8]; 4] = [ + &icdf::<32>(&[44, 34, 30, 19, 21, 12, 11, 3, 3, 2, 16, 2, 2, 1, 5, 2, 1, 3, 3, 1, 1, 2, 2, 2, 3, 1, 9, 9, 2, 7, 2, 1]), + &icdf::<32>(&[1, 10, 1, 8, 3, 8, 8, 14, 13, 14, 1, 14, 12, 13, 11, 11, 12, 11, 10, 10, 11, 8, 9, 8, 7, 8, 1, 1, 6, 1, 6, 5]), + &icdf::<32>(&[31, 21, 3, 17, 1, 8, 17, 4, 1, 18, 16, 4, 2, 3, 1, 10, 1, 3, 16, 11, 16, 2, 2, 3, 2, 11, 1, 4, 9, 8, 7, 3]), + &icdf::<32>(&[1, 4, 16, 5, 18, 11, 5, 14, 15, 1, 3, 12, 13, 14, 14, 6, 14, 12, 2, 6, 1, 12, 12, 11, 10, 3, 10, 5, 1, 1, 1, 3]), +]; +pub const NLSF_STAGE2: [&[u8]; 16] = [ + &icdf::<9>(&[1, 1, 1, 15, 224, 11, 1, 1, 1]), + &icdf::<9>(&[1, 1, 2, 34, 183, 32, 1, 1, 1]), + &icdf::<9>(&[1, 1, 4, 42, 149, 55, 2, 1, 1]), + &icdf::<9>(&[1, 1, 8, 52, 123, 61, 8, 1, 1]), + &icdf::<9>(&[1, 3, 16, 53, 101, 74, 6, 1, 1]), + &icdf::<9>(&[1, 3, 17, 55, 90, 73, 15, 1, 1]), + &icdf::<9>(&[1, 7, 24, 53, 74, 67, 26, 3, 1]), + &icdf::<9>(&[1, 1, 18, 63, 78, 58, 30, 6, 1]), + &icdf::<9>(&[1, 1, 1, 9, 232, 9, 1, 1, 1]), + &icdf::<9>(&[1, 1, 2, 28, 186, 35, 1, 1, 1]), + &icdf::<9>(&[1, 1, 3, 42, 152, 53, 2, 1, 1]), + &icdf::<9>(&[1, 1, 10, 49, 126, 65, 2, 1, 1]), + &icdf::<9>(&[1, 4, 19, 48, 100, 77, 5, 1, 1]), + &icdf::<9>(&[1, 1, 14, 54, 100, 72, 12, 1, 1]), + &icdf::<9>(&[1, 1, 15, 61, 87, 61, 25, 4, 1]), + &icdf::<9>(&[1, 7, 21, 50, 77, 81, 17, 1, 1]), +]; +pub const NLSF_SEL_NBMB: [[u8; 10]; 32] = [ + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0], + [1, 3, 1, 2, 2, 1, 2, 1, 1, 1], + [2, 1, 1, 1, 1, 1, 1, 1, 1, 1], + [1, 2, 2, 2, 2, 1, 2, 1, 1, 1], + [2, 3, 3, 3, 3, 2, 2, 2, 2, 2], + [0, 5, 3, 3, 2, 2, 2, 2, 1, 1], + [0, 2, 2, 2, 2, 2, 2, 2, 2, 1], + [2, 3, 6, 4, 4, 4, 5, 4, 5, 5], + [2, 4, 5, 5, 4, 5, 4, 6, 4, 4], + [2, 4, 4, 7, 4, 5, 4, 5, 5, 4], + [4, 3, 3, 3, 2, 3, 2, 2, 2, 2], + [1, 5, 5, 6, 4, 5, 4, 5, 5, 5], + [2, 7, 4, 6, 5, 5, 5, 5, 5, 5], + [2, 7, 5, 5, 5, 5, 5, 6, 5, 4], + [3, 3, 5, 4, 4, 5, 4, 5, 4, 4], + [2, 3, 3, 5, 5, 4, 4, 4, 4, 4], + [2, 4, 4, 6, 4, 5, 4, 5, 5, 5], + [2, 5, 4, 6, 5, 5, 5, 4, 5, 4], + [2, 7, 4, 5, 4, 5, 4, 5, 5, 5], + [2, 5, 4, 6, 7, 6, 5, 6, 5, 4], + [3, 6, 7, 4, 6, 5, 5, 6, 4, 5], + [2, 7, 6, 4, 4, 4, 5, 4, 5, 5], + [4, 5, 5, 4, 6, 6, 5, 6, 5, 4], + [2, 5, 5, 6, 5, 6, 4, 6, 4, 4], + [4, 5, 5, 5, 3, 7, 4, 5, 5, 4], + [2, 3, 4, 5, 5, 6, 4, 5, 5, 4], + [2, 3, 2, 3, 3, 4, 2, 3, 3, 3], + [1, 1, 2, 2, 2, 2, 2, 3, 2, 2], + [4, 5, 5, 6, 6, 6, 5, 6, 4, 5], + [3, 5, 5, 4, 4, 4, 4, 3, 3, 2], + [2, 5, 3, 7, 5, 5, 4, 4, 5, 4], + [4, 4, 5, 4, 5, 6, 5, 6, 5, 4], +]; +pub const NLSF_SEL_WB: [[u8; 16]; 32] = [ + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], + [2, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 1, 1, 1, 0, 3], + [2, 5, 5, 3, 7, 4, 4, 5, 2, 5, 4, 5, 5, 4, 3, 3], + [0, 2, 1, 2, 2, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1], + [0, 6, 5, 4, 6, 4, 7, 5, 4, 4, 4, 5, 5, 4, 4, 3], + [0, 3, 5, 5, 4, 3, 3, 5, 3, 3, 3, 3, 3, 3, 2, 4], + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], + [0, 2, 6, 3, 7, 2, 5, 3, 4, 5, 5, 4, 3, 3, 2, 3], + [0, 6, 2, 6, 6, 4, 5, 4, 6, 5, 4, 4, 5, 3, 3, 3], + [2, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], + [0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1], + [2, 2, 3, 4, 5, 3, 3, 3, 3, 3, 3, 3, 2, 2, 1, 3], + [2, 2, 3, 3, 4, 3, 3, 3, 3, 3, 3, 3, 3, 2, 1, 3], + [3, 4, 4, 4, 6, 4, 4, 5, 3, 5, 4, 4, 5, 4, 3, 4], + [0, 6, 4, 5, 4, 7, 5, 2, 6, 5, 7, 4, 4, 3, 5, 3], + [0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, 0], + [1, 6, 5, 7, 5, 4, 5, 3, 4, 5, 4, 4, 4, 3, 3, 4], + [1, 3, 3, 4, 4, 3, 3, 5, 2, 3, 3, 5, 5, 5, 3, 4], + [2, 3, 3, 2, 2, 2, 3, 2, 1, 2, 1, 2, 1, 1, 1, 4], + [0, 2, 3, 5, 3, 3, 2, 2, 2, 1, 1, 0, 0, 0, 0, 0], + [3, 4, 3, 5, 3, 3, 2, 2, 1, 1, 1, 1, 1, 2, 2, 4], + [2, 6, 3, 7, 7, 4, 5, 4, 5, 3, 5, 3, 3, 2, 3, 3], + [2, 3, 5, 6, 6, 3, 5, 3, 4, 4, 3, 3, 3, 3, 2, 4], + [1, 3, 3, 4, 4, 4, 4, 3, 5, 5, 5, 3, 1, 1, 1, 1], + [2, 5, 3, 6, 6, 4, 7, 4, 4, 5, 3, 4, 4, 3, 3, 3], + [0, 6, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], + [0, 6, 6, 3, 5, 2, 5, 5, 3, 4, 4, 7, 7, 4, 4, 4], + [3, 3, 7, 3, 5, 4, 3, 3, 3, 2, 2, 3, 3, 3, 2, 3], + [0, 0, 1, 0, 0, 0, 2, 1, 2, 1, 1, 2, 2, 2, 1, 1], + [0, 3, 2, 5, 3, 3, 2, 3, 2, 1, 0, 0, 1, 0, 0, 1], + [3, 5, 5, 4, 7, 5, 3, 3, 2, 3, 2, 2, 1, 0, 1, 0], + [2, 3, 5, 3, 4, 3, 3, 3, 2, 1, 2, 6, 4, 0, 0, 0], +]; +pub const NLSF_EXT: [u8; 7] = icdf(&[156, 60, 24, 9, 4, 2, 1]); +pub const NLSF_PRED_NBMB: [[i32; 9]; 2] = [[179, 138, 140, 148, 151, 149, 153, 151, 163], [116, 67, 82, 59, 92, 72, 100, 89, 92]]; +pub const NLSF_PRED_WB: [[i32; 15]; 2] = [ + [175, 148, 160, 176, 178, 173, 174, 164, 177, 174, 196, 182, 198, 192, 182], + [68, 62, 66, 60, 72, 117, 85, 90, 118, 136, 151, 142, 160, 142, 155], +]; +pub const NLSF_PSEL_NBMB: [[u8; 9]; 32] = [ + [0, 1, 0, 0, 0, 0, 0, 0, 0], + [1, 0, 0, 0, 0, 0, 0, 0, 0], + [0, 0, 0, 0, 0, 0, 0, 0, 0], + [1, 1, 1, 0, 0, 0, 0, 1, 0], + [0, 1, 0, 0, 0, 0, 0, 0, 0], + [0, 1, 0, 0, 0, 0, 0, 0, 0], + [1, 0, 1, 1, 0, 0, 0, 1, 0], + [0, 1, 1, 0, 0, 1, 1, 0, 0], + [0, 0, 1, 1, 0, 1, 0, 1, 1], + [0, 0, 1, 1, 0, 0, 1, 1, 1], + [0, 0, 0, 0, 0, 0, 0, 0, 0], + [0, 1, 0, 1, 1, 1, 1, 1, 0], + [0, 1, 0, 1, 1, 1, 1, 1, 0], + [0, 1, 1, 1, 1, 1, 1, 1, 0], + [1, 0, 1, 1, 0, 1, 1, 1, 1], + [0, 1, 1, 1, 1, 1, 0, 1, 0], + [0, 0, 1, 1, 0, 1, 0, 1, 0], + [0, 0, 1, 1, 1, 0, 1, 1, 1], + [0, 1, 1, 0, 0, 1, 1, 1, 0], + [0, 0, 0, 1, 1, 1, 0, 1, 0], + [0, 1, 1, 0, 0, 1, 0, 1, 0], + [0, 1, 1, 0, 0, 0, 1, 1, 0], + [0, 0, 0, 0, 0, 1, 1, 1, 1], + [0, 0, 1, 1, 0, 0, 0, 1, 1], + [0, 0, 0, 1, 0, 1, 1, 1, 1], + [0, 1, 1, 1, 1, 1, 1, 1, 0], + [0, 0, 0, 0, 0, 0, 0, 0, 0], + [0, 0, 0, 0, 0, 0, 0, 0, 0], + [0, 0, 1, 0, 1, 1, 0, 1, 0], + [1, 0, 0, 1, 0, 0, 0, 0, 0], + [0, 0, 0, 1, 1, 0, 1, 0, 1], + [1, 0, 1, 1, 0, 1, 1, 1, 1], +]; +pub const NLSF_PSEL_WB: [[u8; 15]; 32] = [ + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1], + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], + [0, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 1, 0, 0], + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0], + [0, 1, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 1, 1, 0], + [0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], + [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0], + [0, 1, 1, 0, 0, 0, 1, 0, 1, 1, 1, 0, 1, 0, 1], + [0, 1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 1, 1, 1, 1], + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1], + [0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], + [0, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 0], + [0, 0, 1, 0, 0, 1, 0, 1, 0, 1, 0, 0, 1, 0, 0], + [0, 0, 0, 0, 1, 1, 0, 1, 0, 1, 1, 1, 1, 0, 0], + [0, 1, 0, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1, 1, 1], + [0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0], + [0, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 0, 1, 0, 0], + [0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 1, 1, 1, 0, 0], + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1], + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0], + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], + [0, 1, 0, 1, 0, 1, 1, 0, 1, 0, 1, 0, 1, 1, 0], + [0, 0, 1, 1, 1, 1, 0, 1, 1, 0, 0, 1, 1, 0, 0], + [0, 1, 1, 0, 1, 0, 1, 0, 1, 0, 0, 0, 0, 1, 0], + [0, 0, 0, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1], + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1], + [0, 1, 1, 0, 0, 0, 1, 1, 0, 0, 1, 1, 1, 1, 1], + [0, 0, 0, 0, 0, 1, 0, 1, 1, 1, 1, 0, 1, 1, 1], + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1], + [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1], + [1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0], + [0, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 1, 0], +]; +pub const NLSF_CB_NBMB: [[i32; 10]; 32] = [ + [12, 35, 60, 83, 108, 132, 157, 180, 206, 228], + [15, 32, 55, 77, 101, 125, 151, 175, 201, 225], + [19, 42, 66, 89, 114, 137, 162, 184, 209, 230], + [12, 25, 50, 72, 97, 120, 147, 172, 200, 223], + [26, 44, 69, 90, 114, 135, 159, 180, 205, 225], + [13, 22, 53, 80, 106, 130, 156, 180, 205, 228], + [15, 25, 44, 64, 90, 115, 142, 168, 196, 222], + [19, 24, 62, 82, 100, 120, 145, 168, 190, 214], + [22, 31, 50, 79, 103, 120, 151, 170, 203, 227], + [21, 29, 45, 65, 106, 124, 150, 171, 196, 224], + [30, 49, 75, 97, 121, 142, 165, 186, 209, 229], + [19, 25, 52, 70, 93, 116, 143, 166, 192, 219], + [26, 34, 62, 75, 97, 118, 145, 167, 194, 217], + [25, 33, 56, 70, 91, 113, 143, 165, 196, 223], + [21, 34, 51, 72, 97, 117, 145, 171, 196, 222], + [20, 29, 50, 67, 90, 117, 144, 168, 197, 221], + [22, 31, 48, 66, 95, 117, 146, 168, 196, 222], + [24, 33, 51, 77, 116, 134, 158, 180, 200, 224], + [21, 28, 70, 87, 106, 124, 149, 170, 194, 217], + [26, 33, 53, 64, 83, 117, 152, 173, 204, 225], + [27, 34, 65, 95, 108, 129, 155, 174, 210, 225], + [20, 26, 72, 99, 113, 131, 154, 176, 200, 219], + [34, 43, 61, 78, 93, 114, 155, 177, 205, 229], + [23, 29, 54, 97, 124, 138, 163, 179, 209, 229], + [30, 38, 56, 89, 118, 129, 158, 178, 200, 231], + [21, 29, 49, 63, 85, 111, 142, 163, 193, 222], + [27, 48, 77, 103, 133, 158, 179, 196, 215, 232], + [29, 47, 74, 99, 124, 151, 176, 198, 220, 237], + [33, 42, 61, 76, 93, 121, 155, 174, 207, 225], + [29, 53, 87, 112, 136, 154, 170, 188, 208, 227], + [24, 30, 52, 84, 131, 150, 166, 186, 203, 229], + [37, 48, 64, 84, 104, 118, 156, 177, 201, 230], +]; +pub const NLSF_CB_WB: [[i32; 16]; 32] = [ + [7, 23, 38, 54, 69, 85, 100, 116, 131, 147, 162, 178, 193, 208, 223, 239], + [13, 25, 41, 55, 69, 83, 98, 112, 127, 142, 157, 171, 187, 203, 220, 236], + [15, 21, 34, 51, 61, 78, 92, 106, 126, 136, 152, 167, 185, 205, 225, 240], + [10, 21, 36, 50, 63, 79, 95, 110, 126, 141, 157, 173, 189, 205, 221, 237], + [17, 20, 37, 51, 59, 78, 89, 107, 123, 134, 150, 164, 184, 205, 224, 240], + [10, 15, 32, 51, 67, 81, 96, 112, 129, 142, 158, 173, 189, 204, 220, 236], + [8, 21, 37, 51, 65, 79, 98, 113, 126, 138, 155, 168, 179, 192, 209, 218], + [12, 15, 34, 55, 63, 78, 87, 108, 118, 131, 148, 167, 185, 203, 219, 236], + [16, 19, 32, 36, 56, 79, 91, 108, 118, 136, 154, 171, 186, 204, 220, 237], + [11, 28, 43, 58, 74, 89, 105, 120, 135, 150, 165, 180, 196, 211, 226, 241], + [6, 16, 33, 46, 60, 75, 92, 107, 123, 137, 156, 169, 185, 199, 214, 225], + [11, 19, 30, 44, 57, 74, 89, 105, 121, 135, 152, 169, 186, 202, 218, 234], + [12, 19, 29, 46, 57, 71, 88, 100, 120, 132, 148, 165, 182, 199, 216, 233], + [17, 23, 35, 46, 56, 77, 92, 106, 123, 134, 152, 167, 185, 204, 222, 237], + [14, 17, 45, 53, 63, 75, 89, 107, 115, 132, 151, 171, 188, 206, 221, 240], + [9, 16, 29, 40, 56, 71, 88, 103, 119, 137, 154, 171, 189, 205, 222, 237], + [16, 19, 36, 48, 57, 76, 87, 105, 118, 132, 150, 167, 185, 202, 218, 236], + [12, 17, 29, 54, 71, 81, 94, 104, 126, 136, 149, 164, 182, 201, 221, 237], + [15, 28, 47, 62, 79, 97, 115, 129, 142, 155, 168, 180, 194, 208, 223, 238], + [8, 14, 30, 45, 62, 78, 94, 111, 127, 143, 159, 175, 192, 207, 223, 239], + [17, 30, 49, 62, 79, 92, 107, 119, 132, 145, 160, 174, 190, 204, 220, 235], + [14, 19, 36, 45, 61, 76, 91, 108, 121, 138, 154, 172, 189, 205, 222, 238], + [12, 18, 31, 45, 60, 76, 91, 107, 123, 138, 154, 171, 187, 204, 221, 236], + [13, 17, 31, 43, 53, 70, 83, 103, 114, 131, 149, 167, 185, 203, 220, 237], + [17, 22, 35, 42, 58, 78, 93, 110, 125, 139, 155, 170, 188, 206, 224, 240], + [8, 15, 34, 50, 67, 83, 99, 115, 131, 146, 162, 178, 193, 209, 224, 239], + [13, 16, 41, 66, 73, 86, 95, 111, 128, 137, 150, 163, 183, 206, 225, 241], + [17, 25, 37, 52, 63, 75, 92, 102, 119, 132, 144, 160, 175, 191, 212, 231], + [19, 31, 49, 65, 83, 100, 117, 133, 147, 161, 174, 187, 200, 213, 227, 242], + [18, 31, 52, 68, 88, 103, 117, 126, 138, 149, 163, 177, 192, 207, 223, 239], + [16, 29, 47, 61, 76, 90, 106, 119, 133, 147, 161, 176, 193, 209, 224, 240], + [15, 21, 35, 50, 61, 73, 86, 97, 110, 119, 129, 141, 175, 198, 218, 237], +]; +pub const NLSF_MIN_SPACING_NBMB: [i32; 11] = [250, 3, 6, 3, 3, 3, 4, 3, 3, 3, 461]; +pub const NLSF_MIN_SPACING_WB: [i32; 17] = [100, 3, 40, 3, 3, 3, 5, 14, 14, 10, 11, 3, 8, 9, 7, 3, 347]; +pub const NLSF_INTERP: [u8; 5] = icdf(&[13, 22, 29, 11, 181]); +pub const LSF_ORDER_NBMB: [usize; 10] = [0, 9, 6, 3, 4, 5, 8, 1, 2, 7]; +pub const LSF_ORDER_WB: [usize; 16] = [0, 15, 8, 7, 4, 11, 12, 3, 2, 13, 10, 5, 6, 9, 14, 1]; +pub const COS_Q12: [i32; 129] = [ + 4096, 4095, 4091, 4085, 4076, 4065, 4052, 4036, 4017, 3997, 3973, 3948, 3920, 3889, 3857, 3822, 3784, 3745, 3703, 3659, 3613, 3564, 3513, 3461, + 3406, 3349, 3290, 3229, 3166, 3102, 3035, 2967, 2896, 2824, 2751, 2676, 2599, 2520, 2440, 2359, 2276, 2191, 2106, 2019, 1931, 1842, 1751, 1660, + 1568, 1474, 1380, 1285, 1189, 1093, 995, 897, 799, 700, 601, 501, 401, 301, 201, 101, 0, -101, -201, -301, -401, -501, -601, -700, -799, -897, + -995, -1093, -1189, -1285, -1380, -1474, -1568, -1660, -1751, -1842, -1931, -2019, -2106, -2191, -2276, -2359, -2440, -2520, -2599, -2676, -2751, + -2824, -2896, -2967, -3035, -3102, -3166, -3229, -3290, -3349, -3406, -3461, -3513, -3564, -3613, -3659, -3703, -3745, -3784, -3822, -3857, + -3889, -3920, -3948, -3973, -3997, -4017, -4036, -4052, -4065, -4076, -4085, -4091, -4095, -4096, +]; +pub const PITCH_HIGH: [u8; 32] = icdf(&[3, 3, 6, 11, 21, 30, 32, 19, 11, 10, 12, 13, 13, 12, 11, 9, 8, 7, 6, 4, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1]); +pub const PITCH_LOW: [&[u8]; 3] = + [&icdf::<4>(&[64, 64, 64, 64]), &icdf::<6>(&[43, 42, 43, 43, 42, 43]), &icdf::<8>(&[32, 32, 32, 32, 32, 32, 32, 32])]; +pub const PITCH_DELTA: [u8; 21] = icdf(&[46, 2, 2, 3, 4, 6, 10, 15, 26, 38, 30, 22, 15, 10, 7, 6, 4, 4, 2, 2, 2]); +pub const PITCH_CONTOUR: [&[u8]; 4] = [ + &icdf::<3>(&[143, 50, 63]), + &icdf::<11>(&[68, 12, 21, 17, 19, 22, 30, 24, 17, 16, 10]), + &icdf::<12>(&[91, 46, 39, 19, 14, 12, 8, 7, 6, 5, 5, 4]), + &icdf::<34>(&[33, 22, 18, 16, 15, 14, 14, 13, 13, 10, 9, 9, 8, 6, 6, 6, 5, 4, 4, 4, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1]), +]; +pub const CB_NB_10: [[i32; 2]; 3] = [[0, 0], [1, 0], [0, 1]]; +pub const CB_NB_20: [[i32; 4]; 11] = [ + [0, 0, 0, 0], + [2, 1, 0, -1], + [-1, 0, 1, 2], + [-1, 0, 0, 1], + [-1, 0, 0, 0], + [0, 0, 0, 1], + [0, 0, 1, 1], + [1, 1, 0, 0], + [1, 0, 0, 0], + [0, 0, 0, -1], + [1, 0, 0, -1], +]; +pub const CB_WB_10: [[i32; 2]; 12] = [[0, 0], [0, 1], [1, 0], [-1, 1], [1, -1], [-1, 2], [2, -1], [-2, 2], [2, -2], [-2, 3], [3, -2], [-3, 3]]; +pub const CB_WB_20: [[i32; 4]; 34] = [ + [0, 0, 0, 0], + [0, 0, 1, 1], + [1, 1, 0, 0], + [-1, 0, 0, 0], + [0, 0, 0, 1], + [1, 0, 0, 0], + [-1, 0, 0, 1], + [0, 0, 0, -1], + [-1, 0, 1, 2], + [1, 0, 0, -1], + [-2, -1, 1, 2], + [2, 1, 0, -1], + [-2, 0, 0, 2], + [-2, 0, 1, 3], + [2, 1, -1, -2], + [-3, -1, 1, 3], + [2, 0, 0, -2], + [3, 1, 0, -2], + [-3, -1, 2, 4], + [-4, -1, 1, 4], + [3, 1, -1, -3], + [-4, -1, 2, 5], + [4, 2, -1, -3], + [4, 1, -1, -4], + [-5, -1, 2, 6], + [5, 2, -1, -4], + [-6, -2, 2, 6], + [-5, -2, 2, 5], + [6, 2, -1, -5], + [-7, -2, 3, 8], + [6, 2, -2, -6], + [5, 2, -2, -5], + [8, 3, -2, -7], + [-9, -3, 3, 9], +]; +pub const LTP_PERIODICITY: [u8; 3] = icdf(&[77, 80, 99]); +pub const LTP_FILTER: [&[u8]; 3] = [ + &icdf::<8>(&[185, 15, 13, 13, 9, 9, 6, 6]), + &icdf::<16>(&[57, 34, 21, 20, 15, 13, 12, 13, 10, 10, 9, 10, 9, 8, 7, 8]), + &icdf::<32>(&[15, 16, 14, 12, 12, 12, 11, 11, 11, 10, 9, 9, 9, 9, 8, 8, 8, 8, 7, 7, 6, 6, 5, 4, 5, 4, 4, 4, 3, 4, 3, 2]), +]; +pub const LTP_TAPS_0: [[i32; 5]; 8] = [ + [4, 6, 24, 7, 5], + [0, 0, 2, 0, 0], + [12, 28, 41, 13, -4], + [-9, 15, 42, 25, 14], + [1, -2, 62, 41, -9], + [-10, 37, 65, -4, 3], + [-6, 4, 66, 7, -8], + [16, 14, 38, -3, 33], +]; +pub const LTP_TAPS_1: [[i32; 5]; 16] = [ + [13, 22, 39, 23, 12], + [-1, 36, 64, 27, -6], + [-7, 10, 55, 43, 17], + [1, 1, 8, 1, 1], + [6, -11, 74, 53, -9], + [-12, 55, 76, -12, 8], + [-3, 3, 93, 27, -4], + [26, 39, 59, 3, -8], + [2, 0, 77, 11, 9], + [-8, 22, 44, -6, 7], + [40, 9, 26, 3, 9], + [-7, 20, 101, -7, 4], + [3, -8, 42, 26, 0], + [-15, 33, 68, 2, 23], + [-2, 55, 46, -2, 15], + [3, -1, 21, 16, 41], +]; +pub const LTP_TAPS_2: [[i32; 5]; 32] = [ + [-6, 27, 61, 39, 5], + [-11, 42, 88, 4, 1], + [-2, 60, 65, 6, -4], + [-1, -5, 73, 56, 1], + [-9, 19, 94, 29, -9], + [0, 12, 99, 6, 4], + [8, -19, 102, 46, -13], + [3, 2, 13, 3, 2], + [9, -21, 84, 72, -18], + [-11, 46, 104, -22, 8], + [18, 38, 48, 23, 0], + [-16, 70, 83, -21, 11], + [5, -11, 117, 22, -8], + [-6, 23, 117, -12, 3], + [3, -8, 95, 28, 4], + [-10, 15, 77, 60, -15], + [-1, 4, 124, 2, -4], + [3, 38, 84, 24, -25], + [2, 13, 42, 13, 31], + [21, -4, 56, 46, -1], + [-1, 35, 79, -13, 19], + [-7, 65, 88, -9, -14], + [20, 4, 81, 49, -29], + [20, 0, 75, 3, -17], + [5, -9, 44, 92, -8], + [1, -3, 22, 69, 31], + [-6, 95, 41, -12, 5], + [39, 67, 16, -4, 1], + [0, -6, 120, 55, -36], + [-13, 44, 122, 4, -24], + [81, 5, 11, 3, 7], + [2, 0, 9, 10, 88], +]; +pub const LTP_SCALE: [u8; 3] = icdf(&[128, 64, 64]); +pub const UNIFORM4: [u8; 4] = icdf(&[64, 64, 64, 64]); +pub const RATE_LEVEL: [&[u8]; 2] = [&icdf::<9>(&[15, 51, 12, 46, 45, 13, 33, 27, 14]), &icdf::<9>(&[33, 30, 36, 17, 34, 49, 18, 21, 18])]; +pub const PULSE_COUNT: [&[u8]; 11] = [ + &icdf::<18>(&[131, 74, 25, 8, 3, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]), + &icdf::<18>(&[58, 93, 60, 23, 7, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]), + &icdf::<18>(&[43, 51, 46, 33, 24, 16, 11, 8, 6, 3, 3, 3, 2, 1, 1, 2, 1, 2]), + &icdf::<18>(&[17, 52, 71, 57, 31, 12, 5, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]), + &icdf::<18>(&[6, 21, 41, 53, 49, 35, 21, 11, 6, 3, 2, 2, 1, 1, 1, 1, 1, 1]), + &icdf::<18>(&[7, 14, 22, 28, 29, 28, 25, 20, 17, 13, 11, 9, 7, 5, 4, 4, 3, 10]), + &icdf::<18>(&[2, 5, 14, 29, 42, 46, 41, 31, 19, 11, 6, 3, 2, 1, 1, 1, 1, 1]), + &icdf::<18>(&[1, 2, 4, 10, 19, 29, 35, 37, 34, 28, 20, 14, 8, 5, 4, 2, 2, 2]), + &icdf::<18>(&[1, 2, 2, 5, 9, 14, 20, 24, 27, 28, 26, 23, 20, 15, 11, 8, 6, 15]), + &icdf::<18>(&[1, 1, 1, 6, 27, 58, 56, 39, 25, 14, 10, 6, 3, 3, 2, 1, 1, 2]), + &icdf::<18>(&[2, 1, 6, 27, 58, 56, 39, 25, 14, 10, 6, 3, 3, 2, 1, 1, 2, 0]), +]; +pub const SHELL_16: [&[u8]; 16] = [ + &icdf::<2>(&[126, 130]), + &icdf::<3>(&[56, 142, 58]), + &icdf::<4>(&[25, 101, 104, 26]), + &icdf::<5>(&[12, 60, 108, 64, 12]), + &icdf::<6>(&[7, 35, 84, 87, 37, 6]), + &icdf::<7>(&[4, 20, 59, 86, 63, 21, 3]), + &icdf::<8>(&[3, 12, 38, 72, 75, 42, 12, 2]), + &icdf::<9>(&[2, 8, 25, 54, 73, 59, 27, 7, 1]), + &icdf::<10>(&[2, 5, 17, 39, 63, 65, 42, 18, 4, 1]), + &icdf::<11>(&[1, 4, 12, 28, 49, 63, 54, 30, 11, 3, 1]), + &icdf::<12>(&[1, 4, 8, 20, 37, 55, 57, 41, 22, 8, 2, 1]), + &icdf::<13>(&[1, 3, 7, 15, 28, 44, 53, 48, 33, 16, 6, 1, 1]), + &icdf::<14>(&[1, 2, 6, 12, 21, 35, 47, 48, 40, 25, 12, 5, 1, 1]), + &icdf::<15>(&[1, 1, 4, 10, 17, 27, 37, 47, 43, 33, 21, 9, 4, 1, 1]), + &icdf::<16>(&[1, 1, 1, 8, 14, 22, 33, 40, 43, 38, 28, 16, 8, 1, 1, 1]), + &icdf::<17>(&[1, 1, 1, 1, 13, 18, 27, 36, 41, 41, 34, 24, 14, 1, 1, 1, 1]), +]; +pub const SHELL_8: [&[u8]; 16] = [ + &icdf::<2>(&[127, 129]), + &icdf::<3>(&[53, 149, 54]), + &icdf::<4>(&[22, 105, 106, 23]), + &icdf::<5>(&[11, 61, 111, 63, 10]), + &icdf::<6>(&[6, 35, 86, 88, 36, 5]), + &icdf::<7>(&[4, 20, 59, 87, 62, 21, 3]), + &icdf::<8>(&[3, 13, 40, 71, 73, 41, 13, 2]), + &icdf::<9>(&[3, 9, 27, 53, 70, 56, 28, 9, 1]), + &icdf::<10>(&[3, 8, 19, 37, 57, 61, 44, 20, 6, 1]), + &icdf::<11>(&[3, 7, 15, 28, 44, 54, 49, 33, 17, 5, 1]), + &icdf::<12>(&[1, 7, 13, 22, 34, 46, 48, 38, 28, 14, 4, 1]), + &icdf::<13>(&[1, 1, 11, 22, 27, 35, 42, 47, 33, 25, 10, 1, 1]), + &icdf::<14>(&[1, 1, 6, 14, 26, 37, 43, 43, 37, 26, 14, 6, 1, 1]), + &icdf::<15>(&[1, 1, 4, 10, 20, 31, 40, 42, 40, 31, 20, 10, 4, 1, 1]), + &icdf::<16>(&[1, 1, 3, 8, 16, 26, 35, 38, 38, 35, 26, 16, 8, 3, 1, 1]), + &icdf::<17>(&[1, 1, 2, 6, 12, 21, 30, 36, 38, 36, 30, 21, 12, 6, 2, 1, 1]), +]; +pub const SHELL_4: [&[u8]; 16] = [ + &icdf::<2>(&[127, 129]), + &icdf::<3>(&[49, 157, 50]), + &icdf::<4>(&[20, 107, 109, 20]), + &icdf::<5>(&[11, 60, 113, 62, 10]), + &icdf::<6>(&[7, 36, 84, 87, 36, 6]), + &icdf::<7>(&[6, 24, 57, 82, 60, 23, 4]), + &icdf::<8>(&[5, 18, 39, 64, 68, 42, 16, 4]), + &icdf::<9>(&[6, 14, 29, 47, 61, 52, 30, 14, 3]), + &icdf::<10>(&[1, 15, 23, 35, 51, 50, 40, 30, 10, 1]), + &icdf::<11>(&[1, 1, 21, 32, 42, 52, 46, 41, 18, 1, 1]), + &icdf::<12>(&[1, 6, 16, 27, 36, 42, 42, 36, 27, 16, 6, 1]), + &icdf::<13>(&[1, 5, 12, 21, 31, 38, 40, 38, 31, 21, 12, 5, 1]), + &icdf::<14>(&[1, 3, 9, 17, 26, 34, 38, 38, 34, 26, 17, 9, 3, 1]), + &icdf::<15>(&[1, 3, 7, 14, 22, 29, 34, 36, 34, 29, 22, 14, 7, 3, 1]), + &icdf::<16>(&[1, 2, 5, 11, 18, 25, 31, 35, 35, 31, 25, 18, 11, 5, 2, 1]), + &icdf::<17>(&[1, 1, 4, 9, 15, 21, 28, 32, 34, 32, 28, 21, 15, 9, 4, 1, 1]), +]; +pub const SHELL_2: [&[u8]; 16] = [ + &icdf::<2>(&[128, 128]), + &icdf::<3>(&[42, 172, 42]), + &icdf::<4>(&[21, 107, 107, 21]), + &icdf::<5>(&[12, 60, 112, 61, 11]), + &icdf::<6>(&[8, 34, 86, 86, 35, 7]), + &icdf::<7>(&[8, 23, 55, 90, 55, 20, 5]), + &icdf::<8>(&[5, 15, 38, 72, 72, 36, 15, 3]), + &icdf::<9>(&[6, 12, 27, 52, 77, 47, 20, 10, 5]), + &icdf::<10>(&[6, 19, 28, 35, 40, 40, 35, 28, 19, 6]), + &icdf::<11>(&[4, 14, 22, 31, 37, 40, 37, 31, 22, 14, 4]), + &icdf::<12>(&[3, 10, 18, 26, 33, 38, 38, 33, 26, 18, 10, 3]), + &icdf::<13>(&[2, 8, 13, 21, 29, 36, 38, 36, 29, 21, 13, 8, 2]), + &icdf::<14>(&[1, 5, 10, 17, 25, 32, 38, 38, 32, 25, 17, 10, 5, 1]), + &icdf::<15>(&[1, 4, 7, 13, 21, 29, 35, 36, 35, 29, 21, 13, 7, 4, 1]), + &icdf::<16>(&[1, 2, 5, 10, 17, 25, 32, 36, 36, 32, 25, 17, 10, 5, 2, 1]), + &icdf::<17>(&[1, 2, 4, 7, 13, 21, 28, 34, 36, 34, 28, 21, 13, 7, 4, 2, 1]), +]; +pub const EXC_LSB: [u8; 2] = icdf(&[136, 120]); +pub const EXC_SIGN: [&[u8]; 42] = [ + &icdf::<2>(&[2, 254]), + &icdf::<2>(&[207, 49]), + &icdf::<2>(&[189, 67]), + &icdf::<2>(&[179, 77]), + &icdf::<2>(&[174, 82]), + &icdf::<2>(&[163, 93]), + &icdf::<2>(&[157, 99]), + &icdf::<2>(&[58, 198]), + &icdf::<2>(&[245, 11]), + &icdf::<2>(&[238, 18]), + &icdf::<2>(&[232, 24]), + &icdf::<2>(&[225, 31]), + &icdf::<2>(&[220, 36]), + &icdf::<2>(&[211, 45]), + &icdf::<2>(&[1, 255]), + &icdf::<2>(&[210, 46]), + &icdf::<2>(&[190, 66]), + &icdf::<2>(&[178, 78]), + &icdf::<2>(&[169, 87]), + &icdf::<2>(&[162, 94]), + &icdf::<2>(&[152, 104]), + &icdf::<2>(&[48, 208]), + &icdf::<2>(&[242, 14]), + &icdf::<2>(&[235, 21]), + &icdf::<2>(&[224, 32]), + &icdf::<2>(&[214, 42]), + &icdf::<2>(&[205, 51]), + &icdf::<2>(&[190, 66]), + &icdf::<2>(&[1, 255]), + &icdf::<2>(&[162, 94]), + &icdf::<2>(&[152, 104]), + &icdf::<2>(&[147, 109]), + &icdf::<2>(&[144, 112]), + &icdf::<2>(&[141, 115]), + &icdf::<2>(&[138, 118]), + &icdf::<2>(&[8, 248]), + &icdf::<2>(&[203, 53]), + &icdf::<2>(&[187, 69]), + &icdf::<2>(&[176, 80]), + &icdf::<2>(&[168, 88]), + &icdf::<2>(&[161, 95]), + &icdf::<2>(&[154, 102]), +]; +pub const LBRR_FLAGS: [&[u8]; 2] = [&icdf::<3>(&[53, 53, 150]), &icdf::<7>(&[41, 20, 29, 41, 15, 28, 82])]; diff --git a/crates/audio/src/opus/encoder/tables.rs b/crates/audio/src/opus/encoder/tables.rs new file mode 100644 index 000000000..1985202cc --- /dev/null +++ b/crates/audio/src/opus/encoder/tables.rs @@ -0,0 +1,123 @@ +//! Constant tables of the 48 kHz CELT mode (RFC 6716 §4.3, Tables 55–58 and §4.3.2). +//! +//! Transcribed from RFC 6716; copied from FilmCraft's spec-derived `filmcraft-opus` decoder +//! (first-party, MIT OR Apache-2.0) so both sides share the exact same values. + +/// Number of energy bands. +pub const NB_EBANDS: usize = 21; +/// Band edges in units of 2.5 ms MDCT bins (Table 55). +pub const EBANDS: [i32; NB_EBANDS + 1] = [0, 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, 28, 34, 40, 48, 60, 78, 100]; +/// Overlap of the low-overlap window (samples at 48 kHz). +pub const OVERLAP: usize = 120; +/// Largest LM (20 ms frames). +pub const MAX_LM: usize = 3; +/// Pre-emphasis coefficient. +pub const PREEMPH: f32 = 0.850_006_1; + +/// Static allocation table (Table 57), indexed `[quality][band]`, in 1/32 bit per MDCT bin. +pub const ALLOC: [[u8; NB_EBANDS]; 11] = transpose([ + [0, 90, 110, 118, 126, 134, 144, 152, 162, 172, 200], + [0, 80, 100, 110, 119, 127, 137, 145, 155, 165, 200], + [0, 75, 90, 103, 112, 120, 130, 138, 148, 158, 200], + [0, 69, 84, 93, 104, 114, 124, 132, 142, 152, 200], + [0, 63, 78, 86, 95, 103, 113, 123, 133, 143, 200], + [0, 56, 71, 80, 89, 97, 107, 117, 127, 137, 200], + [0, 49, 65, 75, 83, 91, 101, 111, 121, 131, 200], + [0, 40, 58, 70, 78, 85, 95, 105, 115, 125, 200], + [0, 34, 51, 65, 72, 78, 88, 98, 108, 118, 198], + [0, 29, 45, 59, 66, 72, 82, 92, 102, 112, 193], + [0, 20, 39, 53, 60, 66, 76, 86, 96, 106, 188], + [0, 18, 32, 47, 54, 60, 70, 80, 90, 100, 183], + [0, 10, 26, 40, 47, 54, 64, 74, 84, 94, 178], + [0, 0, 20, 31, 39, 47, 57, 67, 77, 87, 173], + [0, 0, 12, 23, 32, 41, 51, 61, 71, 81, 168], + [0, 0, 0, 15, 25, 35, 45, 55, 65, 75, 163], + [0, 0, 0, 4, 17, 29, 39, 49, 59, 69, 158], + [0, 0, 0, 0, 12, 23, 33, 43, 53, 63, 153], + [0, 0, 0, 0, 1, 16, 26, 36, 46, 56, 148], + [0, 0, 0, 0, 0, 10, 15, 20, 30, 45, 129], + [0, 0, 0, 0, 0, 1, 1, 1, 1, 20, 104], +]); + +const fn transpose(t: [[u8; 11]; NB_EBANDS]) -> [[u8; NB_EBANDS]; 11] { + let mut out = [[0u8; NB_EBANDS]; 11]; + let mut b = 0; + while b < NB_EBANDS { + let mut q = 0; + while q < 11 { + out[q][b] = t[b][q]; + q += 1; + } + b += 1; + } + out +} + +/// Mean band energy (log2 domain) removed before coarse quantisation. +pub const E_MEANS: [f32; 25] = [ + 6.4375, 6.25, 5.75, 5.3125, 5.0625, 4.8125, 4.5, 4.375, 4.875, 4.6875, 4.5625, 4.4375, 4.875, 4.625, 4.3125, 4.5, 4.375, 4.625, 4.75, 4.4375, + 3.75, 3.75, 3.75, 3.75, 3.75, +]; + +/// Inter-frame prediction coefficient (alpha) per LM, Q15. +pub const PRED_COEF: [f32; 4] = [29440.0 / 32768.0, 26112.0 / 32768.0, 21248.0 / 32768.0, 16384.0 / 32768.0]; +/// Inter-band prediction coefficient (beta) per LM for inter frames, Q15. +pub const BETA_COEF: [f32; 4] = [30147.0 / 32768.0, 22282.0 / 32768.0, 12124.0 / 32768.0, 6554.0 / 32768.0]; +/// Beta for intra frames. +pub const BETA_INTRA: f32 = 4915.0 / 32768.0; + +/// Laplace model parameters for coarse energy: `[LM][intra][2*band + {0: prob, 1: decay}]`. +pub const E_PROB_MODEL: [[[u8; 42]; 2]; 4] = [ + [ + [ + 72, 127, 65, 129, 66, 128, 65, 128, 64, 128, 62, 128, 64, 128, 64, 128, 92, 78, 92, 79, 92, 78, 90, 79, 116, 41, 115, 40, 114, 40, 132, + 26, 132, 26, 145, 17, 161, 12, 176, 10, 177, 11, + ], + [ + 24, 179, 48, 138, 54, 135, 54, 132, 53, 134, 56, 133, 55, 132, 55, 132, 61, 114, 70, 96, 74, 88, 75, 88, 87, 74, 89, 66, 91, 67, 100, 59, + 108, 50, 120, 40, 122, 37, 97, 43, 78, 50, + ], + ], + [ + [ + 83, 78, 84, 81, 88, 75, 86, 74, 87, 71, 90, 73, 93, 74, 93, 74, 109, 40, 114, 36, 117, 34, 117, 34, 143, 17, 145, 18, 146, 19, 162, 12, + 165, 10, 178, 7, 189, 6, 190, 8, 177, 9, + ], + [ + 23, 178, 54, 115, 63, 102, 66, 98, 69, 99, 74, 89, 71, 91, 73, 91, 78, 89, 86, 80, 92, 66, 93, 64, 102, 59, 103, 60, 104, 60, 117, 52, + 123, 44, 138, 35, 133, 31, 97, 38, 77, 45, + ], + ], + [ + [ + 61, 90, 93, 60, 105, 42, 107, 41, 110, 45, 116, 38, 113, 38, 112, 38, 124, 26, 132, 27, 136, 19, 140, 20, 155, 14, 159, 16, 158, 18, 170, + 13, 177, 10, 187, 8, 192, 6, 175, 9, 159, 10, + ], + [ + 21, 178, 59, 110, 71, 86, 75, 85, 84, 83, 91, 66, 88, 73, 87, 72, 92, 75, 98, 72, 105, 58, 107, 54, 115, 52, 114, 55, 112, 56, 129, 51, + 132, 40, 150, 33, 140, 29, 98, 35, 77, 42, + ], + ], + [ + [ + 42, 121, 96, 66, 108, 43, 111, 40, 117, 44, 123, 32, 120, 36, 119, 33, 127, 33, 134, 34, 139, 21, 147, 23, 152, 20, 158, 25, 154, 26, + 166, 21, 173, 16, 184, 13, 184, 10, 150, 13, 139, 15, + ], + [ + 22, 178, 63, 114, 74, 82, 84, 83, 92, 82, 103, 62, 96, 72, 96, 67, 101, 73, 107, 72, 113, 55, 118, 52, 125, 52, 118, 52, 117, 55, 135, + 49, 137, 39, 157, 32, 145, 29, 97, 33, 77, 40, + ], + ], +]; + +/// Inverse CDF for coarse energy when only a few bits remain. +pub const SMALL_ENERGY_ICDF: [u8; 3] = [2, 1, 0]; +/// Conservative log2 in 1/8 bits of the number of coded bands (intensity reservation). +pub const LOG2_FRAC_TABLE: [i32; 24] = [0, 8, 13, 16, 19, 21, 23, 24, 26, 27, 28, 29, 30, 31, 32, 32, 33, 34, 34, 35, 36, 36, 37, 37]; +/// TF resolution changes: `[LM][4*transient + 2*tf_select + tf_flag]`. +pub const TF_SELECT_TABLE: [[i8; 8]; 4] = + [[0, -1, 0, -1, 0, -1, 0, -1], [0, -1, 0, -2, 1, 0, 1, -1], [0, -2, 0, -3, 2, 0, 1, -1], [0, -2, 0, -3, 3, 0, 1, -1]]; +/// Spreading decision `{7, 2, 21, 2}/32`. +pub const SPREAD_ICDF: [u8; 4] = [25, 23, 2, 0]; +/// Allocation trim (Table 58). +pub const TRIM_ICDF: [u8; 11] = [126, 124, 119, 109, 87, 41, 19, 9, 4, 2, 0]; diff --git a/crates/audio/src/opus/fft.rs b/crates/audio/src/opus/fft.rs new file mode 100644 index 000000000..300c78d48 --- /dev/null +++ b/crates/audio/src/opus/fft.rs @@ -0,0 +1,130 @@ +//! Shared CELT forward complex FFT. + +#[derive(Clone, Copy, Default, Debug)] +pub struct Cpx { + pub re: f32, + pub im: f32, +} + +impl Cpx { + #[inline] + pub(crate) fn mul(self, o: Cpx) -> Cpx { + Cpx { re: self.re * o.re - self.im * o.im, im: self.re * o.im + self.im * o.re } + } + #[inline] + pub(crate) fn add(self, o: Cpx) -> Cpx { + Cpx { re: self.re + o.re, im: self.im + o.im } + } + #[inline] + pub(crate) fn sub(self, o: Cpx) -> Cpx { + Cpx { re: self.re - o.re, im: self.im - o.im } + } +} + +/// Mixed-radix (2, 3, 4, 5) forward complex FFT, unscaled (decimation in time, recursive). +pub struct Fft { + pub(crate) n: usize, + pub(crate) factors: Vec, + pub(crate) twiddle: Vec, +} + +impl Fft { + pub fn new(n: usize) -> Fft { + let mut factors = Vec::new(); + let mut m = n; + for p in [4usize, 2, 3, 5] { + while m.is_multiple_of(p) { + factors.push(p); + m /= p; + } + } + assert_eq!(m, 1, "unsupported FFT size {n}"); + let twiddle = (0..n) + .map(|k| { + let ph = -2.0 * std::f64::consts::PI * k as f64 / n as f64; + Cpx { re: ph.cos() as f32, im: ph.sin() as f32 } + }) + .collect(); + Fft { n, factors, twiddle } + } + + pub fn process(&self, data: &mut [Cpx], scratch: &mut Vec) { + scratch.clear(); + scratch.extend_from_slice(&data[..self.n]); + self.rec(&mut data[..self.n], scratch, 1, self.n, 0, 1); + } + + fn rec(&self, out: &mut [Cpx], inp: &[Cpx], in_stride: usize, n: usize, fi: usize, tw_stride: usize) { + let p = self.factors[fi]; + let m = n / p; + if m == 1 { + for (i, o) in out.iter_mut().enumerate().take(p) { + *o = inp[i * in_stride]; + } + } else { + for q in 0..p { + self.rec(&mut out[q * m..(q + 1) * m], &inp[q * in_stride..], in_stride * p, m, fi + 1, tw_stride * p); + } + } + // The twiddle index q * k * tw_stride is always below n, so no modulo is needed. + let tw = &self.twiddle; + match p { + 2 => { + for k in 0..m { + let a0 = out[k]; + let a1 = out[m + k].mul(tw[k * tw_stride]); + out[k] = a0.add(a1); + out[m + k] = a0.sub(a1); + } + } + 4 => { + for k in 0..m { + let t0 = out[k]; + let t1 = out[m + k].mul(tw[k * tw_stride]); + let t2 = out[2 * m + k].mul(tw[2 * k * tw_stride]); + let t3 = out[3 * m + k].mul(tw[3 * k * tw_stride]); + let (s0, d0) = (t0.add(t2), t0.sub(t2)); + let (s1, d1) = (t1.add(t3), t1.sub(t3)); + out[k] = s0.add(s1); + out[2 * m + k] = s0.sub(s1); + // The odd outputs combine d0 with -j * d1 = (d1.im, -d1.re) and its negative. + out[m + k] = Cpx { re: d0.re + d1.im, im: d0.im - d1.re }; + out[3 * m + k] = Cpx { re: d0.re - d1.im, im: d0.im + d1.re }; + } + } + 3 => { + // Imaginary part of exp(-2 pi i / 3) = -sqrt(3) / 2. + let h = tw[self.n / 3].im; + for k in 0..m { + let t0 = out[k]; + let t1 = out[m + k].mul(tw[k * tw_stride]); + let t2 = out[2 * m + k].mul(tw[2 * k * tw_stride]); + let (s, d) = (t1.add(t2), t1.sub(t2)); + let a = Cpx { re: t0.re - 0.5 * s.re, im: t0.im - 0.5 * s.im }; + out[k] = t0.add(s); + // j * h * d = (-h * d.im, h * d.re) + out[m + k] = Cpx { re: a.re - h * d.im, im: a.im + h * d.re }; + out[2 * m + k] = Cpx { re: a.re + h * d.im, im: a.im - h * d.re }; + } + } + _ => { + // Generic DFT butterfly (radix 5, once per transform). + let roots: [Cpx; 5] = std::array::from_fn(|q| if q < p { tw[q * (self.n / p)] } else { Cpx::default() }); + let mut t = [Cpx::default(); 5]; + for k in 0..m { + t[0] = out[k]; + for q in 1..p { + t[q] = out[q * m + k].mul(tw[q * k * tw_stride]); + } + for q2 in 0..p { + let mut acc = t[0]; + for q in 1..p { + acc = acc.add(t[q].mul(roots[(q * q2) % p])); + } + out[q2 * m + k] = acc; + } + } + } + } + } +} diff --git a/crates/audio/src/opus/mod.rs b/crates/audio/src/opus/mod.rs index cf2accd33..84e81ec8c 100644 --- a/crates/audio/src/opus/mod.rs +++ b/crates/audio/src/opus/mod.rs @@ -1,20 +1,35 @@ //! Clean-room pure-Rust Opus (RFC 6716, updated by RFC 8251; RFC 7845 `OpusHead`). //! -//! Decoding (feature `opus-decode`): [`Decoder`] (multistream, surround), [`StreamDecoder`] (one -//! elementary stream), [`OpusHead`] and the packet parser [`packet`]. From FilmCraft's -//! `filmcraft-opus` (storytold/filmcraft@8fda2f2). +//! - Decoding (feature `opus-decode`): [`Decoder`] (multistream, surround), [`StreamDecoder`] (one +//! elementary stream), [`OpusHead`] and the packet parser [`packet`]. From FilmCraft's +//! `filmcraft-opus` (storytold/filmcraft@8fda2f2). +//! - Encoding (feature `opus-encode`): [`Encoder`] (in [`encoder`]) (SILK, hybrid and CELT modes, 48 kHz, +//! 20 ms, mono or stereo). From EffectCraft's `effectcraft-opusenc` (storytold/effectcraft@c618429). //! -//! Public items keep their names, so FilmCraft can re-export this module. +//! Public items keep their names, so FilmCraft and EffectCraft can re-export this module. The two +//! halves share the FFT, resampling dot product and CELT allocation preparation; range coders and +//! mode-specific MDCT processing remain separate. [`Bandwidth`] is one type for both (`encoder::Bandwidth` is the same type); +//! the two `Mode` enums keep their apps' variant names and convert into each other. -// Index loops and long argument lists follow the specification's pseudo-code (FilmCraft allows -// these lints workspace-wide). +// Index loops and long argument lists follow the specification's pseudo-code (FilmCraft and +// EffectCraft allow these lints workspace-wide). #![allow(clippy::needless_range_loop, clippy::too_many_arguments)] +mod dot; +mod fft; +mod rate_common; + +#[cfg(feature = "opus-decode")] mod decode; +#[cfg(feature = "opus-encode")] +pub mod encoder; +#[cfg(feature = "opus-decode")] pub use decode::{Decoder, Error, Mode, OpusHead, Packet, Result, SAMPLE_RATES, StreamDecoder, Toc, packet}; +#[cfg(feature = "opus-encode")] +pub use encoder::Encoder; -/// Coded audio bandwidth (RFC 6716 §2, Table 1). +/// Coded audio bandwidth (RFC 6716 §2, Table 1), for the decoder and the encoder alike. #[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)] pub enum Bandwidth { /// 4 kHz (8 kHz sampling). @@ -29,9 +44,32 @@ pub enum Bandwidth { Fullband, } +#[cfg(all(feature = "opus-decode", feature = "opus-encode"))] +impl From for encoder::Mode { + fn from(m: Mode) -> Self { + match m { + Mode::SilkOnly => encoder::Mode::Silk, + Mode::Hybrid => encoder::Mode::Hybrid, + Mode::CeltOnly => encoder::Mode::Celt, + } + } +} + +#[cfg(all(feature = "opus-decode", feature = "opus-encode"))] +impl From for Mode { + fn from(m: encoder::Mode) -> Self { + match m { + encoder::Mode::Silk => Mode::SilkOnly, + encoder::Mode::Hybrid => Mode::Hybrid, + encoder::Mode::Celt => Mode::CeltOnly, + } + } +} + /// Channel order for Vorbis-ordered surround (mapping family 1, RFC 7845 §5.1.1.2): output channel /// → decoded channel, giving the WAV/SMPTE order (L R C LFE back… side…) the rest of craft-audio /// uses. From FilmCraft's `filmcraft-codecs`. +#[cfg(feature = "opus-decode")] pub fn wav_order(channels: usize) -> Option<&'static [usize]> { Some(match channels { 3 => &[0, 2, 1], diff --git a/crates/audio/src/opus/rate_common.rs b/crates/audio/src/opus/rate_common.rs new file mode 100644 index 000000000..86b27c90a --- /dev/null +++ b/crates/audio/src/opus/rate_common.rs @@ -0,0 +1,362 @@ +//! CELT pulse cache and allocation caps shared by encoder and decoder. + +#[cfg(not(feature = "opus-encode"))] +use super::decode::celt::tables::{ALLOC, EBANDS, LOG2_FRAC_TABLE, MAX_LM, NB_EBANDS}; +#[cfg(feature = "opus-encode")] +use super::encoder::tables::{ALLOC, EBANDS, LOG2_FRAC_TABLE, MAX_LM, NB_EBANDS}; + +const BITRES: u32 = 3; +const MAX_PSEUDO: i32 = 40; +const LOG_MAX_PSEUDO: i32 = 6; +const MAX_FINE_BITS: i32 = 8; +const FINE_OFFSET: i32 = 21; +const QTHETA_OFFSET: i32 = 4; +const QTHETA_OFFSET_TWOPHASE: i32 = 16; +fn ilog(x: u32) -> i32 { + (32 - x.leading_zeros()) as i32 +} + +/// Number of pulses represented by pseudo-pulse index `i`. +pub fn get_pulses(i: i32) -> i32 { + if i < 8 { i } else { (8 + (i & 7)) << ((i >> 3) - 1) } +} + +/// Conservative (rounded up) log2 of `val` with `frac` fractional bits. +pub fn log2_frac(mut val: u32, mut frac: u32) -> i32 { + let mut l = ilog(val); + if val & val.wrapping_sub(1) != 0 { + if l > 16 { + val = ((val - 1) >> (l - 16)) + 1; + } else { + val <<= 16 - l; + } + let mut l = (l - 1) << frac; + loop { + let b = (val >> 16) as i32; + l += b << frac; + val = (val + b as u32) >> b; + val = (val * val + 0x7FFF) >> 15; + if frac == 0 { + break; + } + frac -= 1; + } + l + (val > 0x8000) as i32 + } else { + l -= 1; + l << frac + } +} + +/// PVQ codebook size V(N, K) (saturating far above 2^32). +pub fn pvq_v(n: usize, k: usize) -> u128 { + if k == 0 { + return 1; + } + if n == 0 { + return 0; + } + let mut row = vec![0u128; k + 1]; + row[0] = 1; + for _ in 0..n { + let mut prev_diag = row[0]; + for kk in 1..=k { + let up = row[kk]; + let v = up.saturating_add(row[kk - 1]).saturating_add(prev_diag); + prev_diag = up; + row[kk] = v.min(u128::MAX >> 8); + } + row[0] = 1; + } + row[k] +} + +fn fits_in32(n: usize, k: i32) -> bool { + pvq_v(n, k as usize) <= u32::MAX as u128 +} + +/// Precomputed per-mode data. +pub struct Mode { + /// log2 of band width in 1/8 bits. + pub log_n: [i32; NB_EBANDS], + pub(crate) cache_index: Vec, + pub(crate) cache_bits: Vec, + pub(crate) caps: Vec, +} + +impl Mode { + pub fn new() -> Mode { + let mut log_n = [0i32; NB_EBANDS]; + for (i, l) in log_n.iter_mut().enumerate() { + *l = log2_frac((EBANDS[i + 1] - EBANDS[i]) as u32, BITRES); + } + let (cache_index, cache_bits) = compute_pulse_cache(MAX_LM as i32); + let mut m = Mode { log_n, cache_index, cache_bits, caps: Vec::new() }; + m.caps = compute_caps(&m); + m + } + + #[inline] + pub fn cache(&self, lm: i32, band: usize) -> &[u8] { + let idx = self.cache_index[((lm + 1) as usize) * NB_EBANDS + band]; + &self.cache_bits[idx.max(0) as usize..] + } + + pub fn bits2pulses(&self, band: usize, lm: i32, bits: i32) -> i32 { + let cache = self.cache(lm, band); + let mut lo = 0i32; + let mut hi = cache[0] as i32; + let bits = bits - 1; + for _ in 0..LOG_MAX_PSEUDO { + let mid = (lo + hi + 1) >> 1; + if cache[mid as usize] as i32 >= bits { + hi = mid; + } else { + lo = mid; + } + } + let lo_bits = if lo == 0 { -1 } else { cache[lo as usize] as i32 }; + if bits - lo_bits <= cache[hi as usize] as i32 - bits { lo } else { hi } + } + + pub fn pulses2bits(&self, band: usize, lm: i32, pulses: i32) -> i32 { + if pulses == 0 { 0 } else { self.cache(lm, band)[pulses as usize] as i32 + 1 } + } + + /// Per-band allocation caps for this frame (`init_caps`). + pub fn init_caps(&self, lm: usize, c: usize) -> [i32; NB_EBANDS] { + let mut cap = [0i32; NB_EBANDS]; + for (i, cp) in cap.iter_mut().enumerate() { + let n = (EBANDS[i + 1] - EBANDS[i]) << lm; + *cp = ((self.caps[NB_EBANDS * (2 * lm + c - 1) + i] as i32 + 64) * c as i32 * n) >> 2; + } + cap + } +} + +fn compute_pulse_cache(lm_max: i32) -> (Vec, Vec) { + let nb = NB_EBANDS; + let rows = (lm_max + 2) as usize; + let mut index = vec![-1i16; nb * rows]; + let mut entries: Vec<(usize, i32, usize)> = Vec::new(); + let mut curr = 0usize; + for i in 0..rows { + for j in 0..nb { + let n = (((EBANDS[j + 1] - EBANDS[j]) << i) >> 1) as usize; + index[i * nb + j] = -1; + 'search: for k in 0..=i { + for nn in 0..nb { + if k == i && nn >= j { + break; + } + if n == (((EBANDS[nn + 1] - EBANDS[nn]) << k) >> 1) as usize { + index[i * nb + j] = index[k * nb + nn]; + break 'search; + } + } + } + if index[i * nb + j] == -1 && n != 0 { + let mut k = 0; + while k < MAX_PSEUDO && fits_in32(n, get_pulses(k + 1)) { + k += 1; + } + index[i * nb + j] = curr as i16; + entries.push((n, k, curr)); + curr += (k + 1) as usize; + } + } + } + let mut bits = vec![0u8; curr]; + for &(n, k, off) in &entries { + let maxk = get_pulses(k) as usize; + let tmp = required_bits(n, maxk, BITRES); + for j in 1..=k { + bits[off + j as usize] = (tmp[get_pulses(j) as usize] - 1) as u8; + } + bits[off] = k as u8; + } + (index, bits) +} + +fn required_bits(n: usize, maxk: usize, frac: u32) -> Vec { + let mut out = vec![0i32; maxk + 1]; + if n == 1 { + for o in out.iter_mut().skip(1) { + *o = 1 << frac; + } + } else { + for (k, o) in out.iter_mut().enumerate().skip(1) { + *o = log2_frac(pvq_v(n, k) as u32, frac); + } + } + out +} + +fn compute_caps(m: &Mode) -> Vec { + let nb = NB_EBANDS; + let mut caps = vec![0u8; (MAX_LM + 1) * 2 * nb]; + for i in 0..=MAX_LM as i32 { + for c in 1..=2i32 { + for j in 0..nb { + let mut n0 = EBANDS[j + 1] - EBANDS[j]; + + let max_bits = if n0 << i == 1 { + (c * (1 + MAX_FINE_BITS)) << BITRES + } else { + let mut lm0 = 0; + if n0 > 2 { + n0 >>= 1; + lm0 -= 1; + } else if n0 <= 1 { + lm0 = i.min(1); + n0 <<= lm0; + } + let pcache = m.cache(lm0, j); + let mut mb = pcache[pcache[0] as usize] as i32 + 1; + let mut n = n0; + for k in 0..(i - lm0) { + mb <<= 1; + let offset = ((m.log_n[j] + ((lm0 + k) << BITRES)) >> 1) - QTHETA_OFFSET; + let num = 459 * ((2 * n - 1) * offset + mb); + let den = ((2 * n - 1) << 9) - 459; + let qb = ((num + (den >> 1)) / den).min(57); + mb += qb; + n <<= 1; + } + if c == 2 { + mb <<= 1; + let offset = ((m.log_n[j] + (i << BITRES)) >> 1) - if n == 2 { QTHETA_OFFSET_TWOPHASE } else { QTHETA_OFFSET }; + let ndof = 2 * n - 1 - (n == 2) as i32; + let p = if n == 2 { 512 } else { 487 }; + let num = p * (mb + ndof * offset); + let den = (ndof << 9) - p; + let qb = ((num + (den >> 1)) / den).min(if n == 2 { 64 } else { 61 }); + mb += qb; + } + let ndof = c * n + if c == 2 && n > 2 { 1 } else { 0 }; + let mut offset = ((m.log_n[j] + (i << BITRES)) >> 1) - FINE_OFFSET; + if n == 2 { + offset += (1 << BITRES) >> 2; + } + let num = mb + ndof * offset; + let den = (ndof - 1) << BITRES; + let qb = ((num + (den >> 1)) / den).min(MAX_FINE_BITS); + mb += (c * qb) << BITRES; + mb + }; + let w = c * ((EBANDS[j + 1] - EBANDS[j]) << i); + let v = (4 * max_bits / w) - 64; + caps[((i * 2 + c - 1) as usize) * nb + j] = v.clamp(0, 255) as u8; + } + } + } + caps +} + +/// Shared allocation search before encoder/decoder signalling diverges. +pub struct Bracket { + pub skip_start: usize, + pub bits1: [i32; NB_EBANDS], + pub bits2: [i32; NB_EBANDS], + pub thresh: [i32; NB_EBANDS], + pub total: i32, + pub skip_rsv: i32, + pub intensity_rsv: i32, + pub dual_stereo_rsv: i32, +} + +pub fn bracket( + start: usize, + end: usize, + offsets: &[i32; NB_EBANDS], + cap: &[i32; NB_EBANDS], + alloc_trim: i32, + total: i32, + c: usize, + lm: usize, +) -> Bracket { + let ci = c as i32; + let lmi = lm as i32; + let mut total = total.max(0); + let mut skip_start = start; + let skip_rsv = if total >= 1 << BITRES { 1 << BITRES } else { 0 }; + total -= skip_rsv; + let mut intensity_rsv = 0; + let mut dual_stereo_rsv = 0; + if c == 2 { + intensity_rsv = LOG2_FRAC_TABLE[end - start]; + if intensity_rsv > total { + intensity_rsv = 0; + } else { + total -= intensity_rsv; + dual_stereo_rsv = if total >= 1 << BITRES { 1 << BITRES } else { 0 }; + total -= dual_stereo_rsv; + } + } + let mut bits1 = [0i32; NB_EBANDS]; + let mut bits2 = [0i32; NB_EBANDS]; + let mut thresh = [0i32; NB_EBANDS]; + let mut trim_offset = [0i32; NB_EBANDS]; + for j in start..end { + let w = EBANDS[j + 1] - EBANDS[j]; + thresh[j] = (ci << BITRES).max(((3 * w) << lmi << BITRES) >> 4); + trim_offset[j] = (ci * w * (alloc_trim - 5 - lmi) * (end as i32 - j as i32 - 1) * (1 << (lmi + BITRES as i32))) >> 6; + if w << lmi == 1 { + trim_offset[j] -= ci << BITRES; + } + } + let nb_alloc = ALLOC.len() as i32; + let mut lo = 1; + let mut hi = nb_alloc - 1; + loop { + let mut done = false; + let mut psum = 0; + let mid = (lo + hi) >> 1; + for j in (start..end).rev() { + let n = EBANDS[j + 1] - EBANDS[j]; + let mut bitsj = (ci * n * (ALLOC[mid as usize][j] as i32)) << lmi >> 2; + if bitsj > 0 { + bitsj = (bitsj + trim_offset[j]).max(0); + } + bitsj += offsets[j]; + if bitsj >= thresh[j] || done { + done = true; + psum += bitsj.min(cap[j]); + } else if bitsj >= ci << BITRES { + psum += ci << BITRES; + } + } + if psum > total { + hi = mid - 1; + } else { + lo = mid + 1; + } + if lo > hi { + break; + } + } + hi = lo; + lo -= 1; + for j in start..end { + let n = EBANDS[j + 1] - EBANDS[j]; + let mut bits1j = (ci * n * (ALLOC[lo as usize][j] as i32)) << lmi >> 2; + let mut bits2j = if hi >= nb_alloc { cap[j] } else { (ci * n * (ALLOC[hi as usize][j] as i32)) << lmi >> 2 }; + if bits1j > 0 { + bits1j = (bits1j + trim_offset[j]).max(0); + } + if bits2j > 0 { + bits2j = (bits2j + trim_offset[j]).max(0); + } + if lo > 0 { + bits1j += offsets[j]; + } + bits2j += offsets[j]; + if offsets[j] > 0 { + skip_start = j; + } + bits2j = (bits2j - bits1j).max(0); + bits1[j] = bits1j; + bits2[j] = bits2j; + } + Bracket { skip_start, bits1, bits2, thresh, total, skip_rsv, intensity_rsv, dual_stereo_rsv } +} diff --git a/crates/audio/src/pcm_export.rs b/crates/audio/src/pcm_export.rs new file mode 100644 index 000000000..618ba281d --- /dev/null +++ b/crates/audio/src/pcm_export.rs @@ -0,0 +1,287 @@ +//! Audio-only files: WAV (RIFF, little-endian PCM; RF64 past 4 GiB) and AIFF (big-endian PCM), +//! 16, 24 or 32-bit integer or 32-bit float ([`SampleFormat`]), for streaming export. +//! Float WAV is `WAVE_FORMAT_IEEE_FLOAT` with a `fact` chunk; float AIFF is AIFF-C (`fl32`). +//! +//! The header is written first from the known length (`header`), then the samples in chunks +//! (`encode`) and the pad byte (`pad`), so an export streams to disk instead of building the +//! file in memory. + +/// PCM representation in an exported file. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum SampleFormat { + /// 16-bit signed integer. + Int16, + /// 24-bit signed integer. + Int24, + /// 32-bit signed integer. + Int32, + /// 32-bit IEEE float. + Float32, +} + +impl SampleFormat { + /// Bits per sample. + pub fn bits(self) -> u16 { + match self { + Self::Int16 => 16, + Self::Int24 => 24, + Self::Int32 | Self::Float32 => 32, + } + } + /// Bytes per sample. + pub fn bytes(self) -> usize { + usize::from(self.bits() / 8) + } + /// Whether this is a floating point format. + pub fn is_float(self) -> bool { + self == Self::Float32 + } + /// Clip and quantize one sample with symmetric integer full scale; NaN becomes silence. + pub fn quantise(self, s: f32) -> i32 { + let s = s.clamp(-1.0, 1.0); + match self { + Self::Int16 => (s * 32_767.0).round() as i32, + Self::Int24 => (s * 8_388_607.0).round() as i32, + Self::Int32 => (f64::from(s) * 2_147_483_647.0).round() as i32, + Self::Float32 => 0, + } + } + /// Sanitize a floating sample: NaN becomes silence, infinity becomes the largest finite value. + pub fn float(s: f32) -> f32 { + if s.is_nan() { 0.0 } else { s.clamp(f32::MIN, f32::MAX) } + } +} + +/// The container of an audio-only export. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum PcmContainer { + /// RIFF/WAVE or RF64. + Wav, + /// AIFF or AIFF-C. + Aiff, +} + +/// Bytes of sample data for `frames` frames, or None when that does not fit 64 bits. +fn data_len(frames: u64, channels: u16, format: SampleFormat) -> Option { + frames.checked_mul(u64::from(channels))?.checked_mul(format.bytes() as u64) +} + +/// The pad byte after an odd-sized data chunk (RIFF and AIFF chunks are word-aligned). +pub fn pad(frames: u64, channels: u16, format: SampleFormat) -> &'static [u8] { + if data_len(frames, channels, format).is_some_and(|n| n & 1 == 1) { &[0] } else { &[] } +} + +/// `KSDATAFORMAT_SUBTYPE_PCM` / `_IEEE_FLOAT` after their first two bytes (the format tag). +const SUBTYPE_TAIL: [u8; 14] = [0x00, 0x00, 0x00, 0x00, 0x10, 0x00, 0x80, 0x00, 0x00, 0xAA, 0x00, 0x38, 0x9B, 0x71]; + +/// AIFF-C format version 1 (`FVER`). +const AIFC_VERSION_1: u32 = 0xA280_5140; + +/// The AIFF-C compression name of `fl32`, as a Pascal string of even length. +const FL32_NAME: &[u8] = b"\x1532-bit floating point"; + +/// Everything before the samples of a file of `frames` frames. A WAV whose RIFF size would not +/// fit 32 bits (4 GiB: 24-bit 5.1 at 96 kHz after about 41 minutes) is written as RF64 (EBU Tech +/// 3306), with the 64-bit sizes in a `ds64` chunk. AIFF has no 64-bit form, so a longer AIFF is +/// an error naming WAV instead. +pub fn header(container: PcmContainer, channels: u16, sample_rate: u32, format: SampleFormat, frames: u64) -> Result, String> { + let too_long = || "the audio is too long to write".to_string(); + if channels == 0 { + return Err("audio needs at least one channel".into()); + } + if sample_rate == 0 { + return Err("audio needs a sample rate".into()); + } + let bits = format.bits(); + let bps = bits / 8; + let float = format.is_float(); + let data = data_len(frames, channels, format).ok_or_else(too_long)?; + let block_align = channels.checked_mul(bps).ok_or("too many audio channels")?; + match container { + PcmContainer::Wav => { + let byte_rate = + u32::try_from(u64::from(sample_rate) * u64::from(block_align)).map_err(|_| "the audio sample rate is too high".to_string())?; + let ext = channels > 2; + // plain PCM has no cbSize; IEEE float has an empty one + let fmt_len: u32 = if ext { + 40 + } else if float { + 18 + } else { + 16 + }; + // every format but integer PCM has a `fact` chunk (the frame count; RF64: in `ds64`) + let fact_len: u64 = if float { 12 } else { 0 }; + let padded = data.checked_add(data & 1).ok_or_else(too_long)?; + // "WAVE" + fmt chunk + fact chunk + data chunk header + samples + let riff = (4 + 8 + u64::from(fmt_len) + fact_len + 8).checked_add(padded).ok_or_else(too_long)?; + let rf64 = riff > u64::from(u32::MAX); + let mut v = Vec::with_capacity(100); + if rf64 { + const DS64: u64 = 8 + 28; + v.extend_from_slice(b"RF64"); + v.extend_from_slice(&u32::MAX.to_le_bytes()); + v.extend_from_slice(b"WAVEds64"); + v.extend_from_slice(&28u32.to_le_bytes()); + v.extend_from_slice(&riff.checked_add(DS64).ok_or_else(too_long)?.to_le_bytes()); + v.extend_from_slice(&data.to_le_bytes()); + v.extend_from_slice(&frames.to_le_bytes()); + v.extend_from_slice(&0u32.to_le_bytes()); // no table entries + } else { + v.extend_from_slice(b"RIFF"); + v.extend_from_slice(&(riff as u32).to_le_bytes()); + v.extend_from_slice(b"WAVE"); + } + let tag: u16 = if float { 3 } else { 1 }; + v.extend_from_slice(b"fmt "); + v.extend_from_slice(&fmt_len.to_le_bytes()); + v.extend_from_slice(&(if ext { 0xFFFE } else { tag }).to_le_bytes()); + v.extend_from_slice(&channels.to_le_bytes()); + v.extend_from_slice(&sample_rate.to_le_bytes()); + v.extend_from_slice(&byte_rate.to_le_bytes()); + v.extend_from_slice(&block_align.to_le_bytes()); + v.extend_from_slice(&bits.to_le_bytes()); + if ext { + // cbSize, valid bits, channel mask, KSDATAFORMAT_SUBTYPE_PCM / _IEEE_FLOAT + v.extend_from_slice(&22u16.to_le_bytes()); + v.extend_from_slice(&bits.to_le_bytes()); + let mask: u32 = if channels == 6 { 0x3F } else { 0 }; + v.extend_from_slice(&mask.to_le_bytes()); + v.extend_from_slice(&tag.to_le_bytes()); + v.extend_from_slice(&SUBTYPE_TAIL); + } else if float { + v.extend_from_slice(&0u16.to_le_bytes()); + } + if float { + v.extend_from_slice(b"fact"); + v.extend_from_slice(&4u32.to_le_bytes()); + v.extend_from_slice(&(if rf64 { u32::MAX } else { u32::try_from(frames).unwrap_or(u32::MAX) }).to_le_bytes()); + } + v.extend_from_slice(b"data"); + v.extend_from_slice(&(if rf64 { u32::MAX } else { data as u32 }).to_le_bytes()); + Ok(v) + } + PcmContainer::Aiff => { + let aiff_too_long = || "AIFF holds at most 4 GB of audio: export WAV, which has no such limit".to_string(); + let frames32 = u32::try_from(frames).map_err(|_| aiff_too_long())?; + // SSND: offset + block size + samples + let ssnd = u32::try_from(data.checked_add(8).ok_or_else(aiff_too_long)?).map_err(|_| aiff_too_long())?; + // floats need AIFF-C: a version chunk, and the compression type and name in COMM + let comm: u32 = if float { 18 + 4 + FL32_NAME.len() as u32 } else { 18 }; + let fver: u32 = if float { 8 + 4 } else { 0 }; + let form = (4u32 + fver + (8 + comm)) + .checked_add(8) + .and_then(|n| n.checked_add(ssnd)) + .and_then(|n| n.checked_add(ssnd & 1)) + .ok_or_else(aiff_too_long)?; + let mut v = Vec::with_capacity(90); + v.extend_from_slice(b"FORM"); + v.extend_from_slice(&form.to_be_bytes()); + if float { + v.extend_from_slice(b"AIFCFVER"); + v.extend_from_slice(&4u32.to_be_bytes()); + v.extend_from_slice(&AIFC_VERSION_1.to_be_bytes()); + } else { + v.extend_from_slice(b"AIFF"); + } + v.extend_from_slice(b"COMM"); + v.extend_from_slice(&comm.to_be_bytes()); + v.extend_from_slice(&channels.to_be_bytes()); + v.extend_from_slice(&frames32.to_be_bytes()); + v.extend_from_slice(&bits.to_be_bytes()); + v.extend_from_slice(&extended(sample_rate)); + if float { + v.extend_from_slice(b"fl32"); + v.extend_from_slice(FL32_NAME); + } + v.extend_from_slice(b"SSND"); + v.extend_from_slice(&ssnd.to_be_bytes()); + v.extend_from_slice(&0u32.to_be_bytes()); // offset + v.extend_from_slice(&0u32.to_be_bytes()); // block size + Ok(v) + } + } +} + +/// Append interleaved samples in `format` (little-endian for WAV, big-endian for AIFF). +pub fn encode(container: PcmContainer, interleaved: &[f32], format: SampleFormat, out: &mut Vec) { + let bps = format.bytes(); + out.reserve(interleaved.len().saturating_mul(bps)); + for &s in interleaved { + let q = if format.is_float() { SampleFormat::float(s).to_bits() as i32 } else { format.quantise(s) }; + match container { + PcmContainer::Wav => out.extend_from_slice(q.to_le_bytes().get(..bps).unwrap_or_default()), + PcmContainer::Aiff => out.extend_from_slice(q.to_be_bytes().get(4 - bps..).unwrap_or_default()), + } + } +} + +/// A whole file in memory (short audio; exports stream with [`header`] and [`encode`]). +fn write_whole(container: PcmContainer, interleaved: &[f32], channels: u16, sample_rate: u32, format: SampleFormat) -> Result, String> { + let per_frame = usize::from(channels.max(1)); + if !interleaved.len().is_multiple_of(per_frame) { + return Err("the samples do not make whole frames".into()); + } + let frames = (interleaved.len() / per_frame) as u64; + let mut v = header(container, channels, sample_rate, format, frames)?; + encode(container, interleaved, format, &mut v); + v.extend_from_slice(pad(frames, channels, format)); + Ok(v) +} + +/// A RIFF/WAVE file of interleaved samples. More than two channels are written as +/// `WAVE_FORMAT_EXTENSIBLE` with the speaker mask of the layout (6 channels: L, R, C, LFE, Ls, Rs +/// = `0x3F`). +pub fn write_wav(interleaved: &[f32], channels: u16, sample_rate: u32, format: SampleFormat) -> Result, String> { + write_whole(PcmContainer::Wav, interleaved, channels, sample_rate, format) +} + +/// IEEE 754 80-bit extended encoding of a sample rate (AIFF `COMM`). +pub fn extended(rate: u32) -> [u8; 10] { + let mut out = [0u8; 10]; + if rate == 0 { + return out; + } + let shift = rate.leading_zeros(); + let mant = (rate as u64) << (32 + shift); + let exp = 16383 + 31 - shift as u16; + out[..2].copy_from_slice(&exp.to_be_bytes()); + out[2..].copy_from_slice(&mant.to_be_bytes()); + out +} + +/// An AIFF file (`FORM`/`AIFF`, `COMM` + `SSND`; AIFF-C for floats) of interleaved samples. +pub fn write_aiff(interleaved: &[f32], channels: u16, sample_rate: u32, format: SampleFormat) -> Result, String> { + write_whole(PcmContainer::Aiff, interleaved, channels, sample_rate, format) +} + +#[cfg(test)] +mod tests { + use super::*; + #[test] + fn streaming_formats_preserve_samples_and_bounds() { + let samples = [1.0, -1.0, f32::NAN]; + let wav = write_wav(&samples, 1, 48_000, SampleFormat::Int24).unwrap(); + assert_eq!(&wav[..4], b"RIFF"); + assert_eq!(&wav[44..53], &[255, 255, 127, 1, 0, 128, 0, 0, 0]); + let h = header(PcmContainer::Wav, 6, 96_000, SampleFormat::Int24, 96_000 * 3600).unwrap(); + assert_eq!(&h[..4], b"RF64"); + assert!(header(PcmContainer::Aiff, 2, 48_000, SampleFormat::Int24, u64::MAX).is_err()); + let aiff = write_aiff(&[0.25, -3.0], 1, 44_100, SampleFormat::Float32).unwrap(); + assert_eq!(&aiff[8..12], b"AIFC"); + assert_eq!(&aiff[aiff.len() - 8..], [0.25f32.to_be_bytes(), (-3.0f32).to_be_bytes()].concat()); + } + + #[test] + fn zero_channels_rates_and_partial_frames_are_refused() { + for container in [PcmContainer::Wav, PcmContainer::Aiff] { + assert!(header(container, 0, 48_000, SampleFormat::Int16, 1).is_err()); + assert!(header(container, 1, 0, SampleFormat::Int16, 1).is_err()); + assert!(header(container, 2, 48_000, SampleFormat::Int16, 0).is_ok(), "zero frames stay valid"); + } + assert!(write_wav(&[0.5], 0, 48_000, SampleFormat::Int16).is_err()); + assert!(write_wav(&[0.1, 0.2, 0.3], 2, 48_000, SampleFormat::Int16).is_err()); + assert!(write_aiff(&[0.1, 0.2, 0.3], 2, 48_000, SampleFormat::Int16).is_err()); + assert!(write_aiff(&[0.1, 0.2], 2, 48_000, SampleFormat::Int16).is_ok()); + } +} diff --git a/crates/audio/src/pcm_interleaved.rs b/crates/audio/src/pcm_interleaved.rs new file mode 100644 index 000000000..f08c19202 --- /dev/null +++ b/crates/audio/src/pcm_interleaved.rs @@ -0,0 +1,96 @@ +//! Interleaved PCM packet decoding into caller-owned planar channels. + +/// The sample representation in an interleaved PCM packet. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct Format { + /// Bits per sample. + pub bits: u16, + /// IEEE floating-point samples. + pub float: bool, + /// Big-endian sample bytes. + pub big_endian: bool, + /// Signed 8-bit integers (wider integer PCM is signed). + pub signed: bool, +} + +/// Append complete interleaved frames to caller-owned planar channels. Empty channel slices and +/// zero-bit formats append nothing. Incomplete frames are ignored; unsupported widths produce +/// silent samples, matching the container adapters that use this API. +pub fn decode_into(data: &[u8], cfg: Format, out: &mut [Vec]) { + let ch = out.len(); + if cfg.bits == 0 || ch == 0 { + return; + } + let bps = usize::from(cfg.bits).div_ceil(8); + let Some(frame_bytes) = bps.checked_mul(ch).filter(|&n| n > 0) else { return }; + let n = data.len() / frame_bytes; + for channel in out.iter_mut() { + channel.reserve(n); + } + // `n` is whole frames within `data`, so every sample slice below is in bounds. + for i in 0..n { + for (c, o) in out.iter_mut().enumerate() { + let off = (i * ch + c) * bps; + let b = &data[off..off + bps]; + let v = match (cfg.bits, cfg.float, cfg.big_endian) { + (32, true, false) => f32::from_le_bytes([b[0], b[1], b[2], b[3]]), + (32, true, true) => f32::from_be_bytes([b[0], b[1], b[2], b[3]]), + (64, true, false) => f64::from_le_bytes(b.try_into().unwrap_or([0; 8])) as f32, + (64, true, true) => f64::from_be_bytes(b.try_into().unwrap_or([0; 8])) as f32, + (8, _, _) => { + if cfg.signed { + b[0] as i8 as f32 / 128.0 + } else { + (b[0] as f32 - 128.0) / 128.0 + } + } + (16, _, false) => i16::from_le_bytes([b[0], b[1]]) as f32 / 32768.0, + (16, _, true) => i16::from_be_bytes([b[0], b[1]]) as f32 / 32768.0, + (24, _, false) => (i32::from_le_bytes([0, b[0], b[1], b[2]]) >> 8) as f32 / 8_388_608.0, + (24, _, true) => (i32::from_be_bytes([b[0], b[1], b[2], 0]) >> 8) as f32 / 8_388_608.0, + (32, false, false) => i32::from_le_bytes([b[0], b[1], b[2], b[3]]) as f32 / 2_147_483_648.0, + (32, false, true) => i32::from_be_bytes([b[0], b[1], b[2], b[3]]) as f32 / 2_147_483_648.0, + _ => 0.0, + }; + o.push(v); + } + } +} + +#[cfg(test)] +mod tests { + use super::*; + #[test] + fn decodes_endian_and_interleaving_without_touching_partial_frames() { + let mut channels = vec![Vec::new(), Vec::new()]; + let fmt = Format { bits: 16, float: false, big_endian: true, signed: true }; + decode_into(&[0x40, 0, 0xc0, 0, 0xff], fmt, &mut channels); + assert_eq!(channels, [vec![0.5], vec![-0.5]]); + decode_into(&[0; 8], Format { bits: 0, ..fmt }, &mut channels); + assert_eq!(channels, [vec![0.5], vec![-0.5]]); + } + #[test] + fn decodes_sample_representations_and_ignores_invalid_inputs() { + let base = Format { bits: 8, float: false, big_endian: false, signed: false }; + let cases: &[(Format, &[u8], f32)] = &[ + (base, &[192], 0.5), + (Format { signed: true, ..base }, &[192], -0.5), + (Format { bits: 16, ..base }, &[0, 64], 0.5), + (Format { bits: 24, ..base }, &[0, 0, 64], 0.5), + (Format { bits: 24, big_endian: true, ..base }, &[0xc0, 0, 0], -0.5), + (Format { bits: 32, ..base }, &[0, 0, 0, 64], 0.5), + (Format { bits: 32, big_endian: true, ..base }, &[0xc0, 0, 0, 0], -0.5), + (Format { bits: 32, float: true, ..base }, &0.5_f32.to_le_bytes(), 0.5), + (Format { bits: 32, float: true, big_endian: true, ..base }, &(-0.5_f32).to_be_bytes(), -0.5), + (Format { bits: 64, float: true, ..base }, &0.5_f64.to_le_bytes(), 0.5), + (Format { bits: 64, float: true, big_endian: true, ..base }, &(-0.5_f64).to_be_bytes(), -0.5), + (Format { bits: 12, ..base }, &[0xff, 0xff], 0.0), + ]; + for &(format, bytes, expected) in cases { + let mut out = [vec![1.0]]; + decode_into(bytes, format, &mut out); + assert_eq!(out[0], [1.0, expected], "{format:?}"); + } + decode_into(&[1], base, &mut []); + } +} diff --git a/crates/audio/src/qoa.rs b/crates/audio/src/qoa.rs new file mode 100644 index 000000000..a04430137 --- /dev/null +++ b/crates/audio/src/qoa.rs @@ -0,0 +1,411 @@ +//! QOA ("Quite OK Audio"): a decoder (feature `qoa-decode`) and an encoder (feature `qoa-encode`) +//! for whole `.qoa` files, 16-bit, 1 to 8 channels. +//! +//! Written from the published QOA specification (qoaformat.org). A file is an 8-byte header (the +//! magic `qoaf` and the number of samples per channel) followed by frames of up to 5120 samples +//! per channel. Each frame starts with its channel count, sample rate, sample count and size, then +//! the state of one 4-tap sign-sign LMS predictor per channel, then 64-bit slices of 20 samples: +//! a 4-bit scale factor and twenty 3-bit quantised residuals, the channels interleaved slice by +//! slice. +//! +//! Streaming files (a sample count of 0 in the header) are not supported. +//! Samples are `i16`, one `Vec` per channel. + +#[cfg(feature = "qoa-decode")] +const MAX_FRAME_SAMPLES: usize = 5120; +const SLICE_SAMPLES: usize = 20; +const LMS_LEN: usize = 4; +const MAGIC: &[u8; 4] = b"qoaf"; +/// The most channels a file may have. +pub const MAX_CHANNELS: usize = 8; + +/// `dequant[scale factor][quantised residual]`: `round(round((sf + 1)^2.75) * [0.75, -0.75, 2.5, +/// -2.5, 4.5, -4.5, 7, -7])`. +const DEQUANT: [[i32; 8]; 16] = [ + [1, -1, 3, -3, 5, -5, 7, -7], + [5, -5, 18, -18, 32, -32, 49, -49], + [16, -16, 53, -53, 95, -95, 147, -147], + [34, -34, 113, -113, 203, -203, 315, -315], + [63, -63, 210, -210, 378, -378, 588, -588], + [104, -104, 345, -345, 621, -621, 966, -966], + [158, -158, 528, -528, 950, -950, 1477, -1477], + [228, -228, 760, -760, 1368, -1368, 2128, -2128], + [316, -316, 1053, -1053, 1895, -1895, 2947, -2947], + [422, -422, 1405, -1405, 2529, -2529, 3934, -3934], + [548, -548, 1828, -1828, 3290, -3290, 5117, -5117], + [696, -696, 2320, -2320, 4176, -4176, 6496, -6496], + [868, -868, 2893, -2893, 5207, -5207, 8099, -8099], + [1064, -1064, 3548, -3548, 6386, -6386, 9933, -9933], + [1286, -1286, 4288, -4288, 7718, -7718, 12005, -12005], + [1536, -1536, 5120, -5120, 9216, -9216, 14336, -14336], +]; + +/// What went wrong reading or writing a QOA file. +#[non_exhaustive] +#[derive(Debug, thiserror::Error, PartialEq, Eq)] +pub enum Error { + /// The input is not a QOA file, or is damaged or truncated. + #[error("malformed QOA: {0}")] + Malformed(&'static str), + /// A valid file this module does not handle (a streaming file). + #[error("unsupported QOA: {0}")] + Unsupported(&'static str), + /// The samples or settings given to the encoder cannot be written as QOA. + #[error("QOA cannot encode: {0}")] + InvalidInput(&'static str), +} + +/// A decoded file. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct Audio { + /// Sample rate in Hz (1 to 16 777 215). + pub sample_rate: u32, + /// One `Vec` of samples per channel, all the same length. + pub channels: Vec>, +} + +/// The predictor: the last four samples and four weights. +#[derive(Clone, Copy)] +struct Lms { + history: [i32; LMS_LEN], + weights: [i32; LMS_LEN], +} + +impl Lms { + fn new() -> Self { + Lms { history: [0; LMS_LEN], weights: [0, 0, -(1 << 13), 1 << 14] } + } + + fn predict(&self) -> i32 { + let mut p = 0i32; + for i in 0..LMS_LEN { + p = p.wrapping_add(self.weights[i].wrapping_mul(self.history[i])); + } + p >> 13 + } + + fn update(&mut self, sample: i32, residual: i32) { + let delta = residual >> 4; + for i in 0..LMS_LEN { + self.weights[i] += if self.history[i] < 0 { -delta } else { delta }; + } + self.history.copy_within(1.., 0); + self.history[LMS_LEN - 1] = sample; + } +} + +fn clamp16(v: i32) -> i32 { + v.clamp(-32768, 32767) +} + +#[cfg(feature = "qoa-decode")] +mod decode { + use super::{Audio, DEQUANT, Error, LMS_LEN, Lms, MAGIC, MAX_CHANNELS, MAX_FRAME_SAMPLES, SLICE_SAMPLES, clamp16}; + + fn be16(b: &[u8], i: usize) -> u16 { + u16::from_be_bytes([b[i], b[i + 1]]) + } + + /// Decode a whole `.qoa` file. + /// + /// The declared length is checked against the input before anything is allocated: a slice of + /// 8 bytes holds at most 20 samples, so a file cannot declare more samples than that. + pub fn decode(bytes: &[u8]) -> Result { + if bytes.len() < 8 || &bytes[..4] != MAGIC { + return Err(Error::Malformed("missing the qoaf magic")); + } + let total = u32::from_be_bytes([bytes[4], bytes[5], bytes[6], bytes[7]]) as usize; + if total == 0 { + return Err(Error::Unsupported("streaming files (a sample count of 0)")); + } + if bytes.len() < 16 { + return Err(Error::Malformed("truncated frame header")); + } + let channels = usize::from(bytes[8]); + let sample_rate = u32::from_be_bytes([0, bytes[9], bytes[10], bytes[11]]); + if channels == 0 || channels > MAX_CHANNELS || sample_rate == 0 { + return Err(Error::Malformed("bad channel count or sample rate")); + } + // 8 bytes (20 samples) per channel at the very least, plus the frame headers. + let room = (bytes.len() - 8) / (8 * channels) * SLICE_SAMPLES; + if total > room { + return Err(Error::Malformed("the file is too short for its declared length")); + } + let mut out: Vec> = Vec::new(); + for _ in 0..channels { + let mut v = Vec::new(); + v.try_reserve_exact(total).map_err(|_| Error::Malformed("too many samples"))?; + out.push(v); + } + let mut pos = 8; + let mut done = 0usize; + while done < total { + let head = bytes.get(pos..pos + 8).ok_or(Error::Malformed("truncated frame header"))?; + let rate = u32::from_be_bytes([0, head[1], head[2], head[3]]); + if usize::from(head[0]) != channels || rate != sample_rate { + return Err(Error::Malformed("the channel count or sample rate changes between frames")); + } + let samples = usize::from(be16(head, 4)); + let size = usize::from(be16(head, 6)); + let slices = samples.div_ceil(SLICE_SAMPLES); + if samples == 0 || samples > MAX_FRAME_SAMPLES || samples > total - done { + return Err(Error::Malformed("bad frame sample count")); + } + if size != 8 + 16 * channels + 8 * slices * channels { + return Err(Error::Malformed("frame size does not match its sample count")); + } + let frame = bytes.get(pos..pos + size).ok_or(Error::Malformed("truncated frame"))?; + let mut lms = [Lms::new(); MAX_CHANNELS]; + for (c, state) in lms.iter_mut().take(channels).enumerate() { + let at = 8 + c * 16; + for i in 0..LMS_LEN { + state.history[i] = i32::from(be16(frame, at + 2 * i) as i16); + state.weights[i] = i32::from(be16(frame, at + 8 + 2 * i) as i16); + } + } + let mut at = 8 + 16 * channels; + for s in 0..slices { + let n = (samples - s * SLICE_SAMPLES).min(SLICE_SAMPLES); + for (c, state) in lms.iter_mut().take(channels).enumerate() { + let mut slice = u64::from_be_bytes([ + frame[at], + frame[at + 1], + frame[at + 2], + frame[at + 3], + frame[at + 4], + frame[at + 5], + frame[at + 6], + frame[at + 7], + ]); + at += 8; + let table = &DEQUANT[(slice >> 60) as usize & 15]; + for _ in 0..n { + let dequantised = table[(slice >> 57) as usize & 7]; + slice <<= 3; + let sample = clamp16(state.predict() + dequantised); + out[c].push(sample as i16); + state.update(sample, dequantised); + } + } + } + pos += size; + done += samples; + } + Ok(Audio { sample_rate, channels: out }) + } +} +#[cfg(feature = "qoa-decode")] +pub use decode::decode; + +#[cfg(feature = "qoa-encode")] +mod encode { + use super::{DEQUANT, Error, Lms, MAGIC, MAX_CHANNELS, SLICE_SAMPLES, clamp16}; + + /// The quantised residual (0..8) whose dequantised value is nearest `residual`. + fn quantise(table: &[i32; 8], residual: i32) -> usize { + let mut best = 0; + let mut best_err = i64::MAX; + for (q, &d) in table.iter().enumerate() { + let e = i64::from(residual - d).pow(2); + if e < best_err { + best = q; + best_err = e; + } + } + best + } + + /// Encode one slice of one channel with every scale factor and keep the one with the least + /// squared error. Returns the 64-bit slice; `lms` is advanced to the chosen result. + fn encode_slice(lms: &mut Lms, samples: &[i16]) -> u64 { + let mut best: Option<(i64, u64, Lms)> = None; + for (sf, table) in DEQUANT.iter().enumerate() { + let mut trial = *lms; + let mut slice = sf as u64; + let mut err = 0i64; + for k in 0..SLICE_SAMPLES { + let mut q = 0; + if let Some(&s) = samples.get(k) { + let predicted = trial.predict(); + q = quantise(table, i32::from(s) - predicted); + let sample = clamp16(predicted + table[q]); + err += i64::from(i32::from(s) - sample).pow(2); + trial.update(sample, table[q]); + } + slice = slice << 3 | q as u64; + } + // 4 + 60 bits are in `slice`; the scale factor belongs on top. + let slice = slice & ((1 << 60) - 1) | (sf as u64) << 60; + if best.as_ref().is_none_or(|b| err < b.0) { + best = Some((err, slice, trial)); + } + } + match best { + Some((_, slice, state)) => { + *lms = state; + slice + } + None => 0, + } + } + + /// Encode `channels` (one `Vec` of samples per channel, all the same length) as a `.qoa` file. + pub fn encode(sample_rate: u32, channels: &[Vec]) -> Result, Error> { + let n = channels.len(); + if n == 0 || n > MAX_CHANNELS { + return Err(Error::InvalidInput("QOA stores 1 to 8 channels")); + } + if sample_rate == 0 || sample_rate > 0x00ff_ffff { + return Err(Error::InvalidInput("the sample rate must be 1 to 16777215 Hz")); + } + let total = channels[0].len(); + if channels.iter().any(|c| c.len() != total) { + return Err(Error::InvalidInput("the channels differ in length")); + } + let total32 = u32::try_from(total).map_err(|_| Error::InvalidInput("too many samples"))?; + if total == 0 { + return Err(Error::InvalidInput("QOA cannot store an empty file")); + } + let mut out = Vec::with_capacity(8 + total * n / 2); + out.extend_from_slice(MAGIC); + out.extend_from_slice(&total32.to_be_bytes()); + let mut lms = [Lms::new(); MAX_CHANNELS]; + let mut start = 0; + while start < total { + let len = (total - start).min(5120); + let slices = len.div_ceil(SLICE_SAMPLES); + let size = 8 + 16 * n + 8 * slices * n; + out.push(n as u8); + out.extend_from_slice(&sample_rate.to_be_bytes()[1..]); + out.extend_from_slice(&(len as u16).to_be_bytes()); + out.extend_from_slice(&(size as u16).to_be_bytes()); + // The state a frame starts from is written as 16-bit values, so the predictor + // continues from exactly what a decoder reads. + for state in lms.iter_mut().take(n) { + for v in state.history.iter_mut().chain(state.weights.iter_mut()) { + *v = i32::from(*v as i16); + out.extend_from_slice(&(*v as i16).to_be_bytes()); + } + } + for s in 0..slices { + let from = start + s * SLICE_SAMPLES; + let to = (from + SLICE_SAMPLES).min(start + len); + for (c, state) in lms.iter_mut().take(n).enumerate() { + out.extend_from_slice(&encode_slice(state, &channels[c][from..to]).to_be_bytes()); + } + } + start += len; + } + Ok(out) + } +} +#[cfg(feature = "qoa-encode")] +pub use encode::encode; + +#[cfg(all(test, feature = "qoa-decode", feature = "qoa-encode"))] +mod tests { + use super::*; + + fn tone(frames: usize, ch: usize) -> Vec> { + (0..ch) + .map(|c| (0..frames).map(|i| (((i as f64) * (0.03 + 0.01 * c as f64)).sin() * 12000.0 + ((i * 7919) % 200) as f64) as i16).collect()) + .collect() + } + + fn snr(a: &[i16], b: &[i16]) -> f64 { + let (mut s, mut e) = (0f64, 0f64); + for (&x, &y) in a.iter().zip(b) { + s += f64::from(x).powi(2); + e += (f64::from(x) - f64::from(y)).powi(2); + } + 10.0 * (s / e.max(1.0)).log10() + } + + #[test] + fn dequant_table_matches_its_formula() { + for (sf, row) in DEQUANT.iter().enumerate() { + let scale = ((sf as f64 + 1.0).powf(2.75)).round(); + for (q, base) in [0.75, -0.75, 2.5, -2.5, 4.5, -4.5, 7.0, -7.0].into_iter().enumerate() { + assert_eq!(f64::from(row[q]), (scale * base).round(), "sf {sf} q {q}"); + } + } + } + + #[test] + fn round_trips_with_a_good_signal_to_noise_ratio() { + for (frames, ch) in [(1, 1), (19, 2), (20, 1), (21, 2), (5120, 1), (5121, 2), (12_345, 8)] { + let src = tone(frames, ch); + let bytes = encode(44_100, &src).unwrap(); + let back = decode(&bytes).unwrap(); + assert_eq!((back.sample_rate, back.channels.len(), back.channels[0].len()), (44_100, ch, frames)); + if frames > 1000 { + for (c, (a, b)) in src.iter().zip(&back.channels).enumerate() { + assert!(snr(a, b) > 25.0, "{frames}x{ch} channel {c}"); + } + } + } + } + + #[test] + fn silence_and_full_scale_extremes_survive() { + let silent = vec![vec![0i16; 3000]]; + // No residual value is zero, so silence decodes to a tiny residue, never exactly 0. + let quiet = decode(&encode(8000, &silent).unwrap()).unwrap(); + assert!(quiet.channels[0].iter().all(|&s| s.abs() <= 2), "silence stays within 2 LSB"); + let square: Vec = (0..4000).map(|i| if (i / 50) % 2 == 0 { i16::MAX } else { i16::MIN }).collect(); + let back = decode(&encode(48_000, std::slice::from_ref(&square)).unwrap()).unwrap(); + assert_eq!(back.channels[0].len(), square.len()); + } + + #[test] + fn the_file_has_the_documented_layout() { + let bytes = encode(48_000, &tone(100, 2)).unwrap(); + assert_eq!(&bytes[..4], b"qoaf"); + assert_eq!(u32::from_be_bytes(bytes[4..8].try_into().unwrap()), 100); + assert_eq!(bytes[8], 2); + assert_eq!(&bytes[9..12], &48_000u32.to_be_bytes()[1..]); + assert_eq!(u16::from_be_bytes([bytes[12], bytes[13]]), 100); + // 8 header + 2 channels * 16 state + 5 slices * 2 channels * 8 + assert_eq!(usize::from(u16::from_be_bytes([bytes[14], bytes[15]])), 8 + 32 + 80); + assert_eq!(bytes.len(), 8 + 8 + 32 + 80); + // The first frame starts from the specified predictor state. + assert_eq!(&bytes[16 + 8..16 + 16], &[0, 0, 0, 0, 0xe0, 0, 0x40, 0]); + } + + #[test] + fn invalid_input_to_the_encoder_is_refused() { + assert!(encode(44_100, &[]).is_err()); + assert!(encode(0, &[vec![0; 10]]).is_err()); + assert!(encode(1 << 24, &[vec![0; 10]]).is_err()); + assert!(encode(44_100, &[vec![0; 10], vec![0; 9]]).is_err()); + assert!(encode(44_100, &[vec![]]).is_err()); + assert!(encode(44_100, &vec![vec![0; 10]; 9]).is_err()); + } + + #[test] + fn hostile_input_is_an_error_not_a_panic() { + let good = encode(44_100, &tone(6000, 2)).unwrap(); + for len in 0..good.len().min(600) { + let _ = decode(&good[..len]); + } + assert!(decode(&good[..good.len() - 1]).is_err()); + // A header that declares more samples than the bytes can hold. + let mut huge = good.clone(); + huge[4..8].copy_from_slice(&u32::MAX.to_be_bytes()); + assert!(decode(&huge).is_err()); + // A streaming file, and a frame whose size field lies. + let mut streaming = good.clone(); + streaming[4..8].fill(0); + assert_eq!(decode(&streaming), Err(Error::Unsupported("streaming files (a sample count of 0)"))); + let mut lying = good.clone(); + lying[14] ^= 1; + assert!(decode(&lying).is_err()); + let mut zero_rate = good.clone(); + zero_rate[9..12].fill(0); + assert!(decode(&zero_rate).is_err()); + let mut changed = good; + let second = 8 + usize::from(u16::from_be_bytes([changed[14], changed[15]])); + changed[second] = 1; + assert!(decode(&changed).is_err()); + } +} diff --git a/crates/audio/src/read/native.rs b/crates/audio/src/read/native.rs index cc0fac4c4..5502ba7e1 100644 --- a/crates/audio/src/read/native.rs +++ b/crates/audio/src/read/native.rs @@ -413,7 +413,7 @@ impl<'a> Native<'a> { let Some(next) = self.packets.next() else { // A track that declares no frames (an empty WAV or AIFF) is empty, not broken. let empty = self.frames_known && self.info.frames == 0 && self.errors == 0; - if !self.decoded && !empty && !(self.seeked && self.errors == 0) { + if !(self.decoded || empty || self.seeked && self.errors == 0) { return Err(self.no_audio("stream contains no audio")); } return Ok(None); diff --git a/crates/audio/src/read/wav.rs b/crates/audio/src/read/wav.rs index bd9c2301a..c1552d7e2 100644 --- a/crates/audio/src/read/wav.rs +++ b/crates/audio/src/read/wav.rs @@ -115,6 +115,50 @@ fn parse_bext(c: &[u8]) -> Option { }) } +/// Cap optional metadata independently of audio length, including on range-backed sources. +const IXML_MAX: usize = 1 << 20; + +/// iXML's BEXT mirror is a fallback only: a real `bext` always wins, including timestamp zero. +/// Read the two unsigned 32-bit words, not arbitrary text that merely contains their tag names. +fn ixml_time_reference(bytes: &[u8]) -> Option { + let options = craft_xml::Options { + doctype: craft_xml::Doctype::Reject, + ..craft_xml::Options::default().strict().with_limits(craft_xml::Limits { + max_input: IXML_MAX, + max_depth: 32, + max_nodes: 8192, + max_attrs: 64, + max_attrs_total: 8192, + ..craft_xml::Limits::default() + }) + }; + let root = options.parse(bytes).ok()?; + if root.name != "BWFXML" { + return None; + } + let mut bexts = root.children("BEXT"); + let bext = bexts.next()?; + if bexts.next().is_some() { + return None; + } + let word = |name: &str| -> Option> { + let mut fields = bext.children(name); + let Some(field) = fields.next() else { return Some(None) }; + if fields.next().is_some() || field.children.iter().any(|n| matches!(n, craft_xml::Node::El(_))) { + return None; + } + let text = field.text(); + let text = text.trim(); + if text.is_empty() || !text.bytes().all(|b| b.is_ascii_digit()) { + return None; + } + Some(Some(text.parse().ok()?)) + }; + let low = word("BWF_TIME_REFERENCE_LOW")??; + let high = word("BWF_TIME_REFERENCE_HIGH")?.unwrap_or(0); + Some((u64::from(high) << 32) | u64::from(low)) +} + /// Refuse the WAVE file at `at`, which symphonia reads (A-law, ADPCM, ...), if an INFO entry /// claims more bytes than the file holds: symphonia allocates each entry's length. Chunks are /// walked where symphonia reads them: after the bytes a chunk's parser used, not its length. @@ -239,7 +283,7 @@ fn pcm_layout(fmt: &Fmt) -> Result> { } /// Parse a WAV file's chunks: its format, `bext` metadata and where its samples are. Only chunk -/// headers and the small `fmt `, `bext` and `ds64` bodies are read. +/// headers, small `fmt `, `bext` and `ds64` bodies, and at most one 1 MiB iXML document are read. pub(crate) fn parse(src: &Src<'_>) -> Result { if !is_wav(&src.bytes_upto(0, 12)) { return Err(AudioError::Malformed("not a RIFF/RF64 WAVE file".into())); @@ -249,6 +293,8 @@ pub(crate) fn parse(src: &Src<'_>) -> Result { let mut ds64_data: Option = None; let mut fmt: Option = None; let mut bwf: Option = None; + let mut ixml_time = None; + let mut ixml_seen = false; let mut data: Option<(u64, u64)> = None; let mut pos: u64 = 12; @@ -288,7 +334,18 @@ pub(crate) fn parse(src: &Src<'_>) -> Result { let small = |n: u64| src.bytes(body, usize::try_from(declared.min(n)).unwrap_or(0)).unwrap_or_default(); match &id { b"fmt " => fmt = Some(parse_fmt(&small(256))?), - b"bext" => bwf = parse_bext(&small(4096)), + b"bext" => { + if let Some(info) = parse_bext(&small(4096)) { + bwf = Some(info); + } + } + b"iXML" if !ixml_seen => { + // Inspect at most one bounded document, not 100,000 individually bounded trees. + ixml_seen = true; + if declared <= IXML_MAX as u64 { + ixml_time = ixml_time_reference(&small(IXML_MAX as u64)); + } + } b"ds64" => ds64_data = le_u64(&small(64), 8), _ => {} } @@ -297,6 +354,9 @@ pub(crate) fn parse(src: &Src<'_>) -> Result { pos = next; } + if bwf.is_none() { + bwf = ixml_time.map(|time_reference| BwfInfo { time_reference, ..BwfInfo::default() }); + } let fmt = fmt.ok_or_else(|| AudioError::Malformed("WAV has no fmt chunk".into()))?; if fmt.channels == 0 { return Err(AudioError::Malformed("WAV declares zero channels".into())); diff --git a/crates/audio/src/track.rs b/crates/audio/src/track.rs index 920f571ae..e522e6f68 100644 --- a/crates/audio/src/track.rs +++ b/crates/audio/src/track.rs @@ -24,8 +24,10 @@ pub const LOOKAHEAD_PACKETS: usize = 8; enum Inner { #[cfg(feature = "aac-decode")] Aac { + /// The decoder, with SBR on for HE-AAC. dec: Box, - up: Option, + /// HE-AAC: 2048 samples per access unit, at twice the core rate. + sbr: bool, order: Option<&'static [usize]>, /// The format ADTS packets must keep, when their headers gave the configuration. adts_format: Option, @@ -74,7 +76,7 @@ pub(crate) type PacketFrames = fn(&[u8], u64) -> Option; /// /// It takes the track's [`CodecParams`] (from `isobmff::Track::codec_params`, /// `matroska::Track::codec_params`, `ogg::Stream::codec_params`, or built by hand for another -/// container) and decodes AAC (LC; HE-AAC as its upsampled LC core), ALAC, Opus, AC-3/E-AC-3, +/// container) and decodes AAC (LC, and HE-AAC with its SBR high band), ALAC, Opus, AC-3/E-AC-3, /// MP1/MP2, MP3 and PCM. Container timing (AAC priming, Opus pre-skip, end trimming) is the caller's: /// every decoded sample is returned. /// @@ -168,17 +170,18 @@ impl TrackDecoder { params.extradata.clone() }; let asc = asc.as_slice(); - let dec = aac::Decoder::new(asc)?; + let mut dec = aac::Decoder::new(asc)?; let out = aac::output_rate(asc, payloads.iter().copied()).unwrap_or(dec.sample_rate()); - let up = (dec.sample_rate().checked_mul(2) == Some(out)).then(|| aac::Upsample2x::new(dec.channels())); - let codec = if up.is_some() { CodecId::HeAac } else { CodecId::Aac }; - // Only twice the core rate is upsampled to; any other signalled rate plays the core. - let out = if up.is_some() { out } else { dec.sample_rate() }; + let sbr = dec.sample_rate().checked_mul(2) == Some(out); + let codec = if sbr { CodecId::HeAac } else { CodecId::Aac }; + // Only SBR at twice the core rate is decoded; any other signalled rate plays the core. + dec.set_sbr(sbr); + let out = dec.sample_rate(); // a program config element that spells out a configuration has its speakers let cfg = aac::speaker_config(dec.config()); let layout = aac_layout(cfg, dec.channels()); let order = layout.is_ordered().then(|| aac::wav_order(cfg)).flatten(); - (Inner::Aac { dec: Box::new(dec), up, order, adts_format }, codec, out, layout, SampleFormat::Compressed) + (Inner::Aac { dec: Box::new(dec), sbr, order, adts_format }, codec, out, layout, SampleFormat::Compressed) } #[cfg(feature = "alac-decode")] CodecId::Alac => { @@ -320,7 +323,7 @@ impl TrackDecoder { let mut partial = false; let channels = match &mut self.inner { #[cfg(feature = "aac-decode")] - Inner::Aac { dec, up, order, adts_format } => { + Inner::Aac { dec, order, adts_format, .. } => { let mut out: Vec> = Vec::new(); let units = match adts_frames(packet) { Some(frames) => adts_units(frames, *adts_format)?, @@ -328,17 +331,11 @@ impl TrackDecoder { }; // A failed frame leaves the decoder as it was, so the frames before it in the // packet must too: their audio is replaced by the packet's concealment. - let before = (units.len() > 1).then(|| (dec.clone(), up.clone())); - let decoded = units.into_iter().try_for_each(|unit| { - let mut block = dec.decode(unit)?; - if let Some(up) = up.as_mut() { - block = up.process(block); - } - append(&mut out, block) - }); + let before = (units.len() > 1).then(|| dec.clone()); + let decoded = units.into_iter().try_for_each(|unit| append(&mut out, dec.decode(unit)?)); if let Err(e) = decoded { if let Some(before) = before { - (*dec, *up) = before; + *dec = before; } return Err(e); } @@ -440,8 +437,8 @@ impl TrackDecoder { None } #[cfg(feature = "aac-decode")] - Inner::Aac { up, .. } => { - let per_unit = aac::FRAME_LEN * if up.is_some() { 2 } else { 1 }; + Inner::Aac { sbr, .. } => { + let per_unit = aac::FRAME_LEN * if *sbr { 2 } else { 1 }; match adts_frames(data) { Some(f) => Some(f.iter().map(|(h, _)| usize::from(h.raw_data_blocks)).sum::().min(MAX_ADTS_FRAMES_PER_PACKET) * per_unit), // A packet lost whole: as long as the last one, in a stream of multi-frame @@ -488,7 +485,7 @@ impl TrackDecoder { pub fn frames_per_packet(&self) -> Option { match &self.inner { #[cfg(feature = "aac-decode")] - Inner::Aac { up, .. } => Some(aac::FRAME_LEN * if up.is_some() { 2 } else { 1 }), + Inner::Aac { sbr, .. } => Some(aac::FRAME_LEN * if *sbr { 2 } else { 1 }), #[cfg(feature = "alac-decode")] Inner::Alac { .. } => None, #[cfg(feature = "ac3-decode")] @@ -544,7 +541,7 @@ impl TrackDecoder { pub fn preroll(&self) -> usize { match &self.inner { #[cfg(feature = "aac-decode")] - Inner::Aac { up, .. } => 2 * aac::FRAME_LEN * if up.is_some() { 2 } else { 1 }, + Inner::Aac { sbr, .. } => 2 * aac::FRAME_LEN * if *sbr { 2 } else { 1 }, #[cfg(feature = "alac-decode")] Inner::Alac { .. } => 0, #[cfg(feature = "ac3-decode")] @@ -570,12 +567,7 @@ impl TrackDecoder { self.last_frames = 0; match &mut self.inner { #[cfg(feature = "aac-decode")] - Inner::Aac { dec, up, .. } => { - dec.reset(); - if let Some(up) = up { - up.reset(); - } - } + Inner::Aac { dec, .. } => dec.reset(), #[cfg(feature = "alac-decode")] Inner::Alac { .. } => {} #[cfg(feature = "ac3-decode")] diff --git a/crates/audio/src/vorbis/encoder/book.rs b/crates/audio/src/vorbis/encoder/book.rs new file mode 100644 index 000000000..b392cf27c --- /dev/null +++ b/crates/audio/src/vorbis/encoder/book.rs @@ -0,0 +1,287 @@ +//! Bit packing (Vorbis I §2) and codebooks (§3): Huffman lengths, the codeword assignment the +//! decoder derives from them, VQ lattices (lookup type 1) and the header encoding. + +use super::{Error, Result}; + +/// LSB-first bit packer (Vorbis I §2.1.4): bits fill each byte from its least significant end. +#[derive(Clone, Debug, Default)] +pub struct BitWriter { + buf: Vec, + bits: u64, +} + +impl BitWriter { + pub fn new() -> BitWriter { + BitWriter::default() + } + + /// Writes the low `n` bits of `v` (`n` ≤ 32). + pub fn write(&mut self, v: u32, n: u32) { + let n = n.min(32); + for i in 0..n { + let bit = (v >> i) & 1; + let at = (self.bits / 8) as usize; + if at == self.buf.len() { + self.buf.push(0); + } + if bit != 0 + && let Some(b) = self.buf.get_mut(at) + { + *b |= 1 << (self.bits % 8); + } + self.bits += 1; + } + } + + pub fn flag(&mut self, b: bool) { + self.write(b as u32, 1); + } + + /// Writes a codeword MSB first: the decoder walks the tree from its first bit. + fn codeword(&mut self, word: u32, len: u32) { + for i in (0..len).rev() { + self.write((word >> i) & 1, 1); + } + } + + pub fn bytes(&mut self, data: &[u8]) { + for &b in data { + self.write(u32::from(b), 8); + } + } + + pub fn finish(self) -> Vec { + self.buf + } +} + +/// `ilog` of the specification (§9.2.1): the position of the highest set bit, 0 for 0. +pub fn ilog(v: u32) -> u32 { + 32 - v.leading_zeros() +} + +/// The 32-bit float format of codebook headers (§9.2.2), for values it represents exactly. +pub fn float32_pack(v: f64) -> u32 { + if v == 0.0 || !v.is_finite() { + return 0; + } + let sign = if v < 0.0 { 0x8000_0000u32 } else { 0 }; + let mut m = v.abs(); + let mut exp: i32 = 0; + // a 21-bit mantissa: scale into [2^20, 2^21) + while m >= (1 << 21) as f64 { + m /= 2.0; + exp += 1; + } + while m < (1 << 20) as f64 { + m *= 2.0; + exp -= 1; + } + let mant = (m.round() as u32).min((1 << 21) - 1); + let biased = (exp + 788).clamp(0, 1023) as u32; + sign | (biased << 21) | mant +} + +/// Huffman code lengths (1..=`max_len`) for `weights`, every entry coded. +pub fn huffman_lengths(weights: &[f64], max_len: u32) -> Vec { + let n = weights.len(); + if n <= 1 { + return vec![1; n]; + } + let total: f64 = weights.iter().map(|w| if w.is_finite() { w.max(0.0) } else { 0.0 }).sum::().max(1e-300); + let mut floor = 1e-9; + loop { + let w: Vec = weights.iter().map(|&x| (if x.is_finite() { x.max(0.0) } else { 0.0 }) / total + floor).collect(); + let lens = huffman(&w); + if lens.iter().all(|&l| l <= max_len) || floor > 1.0 { + return lens.into_iter().map(|l| l.clamp(1, max_len)).collect(); + } + floor *= 4.0; + } +} + +fn huffman(w: &[f64]) -> Vec { + // nodes: (weight, children) with leaves first + let n = w.len(); + let mut parent = vec![usize::MAX; 2 * n]; + let mut weight: Vec = w.to_vec(); + let mut alive: Vec = (0..n).collect(); + while alive.len() > 1 { + alive.sort_by(|&a, &b| weight[b].partial_cmp(&weight[a]).unwrap_or(std::cmp::Ordering::Equal).then(b.cmp(&a))); + let (Some(a), Some(b)) = (alive.pop(), alive.pop()) else { break }; + let id = weight.len(); + weight.push(weight[a] + weight[b]); + if parent.len() <= id { + parent.resize(id + 1, usize::MAX); + } + parent[a] = id; + parent[b] = id; + alive.push(id); + } + (0..n) + .map(|i| { + let mut d = 0; + let mut x = i; + while let Some(&p) = parent.get(x) { + if p == usize::MAX { + break; + } + d += 1; + x = p; + } + d + }) + .collect() +} + +/// The codewords the decoder assigns to `lengths` (§3.2.1): in entry order, each entry gets the +/// lowest available codeword of its length (no prefix of an assigned codeword, and no assigned +/// codeword a prefix of it). An error when the lengths do not fit a prefix code. +pub fn assign_codewords(lengths: &[u32]) -> Result> { + // A trie of the assigned codewords: `children[node]` (0 = none), `leaf[node]`. + let mut children: Vec<[usize; 2]> = vec![[0, 0]]; + let mut leaf = vec![false]; + let mut words = Vec::with_capacity(lengths.len()); + for &len in lengths { + if len == 0 || len > 32 { + return Err(Error::Internal("codeword length")); + } + let bits = lowest_free(&children, &leaf, 0, len).ok_or(Error::Internal("over-subscribed codebook"))?; + let mut node = 0usize; + let mut word = 0u32; + for &bit in &bits { + word = (word << 1) | bit as u32; + node = match children[node][bit] { + 0 => { + children.push([0, 0]); + leaf.push(false); + let id = children.len() - 1; + children[node][bit] = id; + id + } + c => c, + }; + } + leaf[node] = true; + words.push(word); + } + Ok(words) +} + +/// The leftmost free path of `depth` bits below `node` (depth-first, 0 before 1). +fn lowest_free(children: &[[usize; 2]], leaf: &[bool], node: usize, depth: u32) -> Option> { + if leaf[node] || depth == 0 { + return None; + } + for bit in 0..2 { + let c = children[node][bit]; + if c == 0 { + let mut p = vec![bit]; + p.resize(depth as usize, 0); + return Some(p); + } + if let Some(mut rest) = lowest_free(children, leaf, c, depth - 1) { + rest.insert(0, bit); + return Some(rest); + } + } + None +} + +/// A codebook: entry lengths and codewords, and for VQ books a lattice (lookup type 1). +#[derive(Clone, Debug)] +pub struct Codebook { + pub dimensions: u32, + pub lengths: Vec, + pub words: Vec, + /// Lattice values per dimension (lookup type 1: `values^dimensions == entries`). + pub lattice: Option, +} + +#[derive(Clone, Copy, Debug)] +pub struct Lattice { + pub minimum: f64, + pub delta: f64, + /// Values per dimension. + pub values: u32, +} + +impl Codebook { + /// A scalar book (no lookup) of `weights.len()` entries. + pub fn scalar(dimensions: u32, weights: &[f64]) -> Result { + let lengths = huffman_lengths(weights, 20); + let words = assign_codewords(&lengths)?; + Ok(Codebook { dimensions, lengths, words, lattice: None }) + } + + /// A lattice VQ book: `values` integers per dimension from `minimum` in steps of `delta`; + /// `weight(v)` is the relative probability of one component `v`. + pub fn lattice(dimensions: u32, values: u32, minimum: i32, delta: i32, weight: impl Fn(i32) -> f64) -> Result { + let entries = values.checked_pow(dimensions).ok_or(Error::Internal("codebook size"))?; + let w: Vec = (0..entries) + .map(|e| { + let mut p = 1.0; + let mut rest = e; + for _ in 0..dimensions { + let t = (rest % values) as i32; + rest /= values; + p *= weight(minimum + t * delta); + } + p + }) + .collect(); + let lengths = huffman_lengths(&w, 20); + let words = assign_codewords(&lengths)?; + Ok(Codebook { dimensions, lengths, words, lattice: Some(Lattice { minimum: f64::from(minimum), delta: f64::from(delta), values }) }) + } + + pub fn entries(&self) -> usize { + self.lengths.len() + } + + /// Writes entry `e`. + pub fn put(&self, w: &mut BitWriter, e: usize) { + if let (Some(&word), Some(&len)) = (self.words.get(e), self.lengths.get(e)) { + w.codeword(word, len); + } + } + + /// The entry of the lattice vector `v` (component 0 is the least significant digit, §3.2.1). + pub fn entry_of(&self, v: &[i32]) -> usize { + let Some(l) = self.lattice else { return 0 }; + let mut e = 0usize; + let mut scale = 1usize; + for &x in v.iter().take(self.dimensions as usize) { + let t = ((f64::from(x) - l.minimum) / l.delta).round().clamp(0.0, f64::from(l.values - 1)) as usize; + e += t * scale; + scale *= l.values as usize; + } + e + } + + /// The header encoding (§3.2.1). + pub fn write_header(&self, w: &mut BitWriter) { + w.write(0x56_4342, 24); + w.write(self.dimensions, 16); + w.write(self.entries() as u32, 24); + w.flag(false); // not ordered + w.flag(false); // not sparse + for &l in &self.lengths { + w.write(l - 1, 5); + } + match self.lattice { + None => w.write(0, 4), + Some(l) => { + w.write(1, 4); + w.write(float32_pack(l.minimum), 32); + w.write(float32_pack(l.delta), 32); + let bits = ilog(l.values - 1).max(1); + w.write(bits - 1, 4); + w.flag(false); // sequence_p + for m in 0..l.values { + w.write(m, bits); + } + } + } + } +} diff --git a/crates/audio/src/vorbis/encoder/floor.rs b/crates/audio/src/vorbis/encoder/floor.rs new file mode 100644 index 000000000..503385bd3 --- /dev/null +++ b/crates/audio/src/vorbis/encoder/floor.rs @@ -0,0 +1,207 @@ +//! Floor type 1 (Vorbis I §7): a piecewise-linear spectral envelope in the decoder's dB scale. +//! +//! The encoder picks a target value at each of the floor's X positions, codes it against the +//! decoder's prediction from its neighbours (step 2 of §7.2.4, inverted), and renders the curve +//! exactly as the decoder will (`render_line`, integer arithmetic), so the residue is computed +//! against the very floor the decoder multiplies it by. + +use super::book::{BitWriter, Codebook, ilog}; + +/// `floor1_multiplier`: 2, so values run 0..128 and step two entries of the dB table. +pub const MULTIPLIER: u32 = 2; +/// The value range for [`MULTIPLIER`] 2. +pub const RANGE: i32 = 128; +const RANGE_BITS: u32 = 10; +/// Points per partition, and partitions. +pub const CLASS_DIM: usize = 4; +pub const PARTITIONS: usize = 8; + +/// Step (in dB) between entries of `floor1_inverse_dB_table`: it runs from 1.0649863e-07 to 1 +/// over 256 entries. +fn db_step() -> f64 { + (1.0f64 / 1.064_986_3e-7).ln() / 255.0 +} + +/// `floor1_inverse_dB_table[i]`. +pub fn inverse_db(i: i32) -> f32 { + (f64::from(i.clamp(0, 255) - 255) * db_step()).exp() as f32 +} + +/// The table index whose amplitude is nearest (in dB) to `a`. +pub fn db_index(a: f32) -> i32 { + if !a.is_finite() || a <= 0.0 { + return 0; + } + ((f64::from(a)).ln() / db_step() + 255.0).round().clamp(0.0, 255.0) as i32 +} + +/// A floor 1 configuration for blocks of `n/2 = 2^RANGE_BITS` bins. +#[derive(Clone, Debug)] +pub struct Floor1 { + /// X positions: 0, 1024, then the partitions' points. + pub xs: Vec, + /// Index of the book coding the point values. + pub book: usize, + /// For each point (in coding order) its low and high neighbours. + neighbours: Vec<(usize, usize)>, + /// Point indices sorted by X. + sorted: Vec, +} + +impl Floor1 { + /// Points spaced about evenly in log frequency over 1..1024. + pub fn new(book: usize) -> Floor1 { + let count = CLASS_DIM * PARTITIONS; + let mut xs = vec![0, 1 << RANGE_BITS]; + let mut last = 0; + for i in 1..=count { + // geometric from 2 to ~1000 bins, at least one bin apart + let x = (2.0f64 * (500.0f64).powf((i - 1) as f64 / (count - 1) as f64)).round() as i32; + let x = x.max(last + 1).min((1 << RANGE_BITS) - 1); + xs.push(x); + last = x; + } + let neighbours = (0..xs.len()) + .map(|i| { + if i < 2 { + return (0, 0); + } + let (mut lo, mut hi) = (0usize, 1usize); + for j in 0..i { + if xs[j] < xs[i] && xs[j] > xs[lo] { + lo = j; + } + if xs[j] > xs[i] && xs[j] < xs[hi] { + hi = j; + } + } + (lo, hi) + }) + .collect(); + let mut sorted: Vec = (0..xs.len()).collect(); + sorted.sort_by_key(|&i| xs[i]); + Floor1 { xs, book, neighbours, sorted } + } + + /// The setup-header encoding (§7.2.2): one class of `CLASS_DIM` points, no subclasses. + pub fn write_header(&self, w: &mut BitWriter) { + w.write(PARTITIONS as u32, 5); + for _ in 0..PARTITIONS { + w.write(0, 4); + } + w.write(CLASS_DIM as u32 - 1, 3); + w.write(0, 2); // no subclasses + w.write(self.book as u32 + 1, 8); + w.write(MULTIPLIER - 1, 2); + w.write(RANGE_BITS, 4); + for &x in &self.xs[2..] { + w.write(x as u32, RANGE_BITS); + } + } + + /// Codes the target values `want` (one per point, 0..RANGE) into `w`; returns the floor curve + /// the decoder renders (`n2` amplitudes). + pub fn encode(&self, w: &mut BitWriter, book: &Codebook, want: &[i32], n2: usize) -> Vec { + let n = self.xs.len(); + let want = |i: usize| want.get(i).copied().unwrap_or(0).clamp(0, RANGE - 1); + let mut fin = vec![0i32; n]; + let mut used = vec![false; n]; + let mut coded = vec![0u32; n]; + fin[0] = want(0); + fin[1] = want(1); + used[0] = true; + used[1] = true; + for i in 2..n { + let (lo, hi) = self.neighbours[i]; + let pred = render_point(self.xs[lo], fin[lo], self.xs[hi], fin[hi], self.xs[i]); + let f = want(i); + let (highroom, lowroom) = (RANGE - pred, pred); + let room = 2 * highroom.min(lowroom); + let val = if f == pred { + 0 + } else { + let d = f - pred; + let inside = if d > 0 { 2 * d } else { -2 * d - 1 }; + if inside < room { + inside + } else if highroom > lowroom { + f - pred + lowroom + } else { + pred - f + highroom - 1 + } + }; + coded[i] = val.clamp(0, RANGE - 1) as u32; + if val != 0 { + used[lo] = true; + used[hi] = true; + used[i] = true; + fin[i] = f; + } else { + fin[i] = pred; + } + } + let bits = ilog(RANGE as u32 - 1); + w.write(fin[0] as u32, bits); + w.write(fin[1] as u32, bits); + for &c in &coded[2..] { + book.put(w, c as usize); + } + self.render(&fin, &used, n2) + } + + /// The decoder's curve (§7.2.4 step 2, `render_line`) as amplitudes. + fn render(&self, fin: &[i32], used: &[bool], n2: usize) -> Vec { + let mut v = vec![0i32; n2]; + let m = MULTIPLIER as i32; + let (mut lx, mut ly) = (0i32, fin[self.sorted[0]] * m); + let (mut hx, mut hy) = (0i32, ly); + for &i in &self.sorted[1..] { + if used[i] { + hy = fin[i] * m; + hx = self.xs[i]; + render_line(lx, ly, hx, hy, &mut v); + lx = hx; + ly = hy; + } + } + if (hx as usize) < n2 { + render_line(hx, hy, n2 as i32, hy, &mut v); + } + v.iter().map(|&i| inverse_db(i)).collect() + } +} + +fn render_point(x0: i32, y0: i32, x1: i32, y1: i32, x: i32) -> i32 { + let dy = y1 - y0; + let adx = (x1 - x0).max(1); + let err = dy.abs() * (x - x0); + let off = err / adx; + if dy < 0 { y0 - off } else { y0 + off } +} + +fn render_line(x0: i32, y0: i32, x1: i32, y1: i32, v: &mut [i32]) { + let dy = y1 - y0; + let adx = x1 - x0; + if adx <= 0 { + return; + } + let base = dy / adx; + let sy = if dy < 0 { base - 1 } else { base + 1 }; + let ady = dy.abs() - base.abs() * adx; + let (mut y, mut err) = (y0, 0); + if let Some(s) = v.get_mut(x0 as usize) { + *s = y; + } + for x in x0 + 1..x1 { + err += ady; + if err >= adx { + err -= adx; + y += sy; + } else { + y += base; + } + if let Some(s) = v.get_mut(x as usize) { + *s = y; + } + } +} diff --git a/crates/audio/src/vorbis/encoder/mdct.rs b/crates/audio/src/vorbis/encoder/mdct.rs new file mode 100644 index 000000000..3802a24d6 --- /dev/null +++ b/crates/audio/src/vorbis/encoder/mdct.rs @@ -0,0 +1,134 @@ +//! Forward MDCT of a windowed block of `n` samples to `n/2` coefficients: +//! `X[k] = Σ x[j] cos(2Ï€/n · (j + 1/2 + n/4)(k + 1/2))`, the transform the Vorbis decoder inverts +//! (Vorbis I §1.3.2, §4.3.9). Computed as a DCT-IV of the folded block through an `n/8`-point +//! complex FFT. + +use std::f64::consts::PI; + +#[derive(Clone, Copy, Default, Debug)] +struct Cpx { + re: f64, + im: f64, +} + +impl Cpx { + fn mul(self, o: Cpx) -> Cpx { + Cpx { re: self.re * o.re - self.im * o.im, im: self.re * o.im + self.im * o.re } + } +} + +/// A forward MDCT for one block size (a power of two, at least 16). +#[derive(Clone, Debug)] +pub struct Mdct { + n: usize, + twiddle: Vec, + post: Vec, + fft_tw: Vec, + rev: Vec, +} + +impl Mdct { + pub fn new(n: usize) -> Mdct { + let m = n / 2; // DCT-IV length + let h = m / 2; // FFT length + let twiddle = (0..h).map(|j| cis(-PI * (4.0 * j as f64 + 1.0) / (4.0 * m as f64))).collect(); + let post = (0..h).map(|k| cis(-PI * k as f64 / m as f64)).collect(); + let fft_tw = (0..h / 2).map(|j| cis(-2.0 * PI * j as f64 / h as f64)).collect(); + let bits = h.trailing_zeros(); + let rev = (0..h).map(|i| if bits == 0 { 0 } else { i.reverse_bits() >> (usize::BITS - bits) }).collect(); + Mdct { n, twiddle, post, fft_tw, rev } + } + + /// `x`: `n` windowed samples; `out`: `n/2` coefficients (scaled by `scale`). + pub fn forward(&self, x: &[f32], out: &mut [f32], scale: f64) { + let n = self.n; + let m = n / 2; + let q = n / 4; + let at = |i: usize| f64::from(x.get(i).copied().unwrap_or(0.0)); + // fold (a, b, c, d) to the DCT-IV input (-c_r - d, a - b_r) + let mut v = vec![0f64; m]; + for (i, s) in v.iter_mut().enumerate() { + *s = if i < q { -at(3 * q - 1 - i) - at(3 * q + i) } else { at(i - q) - at(3 * q - 1 - i) }; + } + // DCT-IV through a half-length complex FFT + let h = m / 2; + let mut z: Vec = (0..h).map(|j| Cpx { re: v[2 * j], im: v[m - 1 - 2 * j] }.mul(self.twiddle[j])).collect(); + self.fft(&mut z); + for (k, c) in z.iter().enumerate() { + let d = c.mul(self.post[k]); + if let Some(o) = out.get_mut(2 * k) { + *o = (d.re * scale) as f32; + } + if let Some(o) = out.get_mut(m - 1 - 2 * k) { + *o = (-d.im * scale) as f32; + } + } + } + + /// In-place radix-2 FFT (`e^{-2Ï€i jk/h}`). + fn fft(&self, a: &mut [Cpx]) { + let h = a.len(); + for i in 0..h { + let j = self.rev[i]; + if i < j { + a.swap(i, j); + } + } + let mut len = 2; + while len <= h { + let step = h / len; + for start in (0..h).step_by(len) { + for k in 0..len / 2 { + let w = self.fft_tw[k * step]; + let u = a[start + k]; + let t = a[start + k + len / 2].mul(w); + a[start + k] = Cpx { re: u.re + t.re, im: u.im + t.im }; + a[start + k + len / 2] = Cpx { re: u.re - t.re, im: u.im - t.im }; + } + } + len *= 2; + } + } +} + +fn cis(t: f64) -> Cpx { + Cpx { re: t.cos(), im: t.sin() } +} + +/// The Vorbis window (§4.3.1) for a block of `n` between blocks of the same size. +pub fn window(n: usize) -> Vec { + (0..n) + .map(|i| { + let s = (PI * (i as f64 + 0.5) / n as f64).sin(); + (PI / 2.0 * s * s).sin() as f32 + }) + .collect() +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn matches_the_direct_formula() { + for n in [16usize, 64, 256, 2048] { + let x: Vec = (0..n).map(|i| ((i * 7919 % 211) as f32 - 105.0) / 50.0).collect(); + let mut fast = vec![0f32; n / 2]; + Mdct::new(n).forward(&x, &mut fast, 1.0); + for k in [0usize, 1, 2, n / 4, n / 2 - 1] { + let direct: f64 = + (0..n).map(|j| f64::from(x[j]) * (2.0 * PI / n as f64 * (j as f64 + 0.5 + n as f64 / 4.0) * (k as f64 + 0.5)).cos()).sum(); + assert!((direct - f64::from(fast[k])).abs() < 1e-3 * (1.0 + direct.abs()), "n={n} k={k}: {direct} vs {}", fast[k]); + } + } + } + + #[test] + fn the_window_is_power_complementary() { + let w = window(2048); + for i in 0..1024 { + let s = w[i] * w[i] + w[i + 1024] * w[i + 1024]; + assert!((s - 1.0).abs() < 1e-5, "{i}: {s}"); + } + } +} diff --git a/crates/audio/src/vorbis/encoder/mod.rs b/crates/audio/src/vorbis/encoder/mod.rs new file mode 100644 index 000000000..95be88244 --- /dev/null +++ b/crates/audio/src/vorbis/encoder/mod.rs @@ -0,0 +1,494 @@ +//! Pure-Rust Vorbis I encoder, written from the Vorbis I specification (Xiph.Org, 2020-07-04). +//! +//! No `unsafe` and no C code; builds for `wasm32-unknown-unknown`. ffmpeg is used only as an +//! external test oracle. +//! +//! The stream: one mode of 2048-sample blocks; per channel a floor 1 envelope of 34 points; one +//! residue type 2 over the interleaved channels, partitions of 32 values (a multiple of the +//! channel count for 3, 5, 6 and 7 channels) in seven classes (silent, +//! three single-pass lattice books, and two- and three-pass cascades for large values); no channel +//! coupling. Quality (-1 to 10) sets how far the floor sits from the spectral envelope (below it +//! for tonal regions, above it for noise-like ones, from the spectral flatness), and with it the +//! residue's precision: variable bit rate, like other Vorbis encoders. +//! +//! From FilmCraft's `filmcraft-vorbis` by @JordyUSA (storytold/filmcraft#4481). +//! +//! The packets go into Ogg with [`craft_media::ogg::PageWriter`] and a comment header from +//! [`craft_media::ogg::vorbis_comment`]; [`crate::encode_ogg_vorbis`] does all of it for a +//! whole buffer. +//! +//! ``` +//! use craft_audio::vorbis::encoder::{Encoder, EncoderConfig}; +//! +//! let mut enc = Encoder::new(EncoderConfig::new(48_000, 2, 5.0))?; +//! let tone: Vec = (0..4_800).map(|i| (i as f32 * 0.05).sin() * 0.5).collect(); +//! let mut packets = enc.encode(&[&tone, &tone])?; +//! packets.extend(enc.finish()?); +//! // the last granule position is the input's length, so a decoder trims the padding +//! assert_eq!(packets.last().map(|p| p.1), Some(4_800)); +//! assert_eq!(&enc.identification()[1..7], b"vorbis"); +//! # Ok::<(), craft_audio::vorbis::encoder::Error>(()) +//! ``` + +// Index loops follow the specification, as in FilmCraft (which allows this lint workspace-wide). +#![allow(clippy::needless_range_loop)] + +mod book; +mod floor; +mod mdct; + +use book::{BitWriter, Codebook}; +use floor::Floor1; +use mdct::Mdct; + +/// Why an encoder could not be created or input could not be encoded. +#[non_exhaustive] +#[derive(Clone, Debug, PartialEq, Eq, thiserror::Error)] +pub enum Error { + /// A setting or the input is outside what the encoder takes. + #[error("invalid argument: {0}")] + InvalidArgument(&'static str), + /// An encoder invariant failed (a bug, reported instead of panicking). + #[error("internal error: {0}")] + Internal(&'static str), +} + +/// The encoder's result type. +pub type Result = std::result::Result; + +/// Block size (samples); the short block size of the header is never used. +pub const BLOCK: usize = 2048; +const HOP: usize = BLOCK / 2; +/// Residue values per partition for `channels` channels: 32, or for 3, 5, 6 and 7 channels a +/// multiple of the channel count (48, 40, 48, 56). libvorbis decodes each partition of an +/// interleaved (type 2) residue as if it started on channel 0, so a partition must hold whole +/// sample frames; the size also divides `channels Ă— 1024` and is a multiple of every book's +/// dimension (4). +fn partition_size(channels: usize) -> usize { + match channels { + 3 | 6 => 48, + 5 => 40, + 7 => 56, + _ => 32, + } +} +const CLASSES: usize = 8; +/// Largest residue magnitude the cascade codes: 7·225 + 7·15 + 7. +const MAX_RESIDUE: i32 = 1687; + +/// The quality a NaN or infinite setting reads as. +pub const DEFAULT_QUALITY: f32 = 6.0; + +/// Encoder settings. +#[derive(Clone, Copy, Debug, PartialEq)] +pub struct EncoderConfig { + /// Samples per second, 1 to 768 000. + pub sample_rate: u32, + /// 1 to 8, in Vorbis channel order (§4.3.9). + pub channels: usize, + /// -1 (smallest) to 10 (best); clamped. + pub quality: f32, +} + +impl EncoderConfig { + /// Settings for `channels` channels (in Vorbis order) at `sample_rate` and `quality`. + pub fn new(sample_rate: u32, channels: usize, quality: f32) -> EncoderConfig { + EncoderConfig { sample_rate, channels, quality } + } +} + +/// Book indices in the setup header. +const BOOK_CLASS: usize = 0; +const BOOK_FLOOR: usize = 1; +const BOOK_A1_SPARSE: usize = 2; +const BOOK_A1: usize = 3; +const BOOK_A2: usize = 4; +const BOOK_A4: usize = 5; +const BOOK_F: usize = 6; +const BOOK_M: usize = 7; +const BOOK_C: usize = 8; + +/// Books per class and pass (`None`: nothing coded in that pass). +const CASCADE: [[Option; 3]; CLASSES] = [ + [None, None, None], + [Some(BOOK_A1_SPARSE), None, None], + [Some(BOOK_A1), None, None], + [Some(BOOK_A2), None, None], + [Some(BOOK_A4), None, None], + [Some(BOOK_F), None, None], + [Some(BOOK_M), Some(BOOK_F), None], + [Some(BOOK_C), Some(BOOK_M), Some(BOOK_F)], +]; + +/// A Vorbis encoder for one stream: planar `f32` in, Vorbis packets with their granule +/// positions out. +pub struct Encoder { + cfg: EncoderConfig, + books: Vec, + floor: Floor1, + mdct: Mdct, + window: Vec, + /// Input not yet past a block, per channel; the next block starts at `start`. + pending: Vec>, + /// Read cursor into `pending`: consumed hops are dropped once per call, not per block. + start: usize, + samples_in: u64, + blocks: u64, + finished: bool, + /// Floor offset below the envelope (dB) for tonal and for noise-like regions. + tonal_db: f32, + noise_db: f32, + /// Residue magnitudes below this quantise to 0. + dead_zone: f32, +} + +impl Encoder { + /// An encoder for `cfg`: 1 to 8 channels, a sample rate of 1 to 768 000; the quality is + /// clamped to -1..=10, and NaN or infinite reads as [`DEFAULT_QUALITY`]. + pub fn new(cfg: EncoderConfig) -> Result { + if !(1..=8).contains(&cfg.channels) { + return Err(Error::InvalidArgument("Vorbis encoder: 1 to 8 channels")); + } + if !(1..=768_000).contains(&cfg.sample_rate) { + return Err(Error::InvalidArgument("Vorbis encoder: sample rate")); + } + let quality = if cfg.quality.is_finite() { cfg.quality.clamp(-1.0, 10.0) } else { DEFAULT_QUALITY }; + let cfg = EncoderConfig { quality, ..cfg }; + let books = vec![ + Codebook::scalar(2, &(0..CLASSES * CLASSES).map(|e| class_weight(e / CLASSES) * class_weight(e % CLASSES)).collect::>())?, + Codebook::scalar(1, &(0..floor::RANGE).map(|v| 1.0 / (1.0 + f64::from(v)).powf(1.6)).collect::>())?, + Codebook::lattice(4, 3, -1, 1, |v| if v == 0 { 0.92 } else { 0.04 })?, + Codebook::lattice(4, 3, -1, 1, |v| if v == 0 { 0.5 } else { 0.25 })?, + Codebook::lattice(2, 5, -2, 1, |v| 0.6f64.powi(v.abs()))?, + Codebook::lattice(2, 9, -4, 1, |v| 0.7f64.powi(v.abs()))?, + Codebook::lattice(2, 15, -7, 1, |v| 0.8f64.powi(v.abs()))?, + Codebook::lattice(2, 15, -105, 15, |v| 0.55f64.powi((v / 15).abs()))?, + Codebook::lattice(2, 15, -1575, 225, |v| 0.4f64.powi((v / 225).abs()))?, + ]; + Ok(Encoder { + cfg, + books, + floor: Floor1::new(BOOK_FLOOR), + mdct: Mdct::new(BLOCK), + window: mdct::window(BLOCK), + // the first block is centred on sample 0: the decoder's first output is input sample 0 + pending: vec![vec![0.0; HOP]; cfg.channels], + start: 0, + samples_in: 0, + blocks: 0, + finished: false, + // tones 4 dB (-1) to 30.4 dB (10) under their envelope; noise 14 dB above it (-1) + // to 4.1 dB (10) + tonal_db: 4.0 + 2.4 * (quality + 1.0), + noise_db: -14.0 + 0.9 * (quality + 1.0), + dead_zone: 0.8 - 0.025 * (quality + 1.0), + }) + } + + /// Channels per sample frame. + pub fn channels(&self) -> usize { + self.cfg.channels + } + + /// The quality in use, after clamping. + pub fn quality(&self) -> f32 { + self.cfg.quality + } + + /// Input samples per channel taken so far. + pub fn samples_in(&self) -> u64 { + self.samples_in + } + + /// The identification header packet (§4.2.2): the first packet of the stream. + pub fn identification(&self) -> Vec { + let mut w = BitWriter::new(); + w.write(1, 8); + w.bytes(b"vorbis"); + w.write(0, 32); + w.write(self.cfg.channels as u32, 8); + w.write(self.cfg.sample_rate, 32); + w.write(0, 32); // bit rates: unset (variable) + w.write(0, 32); + w.write(0, 32); + w.write(8, 4); // 256-sample short blocks (unused) + w.write(11, 4); // 2048-sample long blocks + w.flag(true); + w.finish() + } + + /// The setup header packet (§4.2.4): codebooks, floor, residue, mapping and mode. It follows + /// the comment header. + pub fn setup(&self) -> Vec { + let mut w = BitWriter::new(); + w.write(5, 8); + w.bytes(b"vorbis"); + w.write(self.books.len() as u32 - 1, 8); + for b in &self.books { + b.write_header(&mut w); + } + w.write(0, 6); // one time-domain transform placeholder + w.write(0, 16); + w.write(0, 6); // one floor + w.write(1, 16); + self.floor.write_header(&mut w); + w.write(0, 6); // one residue, type 2 + w.write(2, 16); + let end = (self.cfg.channels * HOP) as u32; + w.write(0, 24); + w.write(end, 24); + w.write(partition_size(self.cfg.channels) as u32 - 1, 24); + w.write(CLASSES as u32 - 1, 6); + w.write(BOOK_CLASS as u32, 8); + for c in &CASCADE { + let bits: u32 = c.iter().enumerate().map(|(j, b)| u32::from(b.is_some()) << j).sum(); + w.write(bits & 7, 3); + w.flag(bits > 7); + if bits > 7 { + w.write(bits >> 3, 5); + } + } + for c in &CASCADE { + for b in c.iter().flatten() { + w.write(*b as u32, 8); + } + } + w.write(0, 6); // one mapping, type 0 + w.write(0, 16); + w.flag(false); // one submap + w.flag(false); // no coupling + w.write(0, 2); + w.write(0, 8); // submap 0: time, floor 0, residue 0 + w.write(0, 8); + w.write(0, 8); + w.write(0, 6); // one mode: long blocks, mapping 0 + w.flag(true); + w.write(0, 16); + w.write(0, 16); + w.write(0, 8); + w.flag(true); + w.finish() + } + + /// Encodes planar samples (one slice per channel, all the same length, nominal range ±1); + /// returns the packets completed with their granule positions. NaN or infinite samples, or a + /// channel count or length mismatch, are an error. + pub fn encode(&mut self, planar: &[&[f32]]) -> Result, i64)>> { + if self.finished { + return Err(Error::InvalidArgument("Vorbis encoder: encode after finish")); + } + if planar.len() != self.cfg.channels { + return Err(Error::InvalidArgument("Vorbis encoder: one slice per channel")); + } + let len = planar.first().map_or(0, |p| p.len()); + if planar.iter().any(|p| p.len() != len) { + return Err(Error::InvalidArgument("Vorbis encoder: channels of different lengths")); + } + if planar.iter().any(|p| p.iter().any(|v| !v.is_finite())) { + return Err(Error::InvalidArgument("Vorbis encoder: NaN or infinite sample")); + } + for (b, p) in self.pending.iter_mut().zip(planar) { + b.extend_from_slice(p); + } + self.samples_in += len as u64; + self.drain() + } + + /// Encodes what remains, padded with silence; the last packet's granule position is the + /// input's length, so the decoder trims the padding (§A.2). + pub fn finish(&mut self) -> Result, i64)>> { + if self.finished { + return Ok(Vec::new()); + } + self.finished = true; + // block k ends output at k·HOP: blocks 0..=ceil(len / HOP) + let blocks = self.samples_in.div_ceil(HOP as u64) + 1; + let missing = blocks.saturating_sub(self.blocks) as usize; + self.compact(); + let need = (missing + 1) * HOP; + for b in self.pending.iter_mut() { + if b.len() < need { + b.resize(need, 0.0); + } + } + let mut out = Vec::new(); + for _ in 0..missing { + out.push(self.block()?); + } + Ok(out) + } + + fn drain(&mut self) -> Result, i64)>> { + let mut out = Vec::new(); + while self.pending.first().is_some_and(|b| b.len().saturating_sub(self.start) >= BLOCK) { + out.push(self.block()?); + } + self.compact(); + Ok(out) + } + + /// Drops the hops already encoded from the front of every channel buffer. + fn compact(&mut self) { + for b in self.pending.iter_mut() { + b.drain(..self.start.min(b.len())); + } + self.start = 0; + } + + /// Encodes the block at `start` in `pending` and advances by a hop. + fn block(&mut self) -> Result<(Vec, i64)> { + let ch = self.cfg.channels; + let mut w = BitWriter::new(); + w.write(0, 1); // audio packet; one mode, so no mode number + w.flag(true); // previous and next windows long + w.flag(true); + let mut residue = vec![0i32; ch * HOP]; + let mut any = false; + let mut x = vec![0f32; BLOCK]; + let mut coeffs = vec![0f32; HOP]; + for c in 0..ch { + let src = self.pending.get(c).and_then(|b| b.get(self.start..)).ok_or(Error::Internal("channel buffer"))?; + for (i, v) in x.iter_mut().enumerate() { + *v = src.get(i).copied().unwrap_or(0.0).clamp(-8.0, 8.0) * self.window[i]; + } + self.mdct.forward(&x, &mut coeffs, 2.0 / HOP as f64); + let peak = coeffs.iter().fold(0f32, |m, v| m.max(v.abs())); + // (a NaN peak cannot happen: samples are finite and clamped) + if peak <= 1e-7 { + w.flag(false); // floor unused: the channel is silent + continue; + } + any = true; + w.flag(true); + let want = self.floor_targets(&coeffs); + let curve = self.floor.encode(&mut w, &self.books[BOOK_FLOOR], &want, HOP); + for (k, (&v, &f)) in coeffs.iter().zip(&curve).enumerate() { + let r = if f > 0.0 { v / f } else { 0.0 }; + // a dead zone: values just past half a step are noise peaks, not worth a ±1 + let q = if r.abs() < self.dead_zone { 0.0 } else { r.round() }; + residue[k * ch + c] = if q.is_finite() { (q as i32).clamp(-MAX_RESIDUE, MAX_RESIDUE) } else { 0 }; + } + } + if any { + self.write_residue(&mut w, &residue); + } + let granule = if self.blocks == 0 { + 0 + } else { + let natural = self.blocks.saturating_mul(HOP as u64); + if self.finished { natural.min(self.samples_in) } else { natural } + }; + self.start += HOP; + self.blocks += 1; + Ok((w.finish(), i64::try_from(granule).unwrap_or(i64::MAX))) + } + + /// Floor values at the floor's points: the spectral envelope (power smoothed over a quarter of + /// the frequency, at least 8 bins, on each side) less an offset that grows with the region's tonality, from its + /// spectral flatness: noise needs a few dB of precision, tones much more. + fn floor_targets(&self, coeffs: &[f32]) -> Vec { + let n = coeffs.len(); + let mut cum = vec![0f64; n + 1]; + let mut cum_ln = vec![0f64; n + 1]; + for (i, &v) in coeffs.iter().enumerate() { + let p = f64::from(v) * f64::from(v); + cum[i + 1] = cum[i] + p; + cum_ln[i + 1] = cum_ln[i] + (p + 1e-30).ln(); + } + self.floor + .xs + .iter() + .map(|&x| { + let k = (x as usize).min(n - 1); + let half = (k / 4).max(8); + let (lo, hi) = (k.saturating_sub(half), (k + half + 1).min(n)); + let len = (hi - lo) as f64; + let am = (cum[hi] - cum[lo]) / len; + let gm = ((cum_ln[hi] - cum_ln[lo]) / len).exp(); + // spectral flatness: about -2.5 dB for noise, far below for a tone + let sfm_db = 10.0 * (gm / am.max(1e-30)).max(1e-30).log10(); + let tonality = ((-sfm_db - 2.5) / 20.0).clamp(0.0, 1.0); + let offset = tonality * f64::from(self.tonal_db) + (1.0 - tonality) * f64::from(self.noise_db); + let amp = (am.sqrt() * 10f64.powf(-offset / 20.0)).max(1e-7) as f32; + (floor::db_index(amp) + 1) / floor::MULTIPLIER as i32 + }) + .collect() + } + + /// Residue type 2 (§8.6.2 with the interleaved vector): classes per partition, then the + /// cascade's passes in the decoder's order. + fn write_residue(&self, w: &mut BitWriter, r: &[i32]) { + let partition = partition_size(self.cfg.channels); + let parts = r.len() / partition; + let classes: Vec = r.chunks(partition).map(class_of).collect(); + let per_word = self.books[BOOK_CLASS].dimensions as usize; + for pass in 0..3 { + let mut pc = 0; + while pc < parts { + if pass == 0 { + let mut e = 0usize; + for i in 0..per_word { + e = e * CLASSES + classes.get(pc + i).copied().unwrap_or(0); + } + self.books[BOOK_CLASS].put(w, e); + } + for _ in 0..per_word { + if pc >= parts { + break; + } + let class = classes[pc]; + if let Some(b) = CASCADE[class][pass] { + let book = &self.books[b]; + let dims = book.dimensions as usize; + let part = &r[pc * partition..(pc + 1) * partition]; + for chunk in part.chunks(dims) { + let v: Vec = chunk.iter().map(|&x| pass_value(x, class, pass)).collect(); + book.put(w, book.entry_of(&v)); + } + } + pc += 1; + } + } + } + } +} + +/// Relative frequency of each residue class (for the class book's code lengths). +fn class_weight(c: usize) -> f64 { + [0.3, 0.2, 0.12, 0.12, 0.1, 0.08, 0.06, 0.02].get(c).copied().unwrap_or(0.01) +} + +/// The part of residue value `v` coded in `pass` for `class`: the passes add up to `v`. +fn pass_value(v: i32, class: usize, pass: usize) -> i32 { + let split = |v: i32, step: i32| -> (i32, i32) { + let a = ((f64::from(v) / f64::from(step)).round() as i32).clamp(-7, 7); + (a * step, v - a * step) + }; + match (class, pass) { + (6, 0) => split(v, 15).0, + (6, _) => split(v, 15).1, + (7, 0) => split(v, 225).0, + (7, 1) => split(split(v, 225).1, 15).0, + (7, _) => split(split(v, 225).1, 15).1, + _ => v, + } +} + +/// The residue class of a partition: silent; ±1 with at most four nonzero values, or more; then +/// by the largest magnitude. +fn class_of(p: &[i32]) -> usize { + let m = p.iter().map(|v| v.abs()).max().unwrap_or(0); + match m { + 0 => 0, + 1 if p.iter().filter(|&&v| v != 0).count() <= 4 => 1, + 1 => 2, + 2 => 3, + 3..=4 => 4, + 5..=7 => 5, + 8..=112 => 6, + _ => 7, + } +} + +#[cfg(test)] +mod tests; diff --git a/crates/audio/src/vorbis/encoder/tests.rs b/crates/audio/src/vorbis/encoder/tests.rs new file mode 100644 index 000000000..b1ae87463 --- /dev/null +++ b/crates/audio/src/vorbis/encoder/tests.rs @@ -0,0 +1,285 @@ +//! Unit tests, and ffmpeg (as an external oracle only) decoding the encoder's streams. + +#![allow(clippy::unwrap_used)] + +use std::path::Path; +use std::process::Command; + +use super::book::{BitWriter, assign_codewords, float32_pack}; +use super::{BLOCK, Encoder, EncoderConfig, HOP, partition_size}; + +#[test] +fn codewords_follow_the_specification_example() { + // Vorbis I §3.2.1: lengths 2 4 4 4 4 2 3 3 + let words = assign_codewords(&[2, 4, 4, 4, 4, 2, 3, 3]).unwrap(); + assert_eq!(words, [0b00, 0b0100, 0b0101, 0b0110, 0b0111, 0b10, 0b110, 0b111]); + assert!(assign_codewords(&[1, 1, 1]).is_err(), "over-subscribed"); +} + +#[test] +fn every_book_is_a_complete_prefix_code() { + let e = Encoder::new(EncoderConfig::new(48_000, 2, 5.0)).unwrap(); + for (i, b) in e.books.iter().enumerate() { + let kraft: f64 = b.lengths.iter().map(|&l| 0.5f64.powi(l as i32)).sum(); + assert!((kraft - 1.0).abs() < 1e-9, "book {i}: Kraft sum {kraft}"); + assert!(b.lengths.iter().all(|&l| (1..=20).contains(&l))); + for (a, (&wa, &la)) in b.words.iter().zip(&b.lengths).enumerate() { + for (c, (&wc, &lc)) in b.words.iter().zip(&b.lengths).enumerate() { + if a != c && la <= lc { + assert_ne!(wc >> (lc - la), wa, "book {i}: codeword {a} is a prefix of {c}"); + } + } + } + } +} + +#[test] +fn float32_pack_is_exact_for_lattice_values() { + let unpack = |x: u32| { + let mant = f64::from(x & 0x1F_FFFF); + let exp = ((x >> 21) & 0x3FF) as i32 - 788; + let v = mant * 2f64.powi(exp); + if x & 0x8000_0000 != 0 { -v } else { v } + }; + for v in [1.0, -1.0, -2.0, -4.0, -7.0, -105.0, 15.0, -1575.0, 225.0, 0.5, 3.0] { + assert_eq!(unpack(float32_pack(v)), v); + } + assert_eq!(float32_pack(0.0), 0); +} + +#[test] +fn bits_are_packed_lsb_first() { + let mut w = BitWriter::new(); + w.write(0b1, 1); + w.write(0b10, 2); + w.write(0x1F, 5); + w.write(0xABC, 12); + assert_eq!(w.finish(), [0b1111_1101, 0xBC, 0x0A]); +} + +#[test] +fn hostile_input_is_an_error_not_a_panic() { + assert!(Encoder::new(EncoderConfig::new(48_000, 0, 3.0)).is_err()); + assert!(Encoder::new(EncoderConfig::new(48_000, 9, 3.0)).is_err()); + assert!(Encoder::new(EncoderConfig::new(0, 2, 3.0)).is_err()); + assert_eq!(Encoder::new(EncoderConfig::new(48_000, 1, f32::NAN)).unwrap().cfg.quality, 6.0, "NaN quality uses the default, as the export does"); + let mut e = Encoder::new(EncoderConfig::new(44_100, 2, 3.0)).unwrap(); + assert!(e.encode(&[&[0.0; 4]]).is_err()); + assert!(e.encode(&[&[0.0; 4], &[0.0; 3]]).is_err()); + assert!(e.encode(&[&[f32::NAN; 4], &[0.0; 4]]).is_err()); + let loud = vec![1e30f32; 5000]; + assert!(!e.encode(&[&loud, &loud]).unwrap().is_empty()); + assert!(!e.finish().unwrap().is_empty()); + assert!(e.finish().unwrap().is_empty()); + assert!(e.encode(&[&[0.0; 4], &[0.0; 4]]).is_err()); +} + +#[test] +fn granules_count_the_samples_and_trim_the_end() { + for len in [0usize, 1, 1023, 1024, 1025, 48_000] { + let mut e = Encoder::new(EncoderConfig::new(48_000, 1, 3.0)).unwrap(); + let x = vec![0.1f32; len]; + let mut packets = e.encode(&[&x]).unwrap(); + packets.extend(e.finish().unwrap()); + assert_eq!(packets.len(), len.div_ceil(HOP) + 1, "len {len}"); + // packet k ends output at k·1024 samples; the last one at the input's length + let want: Vec = (0..packets.len()).map(|k| ((k * HOP) as i64).min(len as i64)).collect(); + let got: Vec = packets.iter().map(|p| p.1).collect(); + assert_eq!(got, want, "len {len}"); + } +} + +// --- ffmpeg oracle --- + +/// An Ogg Vorbis file of the three headers and the audio packets, through craft-media's writer. +fn ogg(packets: &[(Vec, i64)]) -> Vec { + let mut w = craft_media::ogg::PageWriter::new(7); + let mut out = Vec::new(); + for (i, (p, g)) in packets.iter().enumerate() { + out.extend(if i < 3 { w.header(p) } else { w.packet(p, *g) }); + } + out.extend(w.finish(None)); + out +} + +fn comment() -> Vec { + craft_media::ogg::vorbis_comment("test", &[]) +} + +/// Encodes `input` (planar) and decodes it with ffmpeg's `decoder`; returns the planar output. +/// ffmpeg's decoders to check: its own `vorbis` always, and `libvorbis` when ffmpeg has it. +fn decoders(ff: &Path) -> Vec<&'static str> { + let list = + Command::new(ff).args(["-hide_banner", "-decoders"]).output().map(|o| String::from_utf8_lossy(&o.stdout).into_owned()).unwrap_or_default(); + if list.lines().any(|l| l.split_whitespace().nth(1) == Some("libvorbis")) { + vec!["vorbis", "libvorbis"] + } else { + eprintln!("SKIPPED (libvorbis decoder): this ffmpeg has no libvorbis; checking its own decoder only"); + vec!["vorbis"] + } +} + +fn round_trip(input: &[Vec], rate: u32, quality: f32, decoder: &str, name: &str) -> Option<(Vec>, usize)> { + let ff = craft_media_testkit::ffmpeg_or_skip("vorbis oracle")?; + let mut e = Encoder::new(EncoderConfig::new(rate, input.len(), quality)).unwrap(); + let mut packets = vec![(e.identification(), 0), (comment(), 0), (e.setup(), 0)]; + let n = input[0].len(); + let mut at = 0; + while at < n { + let end = (at + 4410).min(n); + let s: Vec<&[f32]> = input.iter().map(|c| &c[at..end]).collect(); + packets.extend(e.encode(&s).unwrap()); + at = end; + } + packets.extend(e.finish().unwrap()); + let bytes: usize = packets[3..].iter().map(|p| p.0.len()).sum(); + let dir = craft_media_testkit::fixtures_dir("audio/vorbis-encoder"); + let path = dir.join(format!("{name}.ogg")); + std::fs::write(&path, ogg(&packets)).unwrap(); + let out = Command::new(ff) + .args(["-v", "error", "-xerror", "-c:a", decoder, "-i"]) + .arg(&path) + .args(["-f", "f32le", "-c:a", "pcm_f32le", "-"]) + .output() + .unwrap(); + assert!(out.status.success() && out.stderr.is_empty(), "{name} {decoder}: {}", String::from_utf8_lossy(&out.stderr)); + let pcm: Vec = out.stdout.as_chunks::<4>().0.iter().map(|&b| f32::from_le_bytes(b)).collect(); + let ch = input.len(); + Some(((0..ch).map(|c| pcm.iter().skip(c).step_by(ch).copied().collect()).collect(), bytes)) +} + +fn snr_db(a: &[f32], b: &[f32]) -> f64 { + let (mut s, mut e) = (0f64, 0f64); + for (&x, &y) in a.iter().zip(b) { + s += f64::from(x) * f64::from(x); + e += (f64::from(x) - f64::from(y)).powi(2); + } + 10.0 * (s.max(1e-20) / e.max(1e-20)).log10() +} + +fn music(n: usize, seed: u32) -> Vec { + let mut r = seed; + (0..n) + .map(|i| { + let t = i as f32 / 48_000.0; + r = r.wrapping_mul(1_664_525).wrapping_add(1_013_904_223); + let noise = (r >> 8) as f32 / 8_388_608.0 - 1.0; + let chord: f32 = [220.0f32, 277.18, 329.63, 440.0] + .iter() + .enumerate() + .map(|(k, f)| (0.2 / (k + 1) as f32) * (2.0 * std::f32::consts::PI * f * t).sin()) + .sum(); + chord + 0.01 * noise + }) + .collect() +} + +/// Harmonic chords changing every half second, with decaying noise bursts (music-like). +fn chords(n: usize, seed: u32) -> Vec { + let mut r = seed; + let notes = [220.0f32, 277.2, 329.6, 440.0, 349.2, 523.3, 392.0, 659.3]; + (0..n) + .map(|i| { + let t = i as f32 / 48_000.0; + let chord = (t * 2.0) as usize; + let mut s = 0.0; + for k in 0..3 { + let f = notes[(chord * 3 + k * 2) % notes.len()]; + for h in 1..6 { + s += (2.0 * std::f32::consts::PI * f * h as f32 * t).sin() * 0.08 / h as f32; + } + } + r = r.wrapping_mul(1_664_525).wrapping_add(1_013_904_223); + let env = (-(t * 2.0).fract() * 3.0).exp(); + s * (0.5 + 0.5 * env) + 0.02 * ((r >> 8) as f32 / 8_388_608.0 - 1.0) * env + }) + .collect() +} + +/// ffmpeg's own Vorbis decoder and libvorbis read every stream exactly as long as the input, +/// with SNR floors about 6 dB (chords 4 dB) under what the encoder measured when written, and bit +/// rates near libvorbis's for the same quality. +/// A signal, and per quality the SNR floor and the bit-rate range (kbit/s). +type Case = (&'static str, Vec>, [(f32, f64, std::ops::Range); 5]); + +#[test] +fn ffmpeg_decodes_every_quality() { + let Some(ff) = craft_media_testkit::ffmpeg_or_skip("vorbis oracle") else { return }; + let n = 48_000 * 3; + let cases: [Case; 2] = [ + ( + "music", + vec![music(n, 1), music(n, 2).iter().map(|v| v * 0.7).collect()], + [(-1.0, 3.0, 15..90), (0.0, 6.5, 20..110), (3.0, 12.0, 60..180), (6.0, 18.5, 120..280), (10.0, 25.5, 180..400)], + ), + ( + "chords", + vec![chords(n, 1), chords(n, 2).iter().map(|v| v * 0.8).collect()], + [(-1.0, 4.5, 20..90), (0.0, 6.0, 25..110), (3.0, 10.0, 70..200), (6.0, 14.0, 130..300), (10.0, 19.0, 180..420)], + ), + ]; + for (name, input, levels) in &cases { + for (q, floor, kbps) in levels { + for decoder in decoders(&ff) { + let Some((out, bytes)) = round_trip(input, 48_000, *q, decoder, &format!("{name}{q}")) else { return }; + assert!(out.iter().all(|c| c.len() == n), "{name} q{q} {decoder}: length {}", out[0].len()); + let snr = snr_db(&input[0][4096..], &out[0][4096..]); + assert!(snr >= *floor, "{name} q{q} {decoder}: {snr:.2} dB < {floor}"); + let rate = bytes * 8 / 3 / 1000; + assert!(kbps.contains(&rate), "{name} q{q}: {rate} kbit/s outside {kbps:?}"); + } + } + } +} + +/// Mono, 44.1 kHz, 3.0, 5.1 and 6.1 streams decode too, with libvorbis as well (which reads each +/// residue partition from channel 0); a silent channel costs one bit per block. +#[test] +fn channel_counts_and_rates() { + let Some(ff) = craft_media_testkit::ffmpeg_or_skip("vorbis oracle") else { return }; + let n = 44_100; + for (rate, ch) in [(44_100u32, 1usize), (22_050, 2), (48_000, 3), (48_000, 6), (48_000, 7)] { + let mut input: Vec> = (0..ch).map(|c| chords(n, c as u32 + 1).iter().map(|v| v * (1.0 - 0.1 * c as f32)).collect()).collect(); + if ch == 6 { + input[3] = vec![0.0; n]; + } + for decoder in decoders(&ff) { + let Some((mut out, _)) = round_trip(&input, rate, 5.0, decoder, &format!("ch{ch}_{rate}_{decoder}")) else { return }; + if decoder == "vorbis" { + // ffmpeg's decoder gives WAV order (5.1: FL FR FC LFE BL BR); the stream is in Vorbis + // order (L C R Ls Rs LFE), which libvorbis passes through + let wav_to_vorbis: &[usize] = match ch { + 3 => &[0, 2, 1], + 6 => &[0, 2, 1, 4, 5, 3], + 7 => &[0, 2, 1, 5, 6, 4, 3], + _ => &[0, 1], + }; + if ch > 2 { + out = wav_to_vorbis.iter().map(|&i| out[i].clone()).collect(); + } + } + assert_eq!(out.len(), ch); + for (c, (a, b)) in input.iter().zip(&out).enumerate() { + assert_eq!(b.len(), n, "{ch} ch {rate} Hz {decoder}: channel {c} length"); + if a.iter().all(|&v| v == 0.0) { + assert!(b.iter().all(|v| v.abs() < 1e-6), "silent channel {c}"); + } else { + let snr = snr_db(&a[4096..], &b[4096..]); + assert!(snr > 8.0, "{ch} ch {rate} Hz {decoder} channel {c}: {snr:.1} dB"); + } + } + } + } +} + +/// Every partition holds whole sample frames (libvorbis decodes each one from channel 0), splits +/// the interleaved residue evenly and fits every book's dimension. +#[test] +fn partitions_hold_whole_frames() { + for ch in 1..=8 { + let p = partition_size(ch); + assert_eq!(p % ch, 0, "{ch} channels"); + assert_eq!((ch * BLOCK / 2) % p, 0, "{ch} channels"); + assert_eq!(p % 4, 0, "{ch} channels"); + } +} diff --git a/crates/audio/src/vorbis/mod.rs b/crates/audio/src/vorbis/mod.rs new file mode 100644 index 000000000..29c423f1d --- /dev/null +++ b/crates/audio/src/vorbis/mod.rs @@ -0,0 +1,4 @@ +//! Vorbis I (feature `vorbis-encode`): the pure-Rust [`encoder`]. Decoding Vorbis goes through +//! [`crate::decode`]. + +pub mod encoder; diff --git a/crates/audio/src/wav_markers.rs b/crates/audio/src/wav_markers.rs new file mode 100644 index 000000000..0bf4e7568 --- /dev/null +++ b/crates/audio/src/wav_markers.rs @@ -0,0 +1,721 @@ +//! Markers a WAV file carries, read from its RIFF chunks. +//! +//! Two encodings are read into separate lists for the caller to map and merge: +//! - **Cue points** (the standard): a `cue ` chunk with one point per marker, named by `labl` (or +//! `note`) sub-chunks of a `LIST` `adtl` chunk, and given a length by `ltxt`. +//! - **XMP** (Adobe's tools): an `_PMX` chunk holding an XMP packet whose `xmpDM:markers` list has +//! one entry per marker (`xmpDM:startTime`, `xmpDM:duration`, `xmpDM:name`), counted in frames +//! of `xmpDM:frameRate` (`f44100` for samples at 44.1 kHz), or with an explicit +//! FrameCount rate suffix (`15f24`, `15f30000s1001`). Durations with a distinct rate +//! are converted to the marker start rate without app-specific time quantization. +//! +//! RF64 files (WAVs over 4 GB) carry a `0xFFFFFFFF` size for a chunk too big for 32 bits and give its real size in a +//! `ds64` chunk; those sizes are resolved so the chunks after the audio are still found. +//! +//! The file is untrusted input: every read is bounds-checked, sizes use checked arithmetic, and +//! the number of markers, of `adtl` sub-chunks looked at and the size of the XMP packet are capped. + +use std::collections::{HashMap, HashSet}; + +/// Marker coordinates before app-specific time quantization. Explicit XMP duration +/// rates are normalized to the start coordinate rate; bare counts retain the legacy +/// track/file-rate interpretation. +#[derive(Clone, Debug, PartialEq)] +pub struct Marker { + /// Start in frames at `rate`. + pub start: f64, + /// Duration in frames at `rate`. + pub duration: f64, + /// Frames per second; consumers decide how invalid or absent rates map to their timeline. + pub rate: f64, + /// Label text. + pub name: String, +} +/// Separate source lists: rounding, deduplication, ordering and labels belong to the caller. +#[derive(Clone, Debug, Default, PartialEq)] +pub struct Markers { + /// RIFF cue/adtl markers in file order. + pub cue: Vec, + /// XMP markers in packet order. + pub xmp: Vec, + /// A chunk-inventory or cue-label retention limit omitted metadata. Marker coordinates + /// remain available when only their label was omitted. Existing marker/XMP prefix limits + /// are separate and do not set this flag. + pub limited: bool, +} + +/// Most markers read from one file. +pub const MAX_MARKERS: usize = 10_000; +/// Largest XMP packet read (bytes). +const MAX_XMP: usize = 4 << 20; +/// Most `adtl` sub-chunks (`labl`, `note`, `ltxt`) looked at in one file: a few per cue point is plenty. +const MAX_ADTL_RECORDS: usize = 4 * MAX_MARKERS; +/// Most entries read from an RF64 file's `ds64` size table. +const MAX_DS64_TABLE: usize = 1_024; +/// Maximum retained RIFF chunk descriptors (including empty chunks). +pub const MAX_CHUNKS: usize = 100_000; +/// Maximum UTF-8 bytes in one retained cue label. +pub const MAX_LABEL_BYTES: usize = 64 * 1024; +/// Aggregate cue label bytes, enforced separately for the name map and emitted markers. +pub const MAX_LABEL_TOTAL_BYTES: usize = 4 * 1024 * 1024; + +fn u32_at(b: &[u8], at: usize) -> Option { + let s = b.get(at..at.checked_add(4)?)?; + Some(u32::from_le_bytes(s.try_into().ok()?)) +} + +fn u64_at(b: &[u8], at: usize) -> Option { + let s = b.get(at..at.checked_add(8)?)?; + Some(u64::from_le_bytes(s.try_into().ok()?)) +} + +/// The text of a NUL-terminated label (the rest of the chunk), lossily decoded. +fn label(b: &[u8]) -> Option { + // Look only as far as one label can be retained; never copy an arbitrary chunk body. + let end = b.iter().take(MAX_LABEL_BYTES + 1).position(|&c| c == 0).unwrap_or(b.len()); + if end > MAX_LABEL_BYTES { + return None; + } + let text = String::from_utf8_lossy(b.get(..end)?); + let text = text.trim(); + (text.len() <= MAX_LABEL_BYTES).then(|| text.to_string()) +} + +/// Whether `bytes` start a RIFF/WAVE file. +pub fn is_wav(bytes: &[u8]) -> bool { + matches!(bytes.get(0..4), Some(b"RIFF" | b"RF64")) && bytes.get(8..12) == Some(b"WAVE") +} + +/// The sizes an RF64 file's `ds64` chunk gives for chunks too big for the 32-bit size field. +struct Ds64 { + /// The `data` chunk's size. + data: u64, + /// Other chunks: `(id, size)`. + table: Vec<([u8; 4], u64)>, +} + +impl Ds64 { + /// The real size of chunk `id`, when the file says. + fn size_of(&self, id: &[u8; 4]) -> Option { + if id == b"data" { Some(self.data) } else { self.table.iter().find(|(t, _)| t == id).map(|(_, size)| *size) } + } +} + +/// The `ds64` chunk of an RF64 file (`RF64`, size, `WAVE`, then `ds64`); `None` for any other file or one whose +/// `ds64` doesn't fit. +fn ds64(bytes: &[u8]) -> Option { + if bytes.get(0..4) != Some(b"RF64") || bytes.get(12..16) != Some(b"ds64") { + return None; + } + let size = usize::try_from(u32_at(bytes, 16)?).ok()?; + let body = bytes.get(20..20usize.checked_add(size)?)?; + // riff size (8), data size (8), sample count (8), table length (4), then `(id, size)` entries of 12 bytes + let data = u64_at(body, 8)?; + let entries = usize::try_from(u32_at(body, 24)?).ok()?; + let mut table = Vec::new(); + for i in 0..entries.min(MAX_DS64_TABLE) { + let at = 28usize.checked_add(i.checked_mul(12)?)?; + let id = <[u8; 4]>::try_from(body.get(at..at.checked_add(4)?)?).ok()?; + table.push((id, u64_at(body, at + 4)?)); + } + Some(Ds64 { data, table }) +} + +/// The chunks of a WAVE file as `(id, body)`, stopping at the first one that doesn't fit. +fn chunks<'a>(bytes: &'a [u8], limited: &mut bool) -> Vec<([u8; 4], &'a [u8])> { + let big = ds64(bytes); + let mut out = Vec::new(); + let mut at = 12usize; + while let Some(head) = bytes.get(at..at.saturating_add(8)) { + if out.len() == MAX_CHUNKS { + *limited = true; + break; + } + let (Ok(id), Some(field)) = (<[u8; 4]>::try_from(&head[..4]), u32_at(head, 4)) else { break }; + // In an RF64 file a size field of 0xFFFFFFFF stands for the size `ds64` gives. + let size = match &big { + Some(d) if field == u32::MAX => d.size_of(&id).unwrap_or(u64::from(field)), + _ => u64::from(field), + }; + let start = at.saturating_add(8); + let Some(end) = usize::try_from(size).ok().and_then(|n| start.checked_add(n)) else { break }; + let Some(body) = bytes.get(start..end) else { break }; + out.push((id, body)); + // Chunks are padded to an even size. + at = end.saturating_add(end & 1); + } + out +} + +/// Read bounded cue/adtl and XMP metadata without interpreting application timeline policy. +pub fn read(bytes: &[u8]) -> Markers { + if !is_wav(bytes) { + return Markers::default(); + } + let mut limited = false; + let chunks = chunks(bytes, &mut limited); + let rate = chunks.iter().find(|(id, _)| id == b"fmt ").and_then(|(_, b)| u32_at(b, 4)).map_or(0.0, f64::from); + let cue = cue_markers(&chunks, rate, &mut limited); + let xmp = chunks.iter().find(|(id, _)| id == b"_PMX").map_or_else(Vec::new, |(_, xmp)| xmp_markers(xmp, rate)); + Markers { cue, xmp, limited } +} + +/// Cue points named and sized by the `adtl` list. +fn cue_markers(chunks: &[([u8; 4], &[u8])], rate: f64, limited: &mut bool) -> Vec { + let Some((_, cue)) = chunks.iter().find(|(id, _)| id == b"cue ") else { return Vec::new() }; + let count = (u32_at(cue, 0).unwrap_or(0) as usize).min(MAX_MARKERS); + // The cue points first: `(id, offset in samples)`. + let mut points = Vec::new(); + for i in 0..count { + let Some(point) = cue.get(4usize.saturating_add(i.saturating_mul(24))..) else { break }; + let (Some(id), Some(offset)) = (u32_at(point, 0), u32_at(point, 20)) else { break }; + points.push((id, offset)); + } + // Only the names and lengths of those ids are kept, so a hostile `adtl` list cannot grow them past the cue count, + // and each record is a lookup, not a scan. + let wanted: HashSet = points.iter().map(|(id, _)| *id).collect(); + let mut names: HashMap = HashMap::new(); + let mut lengths: HashMap = HashMap::new(); + let mut name_bytes = 0usize; + let mut records = 0usize; + 'lists: for (_, list) in chunks.iter().filter(|(id, b)| id == b"LIST" && b.get(0..4) == Some(b"adtl")) { + let mut at = 4usize; + while let Some(head) = list.get(at..at.saturating_add(8)) { + let (Ok(id), Some(size)) = (<[u8; 4]>::try_from(&head[..4]), u32_at(head, 4)) else { break }; + let start = at.saturating_add(8); + let Some(end) = start.checked_add(size as usize) else { break }; + let Some(body) = list.get(start..end) else { break }; + records += 1; + if records > MAX_ADTL_RECORDS { + break 'lists; + } + if let Some(cue_id) = u32_at(body, 0).filter(|c| wanted.contains(c)) { + match &id { + // A `labl` is the name; a `note` only when there is none. + b"labl" | b"note" if id == *b"labl" || !names.contains_key(&cue_id) => { + let without_old = name_bytes.saturating_sub(names.get(&cue_id).map_or(0, String::len)); + let text = label(body.get(4..).unwrap_or_default()) + .filter(|text| text.len() <= MAX_LABEL_TOTAL_BYTES.saturating_sub(without_old)) + .unwrap_or_else(|| { + *limited = true; + String::new() + }); + name_bytes = without_old + text.len(); + names.insert(cue_id, text); + } + // The first `ltxt` of a cue point gives its length. + b"ltxt" => { + lengths.entry(cue_id).or_insert_with(|| u32_at(body, 4).unwrap_or(0)); + } + _ => {} + } + } + at = end.saturating_add(end & 1); + } + } + let mut emitted_bytes = 0usize; + points + .into_iter() + .map(|(id, offset)| { + let comment = match names.get(&id) { + Some(text) if text.len() <= MAX_LABEL_TOTAL_BYTES.saturating_sub(emitted_bytes) => { + emitted_bytes += text.len(); + text.clone() + } + Some(_) => { + *limited = true; + String::new() + } + None => String::new(), + }; + let length = lengths.get(&id).copied().unwrap_or(0); + Marker { start: f64::from(offset), duration: f64::from(length), rate, name: comment } + }) + .collect() +} + +/// The text of `name` in an XMP entry, as an attribute (`name="v"`) or an element +/// (`v`). +fn xmp_value(entry: &str, name: &str) -> Option { + // Each entry starts just after ` in_tag = true, + b'>' => in_tag = false, + b'\'' | b'"' if in_tag => { + let quote = bytes[at]; + at += 1; + while at < bytes.len() && bytes[at] != quote { + at += 1; + } + } + _ if in_tag && (at == 0 || xml_space(bytes[at - 1])) && bytes[at..].starts_with(name.as_bytes()) => { + let mut value = at + name.len(); + while value < bytes.len() && xml_space(bytes[value]) { + value += 1; + } + if bytes.get(value) == Some(&b'=') { + value += 1; + while value < bytes.len() && xml_space(bytes[value]) { + value += 1; + } + if let Some("e @ (b'\'' | b'"')) = bytes.get(value) { + let start = value + 1; + let end = bytes.get(start..)?.iter().position(|&b| b == quote)? + start; + return Some(unescape(entry.get(start..end)?)); + } + } + } + _ => {} + } + at += 1; + } + let open = format!("<{name}>"); + let i = entry.find(&open)?; + let rest = entry.get(i + open.len()..)?; + Some(unescape(rest.get(..rest.find('<')?)?)) +} + +fn xml_space(b: u8) -> bool { + matches!(b, b' ' | b'\t' | b'\r' | b'\n') +} + +fn unescape(s: &str) -> String { + let mut out = String::with_capacity(s.len()); + let mut rest = s; + while let Some(i) = rest.find('&') { + out.push_str(rest.get(..i).unwrap_or_default()); + let tail = rest.get(i..).unwrap_or_default(); + let Some(semi) = tail.as_bytes().iter().take(11).position(|b| *b == b';') else { + out.push('&'); + rest = tail.get(1..).unwrap_or_default(); + continue; + }; + let ent = tail.get(1..semi).unwrap_or_default(); + match ent { + "amp" => out.push('&'), + "lt" => out.push('<'), + "gt" => out.push('>'), + "quot" => out.push('"'), + "apos" => out.push('\''), + _ => { + let code = ent + .strip_prefix("#x") + .and_then(|h| u32::from_str_radix(h, 16).ok()) + .or_else(|| ent.strip_prefix('#').and_then(|d| d.parse().ok())); + match code.and_then(char::from_u32) { + Some(c) => out.push(c), + None => out.push_str(tail.get(..=semi).unwrap_or_default()), + } + } + } + rest = tail.get(semi + 1..).unwrap_or_default(); + } + out.push_str(rest); + out.trim().to_string() +} + +/// A frame rate as XMP writes it: `f44100`, or `f30000s1001` (rate over scale). +fn xmp_rate(text: &str) -> Option { + let t = text.trim().strip_prefix('f')?; + let (num, den) = t.split_once('s').unwrap_or((t, "1")); + let (n, d): (f64, f64) = (num.parse().ok()?, den.parse().ok()?); + (n.is_finite() && d.is_finite() && n > 0.0 && d > 0.0 && (n / d).is_finite() && n / d > 0.0).then_some(n / d) +} + +/// Adobe XMP FrameCount: a count optionally followed by `fRATE[sSCALE]`. +/// Bare counts retain the caller's legacy track/file-rate interpretation. +fn xmp_count(text: &str, default_rate: f64) -> Option<(f64, f64)> { + let text = text.trim(); + let (count, rate) = match text.find('f') { + Some(at) => (text.get(..at)?, xmp_rate(text.get(at..)?)?), + None => (text, default_rate), + }; + let count: f64 = count.parse().ok()?; + (count.is_finite() && rate.is_finite() && rate > 0.0).then_some((count, rate)) +} + +/// The markers of an XMP packet's `xmpDM:markers` entries. +fn xmp_markers(packet: &[u8], file_rate: f64) -> Vec { + let text = String::from_utf8_lossy(packet.get(..packet.len().min(MAX_XMP)).unwrap_or_default()); + let mut out = Vec::new(); + // The frame rate belongs to the track an entry follows. + let mut rate = file_rate; + for entry in text.split("= MAX_MARKERS { + break; + } + } + out +} + +#[cfg(test)] +mod tests { + use super::*; + + struct TimedMarker { + time: f64, + duration: f64, + comment: String, + } + fn read(bytes: &[u8]) -> Vec { + let metadata = super::read(bytes); + let time = |frames: f64, rate: f64| if rate > 0.0 && frames.is_finite() { (frames / rate).max(0.0) } else { 0.0 }; + let map = |m: Marker| TimedMarker { time: time(m.start, m.rate), duration: time(m.duration, m.rate), comment: m.name }; + let mut out: Vec<_> = metadata.cue.into_iter().map(map).collect(); + let mut seen: HashSet<_> = out.iter().map(|m| (m.comment.clone(), m.time.to_bits())).collect(); + for m in metadata.xmp.into_iter().map(map) { + if seen.insert((m.comment.clone(), m.time.to_bits())) { + out.push(m); + } + } + out.truncate(MAX_MARKERS); + out.sort_by(|a, b| a.time.total_cmp(&b.time)); + out + } + + fn chunk(id: &[u8; 4], body: &[u8]) -> Vec { + let mut v = id.to_vec(); + v.extend((body.len() as u32).to_le_bytes()); + v.extend(body); + if body.len() % 2 == 1 { + v.push(0); + } + v + } + + fn fmt(rate: u32) -> Vec { + let mut b = Vec::new(); + b.extend(1u16.to_le_bytes()); // PCM + b.extend(1u16.to_le_bytes()); // mono + b.extend(rate.to_le_bytes()); + b.extend((rate * 2).to_le_bytes()); + b.extend(2u16.to_le_bytes()); + b.extend(16u16.to_le_bytes()); + chunk(b"fmt ", &b) + } + + fn wav(chunks: &[Vec]) -> Vec { + let body: Vec = chunks.iter().flatten().copied().collect(); + let mut v = b"RIFF".to_vec(); + v.extend((body.len() as u32 + 4).to_le_bytes()); + v.extend(b"WAVE"); + v.extend(body); + v + } + + fn cue(points: &[(u32, u32)]) -> Vec { + let mut b = (points.len() as u32).to_le_bytes().to_vec(); + for (id, offset) in points { + b.extend(id.to_le_bytes()); + b.extend(0u32.to_le_bytes()); + b.extend(b"data"); + b.extend(0u32.to_le_bytes()); + b.extend(0u32.to_le_bytes()); + b.extend(offset.to_le_bytes()); + } + chunk(b"cue ", &b) + } + + fn labl(id: &[u8; 4], cue_id: u32, text: &str) -> Vec { + let mut b = cue_id.to_le_bytes().to_vec(); + b.extend(text.as_bytes()); + b.push(0); + chunk(id, &b) + } + + fn adtl(subs: &[Vec]) -> Vec { + let mut b = b"adtl".to_vec(); + b.extend(subs.iter().flatten()); + chunk(b"LIST", &b) + } + + #[test] + fn cue_points_are_named_by_the_adtl_list_and_timed_in_samples() { + let file = wav(&[ + fmt(44100), + chunk(b"data", &[0; 8]), + cue(&[(2, 88200), (1, 22050)]), + adtl(&[labl(b"labl", 1, "SEQ01_Prologue"), labl(b"labl", 2, "SEQ02_Vallee_P1_Agriculture")]), + ]); + let m = read(&file); + assert_eq!(m.len(), 2); + assert_eq!((m[0].comment.as_str(), m[0].time), ("SEQ01_Prologue", 0.5)); + assert_eq!((m[1].comment.as_str(), m[1].time), ("SEQ02_Vallee_P1_Agriculture", 2.0)); + } + + #[test] + fn a_note_names_a_cue_only_when_it_has_no_label_and_ltxt_gives_a_length() { + let mut ltxt = 1u32.to_le_bytes().to_vec(); + ltxt.extend(44100u32.to_le_bytes()); + ltxt.extend([0; 16]); + let file = wav(&[ + fmt(44100), + cue(&[(1, 0), (2, 44100)]), + adtl(&[labl(b"note", 1, "from note"), chunk(b"ltxt", <xt), labl(b"labl", 2, "named"), labl(b"note", 2, "ignored")]), + ]); + let m = read(&file); + assert_eq!((m[0].comment.as_str(), m[0].duration), ("from note", 1.0)); + assert_eq!(m[1].comment, "named"); + } + + const XMP_ELEMENTS: &str = r#"MarkersCuef44100441000One & Two13230022050Three"#; + + #[test] + fn xmp_markers_are_read_as_elements_in_frames_of_the_track_rate() { + let file = wav(&[fmt(48000), chunk(b"_PMX", XMP_ELEMENTS.as_bytes())]); + let m = read(&file); + assert_eq!(m.len(), 2); + // 44100 frames at f44100 is one second, whatever the file's own rate. + assert_eq!((m[0].comment.as_str(), m[0].time), ("One & Two", 1.0)); + assert_eq!((m[1].comment.as_str(), m[1].time, m[1].duration), ("Three", 3.0, 0.5)); + } + + #[test] + fn xmp_markers_are_read_as_attributes_and_use_the_file_rate_without_a_track_rate() { + let xmp = r#""#; + let m = read(&wav(&[fmt(48000), chunk(b"_PMX", xmp.as_bytes())])); + assert_eq!((m[0].comment.as_str(), m[0].time), ("Attr Ă©", 0.5)); + } + + #[test] + fn a_marker_in_both_encodings_is_listed_once() { + let xmp = r#""#; + let file = wav(&[fmt(44100), cue(&[(1, 44100)]), adtl(&[labl(b"labl", 1, "Same")]), chunk(b"_PMX", xmp.as_bytes())]); + let names: Vec = read(&file).into_iter().map(|m| m.comment).collect(); + assert_eq!(names, ["Same", "Only XMP"]); + } + + #[test] + fn damaged_or_hostile_files_read_as_far_as_they_go_without_panicking() { + assert!(read(b"").is_empty()); + assert!(read(b"RIFF").is_empty()); + assert!(read(b"not a wav file at all").is_empty()); + let good = wav(&[fmt(44100), cue(&[(1, 44100)]), adtl(&[labl(b"labl", 1, "Kept")])]); + // Every truncation is safe. + for n in 0..good.len() { + let _ = read(&good[..n]); + } + // A cue count far beyond the chunk, and chunk sizes past the end of the file. + let mut huge = 0xFFFF_FFFFu32.to_le_bytes().to_vec(); + huge.extend([0; 24]); + let m = read(&wav(&[fmt(44100), chunk(b"cue ", &huge)])); + assert!(m.len() <= 1); + let mut lying = good.clone(); + lying.extend(b"LIST\xff\xff\xff\xffadtl"); + assert_eq!(read(&lying).len(), 1); + // No sample rate: times are zero, not a division by zero. + let m = read(&wav(&[cue(&[(1, 44100)])])); + assert_eq!(m[0].time, 0.0); + } + + /// A chunk whose size field says `field` (an RF64 file's `0xFFFFFFFF`) while its body is `body`. + fn chunk_with_size_field(id: &[u8; 4], field: u32, body: &[u8]) -> Vec { + let mut v = id.to_vec(); + v.extend(field.to_le_bytes()); + v.extend(body); + if body.len() % 2 == 1 { + v.push(0); + } + v + } + + /// An RF64 `ds64` chunk: the `data` chunk's size and a table of other chunks' sizes. + fn ds64_chunk(data_size: u64, table: &[([u8; 4], u64)]) -> Vec { + let mut b = 0u64.to_le_bytes().to_vec(); // riff size (not used) + b.extend(data_size.to_le_bytes()); + b.extend(0u64.to_le_bytes()); // sample count + b.extend((table.len() as u32).to_le_bytes()); + for (id, size) in table { + b.extend(id); + b.extend(size.to_le_bytes()); + } + chunk(b"ds64", &b) + } + + fn rf64(chunks: &[Vec]) -> Vec { + let body: Vec = chunks.iter().flatten().copied().collect(); + let mut v = b"RF64".to_vec(); + v.extend(u32::MAX.to_le_bytes()); + v.extend(b"WAVE"); + v.extend(body); + v + } + + #[test] + fn rf64_markers_after_a_data_chunk_of_sentinel_size_are_read() { + let audio = [0u8; 8]; + let file = rf64(&[ + ds64_chunk(audio.len() as u64, &[]), + fmt(44100), + chunk_with_size_field(b"data", u32::MAX, &audio), + cue(&[(1, 44100)]), + adtl(&[labl(b"labl", 1, "After the audio")]), + ]); + let m = read(&file); + assert_eq!(m.len(), 1); + assert_eq!((m[0].comment.as_str(), m[0].time), ("After the audio", 1.0)); + } + + #[test] + fn rf64_sizes_of_other_chunks_come_from_the_ds64_table() { + let junk = [7u8; 6]; + let tail = || [cue(&[(1, 8000)]), adtl(&[labl(b"labl", 1, "Found")])].concat(); + let with_entry = rf64(&[ds64_chunk(0, &[(*b"junk", junk.len() as u64)]), fmt(8000), chunk_with_size_field(b"junk", u32::MAX, &junk), tail()]); + assert_eq!(read(&with_entry)[0].comment, "Found"); + // Without the table entry the sentinel is all there is, and what follows is not reached. + let without = rf64(&[ds64_chunk(0, &[]), fmt(8000), chunk_with_size_field(b"junk", u32::MAX, &junk), tail()]); + assert!(read(&without).is_empty()); + } + + #[test] + fn hostile_rf64_size_tables_read_as_far_as_they_go_without_panicking() { + // sizes that overflow when added to an offset + for data in [u64::MAX, u64::MAX - 7, 1 << 63] { + let file = + rf64(&[ds64_chunk(data, &[(*b"junk", u64::MAX)]), fmt(8000), chunk_with_size_field(b"data", u32::MAX, &[0; 4]), cue(&[(1, 0)])]); + assert!(read(&file).is_empty()); + } + // a table that claims far more entries than its chunk holds + let mut liar = ds64_chunk(0, &[]); + liar[8 + 24..8 + 28].copy_from_slice(&u32::MAX.to_le_bytes()); + let _ = read(&rf64(&[liar, fmt(8000), cue(&[(1, 0)])])); + // every truncation of a good RF64 file is safe + let good = rf64(&[ + ds64_chunk(4, &[]), + fmt(44100), + chunk_with_size_field(b"data", u32::MAX, &[0; 4]), + cue(&[(1, 44100)]), + adtl(&[labl(b"labl", 1, "Kept")]), + ]); + assert_eq!(read(&good).len(), 1); + for n in 0..good.len() { + let _ = read(&good[..n]); + } + } + + #[test] + fn a_huge_adtl_list_is_read_in_linear_time_and_only_for_the_cue_points_that_exist() { + // Records for ids that have no cue point are skipped, not stored, and the number looked at is capped. + let junk = |n: u32| (1000..1000 + n).map(|id| labl(b"labl", id, "n")).collect::>(); + let started = std::time::Instant::now(); + // the real label comes first: read + let mut subs = vec![labl(b"labl", 1, "Real")]; + subs.extend(junk(100_000)); + let m = read(&wav(&[fmt(44100), cue(&[(1, 44100)]), adtl(&subs)])); + assert_eq!(m[0].comment, "Real"); + // it comes after more records than are looked at: not read, and the marker is still there + let mut subs = junk(MAX_ADTL_RECORDS as u32 + 10); + subs.push(labl(b"labl", 1, "Too late")); + let m = read(&wav(&[fmt(44100), cue(&[(1, 44100)]), adtl(&subs)])); + assert_eq!((m.len(), m[0].comment.as_str()), (1, "")); + assert!(started.elapsed() < std::time::Duration::from_secs(5), "took {:?}", started.elapsed()); + } + + #[test] + fn metadata_keeps_original_coordinates_and_separate_sources() { + let xmp = br#""#; + let file = wav(&[fmt(44100), cue(&[(1, 1)]), adtl(&[labl(b"labl", 1, "Same")]), chunk(b"_PMX", xmp)]); + let data = super::read(&file); + assert_eq!(data.cue[0].start, 1.0); + assert_eq!(data.cue[0].rate, 44100.0); + assert_eq!(data.xmp[0].start, 1.0); + assert_eq!(data.xmp[0].duration, 2.0); + assert_eq!(data.cue.len(), 1); + assert_eq!(data.xmp.len(), 1); + } + + #[test] + fn repeated_unterminated_entities_preserve_text_with_bounded_lookahead() { + let text = "&".repeat(100_000); + assert_eq!(unescape(&text), text); + assert_eq!(unescape("A A & &unknown;"), "A A & &unknown;"); + } + #[test] + fn repeated_cue_ids_cannot_multiply_label_storage() { + let text = "x".repeat(MAX_LABEL_BYTES); + let points: Vec<_> = (0..MAX_MARKERS).map(|i| (1, i as u32)).collect(); + let file = wav(&[fmt(44100), cue(&points), adtl(&[labl(b"labl", 1, &text)])]); + let metadata = super::read(&file); + assert_eq!(metadata.cue.len(), MAX_MARKERS); + assert_eq!(metadata.cue.iter().map(|m| m.name.len()).sum::(), MAX_LABEL_TOTAL_BYTES); + assert!(metadata.limited); + assert_eq!(metadata.cue[0].name, text); + assert!(metadata.cue[MAX_MARKERS - 1].name.is_empty()); + assert_eq!(metadata.cue[MAX_MARKERS - 1].start, (MAX_MARKERS - 1) as f64); + } + + #[test] + fn cue_name_map_and_single_label_are_bounded() { + let count = MAX_LABEL_TOTAL_BYTES / MAX_LABEL_BYTES + 2; + let text = "x".repeat(MAX_LABEL_BYTES); + let points: Vec<_> = (0..count as u32).map(|i| (i, i)).collect(); + let labels: Vec<_> = (0..count as u32).map(|i| labl(b"labl", i, &text)).collect(); + let metadata = super::read(&wav(&[fmt(44100), cue(&points), adtl(&labels)])); + assert!(metadata.limited); + assert_eq!(metadata.cue.iter().map(|m| m.name.len()).sum::(), MAX_LABEL_TOTAL_BYTES); + let oversized = "x".repeat(MAX_LABEL_BYTES + 1); + let metadata = super::read(&wav(&[fmt(44100), cue(&[(1, 3)]), adtl(&[labl(b"labl", 1, &oversized)])])); + assert!(metadata.limited); + assert_eq!(metadata.cue[0].start, 3.0); + assert!(metadata.cue[0].name.is_empty()); + } + + #[test] + fn empty_chunk_inventory_is_bounded() { + let empty = chunk(b"junk", &[]); + let many = empty.repeat(MAX_CHUNKS + 1); + let metadata = super::read(&wav(&[many, fmt(44100), cue(&[(1, 1)])])); + assert!(metadata.limited); + assert!(metadata.cue.is_empty()); + } + + #[test] + fn entity_lookahead_keeps_previous_semicolon_boundary() { + assert_eq!(unescape("A"), "A"); + assert_eq!(unescape("A"), "A"); + } + #[test] + fn xmp_quotes_spacing_and_attribute_boundaries() { + let packet = br#""#; + let markers = xmp_markers(packet, 48000.0); + assert_eq!(markers.len(), 1); + assert_eq!((markers[0].start, markers[0].duration, markers[0].rate), (24000.0, 12000.0, 48000.0)); + assert_eq!(markers[0].name, "single & quoted"); + let spoof = r#" label=" xmpDM:startTime='7'" otherxmpDM:startTime='8' xmpDM:startTime = '9'/>"#; + assert_eq!(xmp_value(spoof, "xmpDM:startTime").as_deref(), Some("9")); + assert!(xmp_value(" xmpDM:startTime='unterminated", "xmpDM:startTime").is_none()); + } + + #[test] + fn xmp_frame_counts_preserve_explicit_and_mixed_rates() { + let packet = br#""#; + let markers = xmp_markers(packet, 44100.0); + assert_eq!(markers.len(), 3); + assert_eq!((markers[0].start, markers[0].rate), (15.0, 24.0)); + assert!((markers[0].duration / markers[0].rate - 0.5005).abs() < 1e-12); + assert_eq!((markers[1].start, markers[1].duration, markers[1].rate), (48000.0, 24000.0, 48000.0)); + assert_eq!((markers[2].start, markers[2].duration, markers[2].rate), (15.0, 12.0, 24.0)); + for invalid in ["15f0", "15f24s0", "15fNaN", "15finf", "NaN", "1e999", "15f24junk"] { + assert!(xmp_count(invalid, 48000.0).is_none(), "{invalid}"); + } + } +} diff --git a/crates/audio/src/wav_range.rs b/crates/audio/src/wav_range.rs new file mode 100644 index 000000000..33762caf2 --- /dev/null +++ b/crates/audio/src/wav_range.rs @@ -0,0 +1,223 @@ +//! Range-backed PCM WAVE reader and canonical 16-bit writer. +//! +//! Reads chunk headers and only the requested sample range; it never loads a whole file. + +use crate::{AudioError, Result}; +use craft_media::ByteSource; + +/// Whether a prefix identifies a RIFF or RF64 WAVE file. +pub fn sniff(b: &[u8]) -> bool { + b.len() >= 12 && (&b[0..4] == b"RIFF" || &b[0..4] == b"RF64") && &b[8..12] == b"WAVE" +} + +const MAX_CHUNKS: usize = 1 << 16; +const IXML_MAX: u64 = 1 << 20; +/// Maximum sample data returned by a single range read. +pub const MAX_RANGE_BYTES: u64 = 4 << 20; + +fn le16(b: &[u8], o: usize) -> u16 { + b.get(o..o.saturating_add(2)).and_then(|s| s.try_into().ok()).map_or(0, u16::from_le_bytes) +} +fn le32(b: &[u8], o: usize) -> u32 { + b.get(o..o.saturating_add(4)).and_then(|s| s.try_into().ok()).map_or(0, u32::from_le_bytes) +} +fn read_range(source: &impl ByteSource, off: u64, len: usize) -> Result> { + let mut bytes = vec![0; len]; + source.read_at(off, &mut bytes)?; + Ok(bytes) +} + +/// A PCM WAVE file held by an arbitrary random-access source. +pub struct WavRange { + source: S, + /// Number of complete sample frames. + pub frames: u64, + /// Channel count. + pub channels: u16, + /// Sample rate in Hz. + pub rate: u32, + /// Container bits per sample. + pub bits: u16, + /// Whether samples are floating point. + pub float: bool, + /// BWF time reference in samples, from bext or iXML. + pub time_reference: Option, + data_off: u64, +} + +impl WavRange { + /// Open with caller-selected bounds on channel count and sample rate. + pub fn open(source: S, max_channels: u32, max_rate: u32) -> Result { + let len = source.len(); + let mut pos: u64 = 12; + let mut fmt = None; + let mut data = None; + let mut bext_tr = None; + let mut ixml_tr = None; + let mut ds64_data = None; + for _ in 0..MAX_CHUNKS { + let Some(body) = pos.checked_add(8).filter(|&b| b <= len) else { break }; + let mut hdr = [0u8; 8]; + source.read_at(pos, &mut hdr)?; + let id = &hdr[0..4]; + let mut clen = u64::from(le32(&hdr, 4)); + let rest = len - body; + if id == b"data" { + if clen == 0xFFFF_FFFF { + clen = ds64_data.unwrap_or(rest); + } + clen = clen.min(rest); + } + if clen > rest { + break; + } + match id { + b"fmt " if clen >= 16 => { + let b = read_range(&source, body, clen.min(40) as usize)?; + let mut tag = le16(&b, 0); + if tag == 0xFFFE && clen >= 40 { + tag = le16(&b, 24); + } + fmt = Some((tag, le16(&b, 2), le32(&b, 4), le16(&b, 14))); + } + b"ds64" if clen >= 16 => { + let b = read_range(&source, body + 8, 8)?; + ds64_data = Some(u64::from(le32(&b, 0)) | u64::from(le32(&b, 4)) << 32); + } + b"data" => data = Some((body, clen)), + b"bext" if clen >= 346 => { + let b = read_range(&source, body + 338, 8)?; + bext_tr = Some(u64::from(le32(&b, 0)) | u64::from(le32(&b, 4)) << 32); + } + b"iXML" => { + let b = read_range(&source, body, clen.min(IXML_MAX) as usize)?; + ixml_tr = ixml_time_reference(&b); + } + _ => {} + } + pos = body.saturating_add(clen).saturating_add(clen & 1); + } + let (tag, channels, rate, bits) = fmt.ok_or_else(|| AudioError::Malformed("missing fmt chunk".into()))?; + let (data_off, dlen) = data.ok_or_else(|| AudioError::Malformed("missing data chunk".into()))?; + let float = tag == 3; + if !(tag == 1 || float) + || channels == 0 + || rate == 0 + || u32::from(channels) > max_channels + || rate > max_rate + || !matches!(bits, 8 | 16 | 24 | 32 | 64) + { + return Err(AudioError::Unsupported(format!("WAV format tag {tag}, {bits} bits"))); + } + let frames = dlen / (u64::from(channels) * u64::from(bits / 8)); + Ok(Self { source, frames, channels, rate, bits, float, time_reference: bext_tr.or(ixml_tr), data_off }) + } + /// Bytes in one whole frame. + pub fn frame_bytes(&self) -> usize { + usize::from(self.channels) * usize::from(self.bits / 8) + } + /// Read the inclusive range of frames, capped to prevent unbounded allocation. + pub fn read_frames(&self, first: u64, last: u64) -> Result> { + if first > last || last >= self.frames { + return Err(AudioError::Malformed("WAV frame range outside data".into())); + } + let n = (last - first + 1).saturating_mul(self.frame_bytes() as u64); + if n > MAX_RANGE_BYTES { + return Err(AudioError::TooLarge("WAV frame range".into())); + } + let off = self.data_off.saturating_add(first.saturating_mul(self.frame_bytes() as u64)); + read_range(&self.source, off, n as usize) + } + /// Decode a channel from a buffer returned by `read_frames`. + pub fn sample(&self, block: &[u8], frame: usize, ch: usize) -> f32 { + if ch >= usize::from(self.channels) { + return 0.0; + } + let bps = usize::from(self.bits / 8); + let Some(o) = frame.checked_mul(usize::from(self.channels)).and_then(|v| v.checked_add(ch)).and_then(|v| v.checked_mul(bps)) else { + return 0.0; + }; + let Some(b) = block.get(o..o.saturating_add(bps)) else { return 0.0 }; + match (self.bits, self.float, b) { + (8, _, &[a]) => (a as f32 - 128.0) / 128.0, + (16, _, &[a, b]) => i16::from_le_bytes([a, b]) as f32 / 32768.0, + (24, _, &[a, b, c]) => (i32::from_le_bytes([0, a, b, c]) >> 8) as f32 / 8_388_608.0, + (32, false, &[a, b, c, d]) => i32::from_le_bytes([a, b, c, d]) as f32 / 2_147_483_648.0, + (32, true, &[a, b, c, d]) => f32::from_le_bytes([a, b, c, d]), + (64, true, b) => f64::from_le_bytes(b.try_into().unwrap_or([0; 8])) as f32, + _ => 0.0, + } + } +} + +fn ixml_time_reference(x: &[u8]) -> Option { + let text = String::from_utf8_lossy(x); + let field = |tag: &str| -> Option { + let open = format!("<{tag}>"); + let a = text.find(&open)? + open.len(); + let b = a + text[a..].find('<')?; + text[a..b].trim().parse().ok() + }; + let low = field("BWF_TIME_REFERENCE_LOW")?; + Some(low | field("BWF_TIME_REFERENCE_HIGH").unwrap_or(0) << 32) +} + +/// Write interleaved samples as 16-bit PCM WAVE, using symmetric 32767 scaling. +#[cfg(feature = "pcm-encode")] +pub fn write_wav16(samples: &[f32], channels: u16, rate: u32) -> Vec { + let data_len = samples.len().saturating_mul(2); + let mut v = crate::pcm_wav_header(u32::try_from(data_len).unwrap_or(u32::MAX), channels, rate, 16); + v.reserve(data_len); + for s in samples { + v.extend_from_slice(&((s.clamp(-1.0, 1.0) * 32767.0).round() as i16).to_le_bytes()); + } + v +} + +#[cfg(all(test, feature = "pcm-encode"))] +mod tests { + use super::*; + use std::sync::{Arc, Mutex}; + + struct Source { + bytes: Vec, + reads: Arc>>, + } + impl ByteSource for Source { + fn len(&self) -> u64 { + self.bytes.len() as u64 + } + fn read_at(&self, offset: u64, buf: &mut [u8]) -> std::io::Result<()> { + self.reads.lock().unwrap_or_else(std::sync::PoisonError::into_inner).push((offset, buf.len())); + self.bytes.read_at(offset, buf) + } + } + #[test] + fn bounded_reads_and_pcm_values() { + let bytes = write_wav16(&[-1.0, -0.5, 0.0, 0.5, 1.0], 1, 48_000); + let reads = Arc::new(Mutex::new(Vec::new())); + let wav = WavRange::open(Source { bytes, reads: reads.clone() }, 64, 384_000).unwrap(); + assert_eq!((wav.frames, wav.channels, wav.rate, wav.bits), (5, 1, 48_000, 16)); + assert!(reads.lock().unwrap().iter().all(|(_, n)| *n <= 40)); + let data = wav.read_frames(1, 3).unwrap(); + assert_eq!(wav.sample(&data, 0, 0), -0.5); + assert_eq!(wav.sample(&data, 2, 0), 0.5); + assert!(wav.read_frames(0, 5).is_err()); + assert!(wav.read_frames(3, 1).is_err()); + assert_eq!(reads.lock().unwrap().last(), Some(&(46, 6))); + } + #[test] + fn ixml_fallback_accepts_flat_fields() { + let base = write_wav16(&[0.0], 1, 48_000); + let mut bytes = base[..36].to_vec(); + let xml = b"51"; + bytes.extend_from_slice(b"iXML"); + bytes.extend_from_slice(&(xml.len() as u32).to_le_bytes()); + bytes.extend_from_slice(xml); + if xml.len() & 1 != 0 { + bytes.push(0); + } + bytes.extend_from_slice(&base[36..]); + assert_eq!(WavRange::open(bytes.as_slice(), 64, 384_000).unwrap().time_reference, Some((1 << 32) + 5)); + } +} diff --git a/crates/audio/src/write/coded.rs b/crates/audio/src/write/coded.rs new file mode 100644 index 000000000..0d0dd626f --- /dev/null +++ b/crates/audio/src/write/coded.rs @@ -0,0 +1,506 @@ +//! The coded formats of [`FileEncoder`](crate::FileEncoder): AAC-LC and ALAC in M4A (through +//! craft-media's MP4 muxer) and MP3, AC-3 and E-AC-3 as raw streams. [`Writer`] maps the caller's +//! WAV-order channels onto the encoder's order, drives the packet encoder, and for M4A records the +//! encoder's priming and the exact length in the edit list. + +// Each codec is its own feature, and so are the two sinks: with only some of them built, the +// other halves of the shared code are unused. +#![cfg_attr( + not(all(feature = "aac-encode", feature = "alac-encode", feature = "mp3-encode", feature = "ac3-encode", feature = "eac3-encode")), + allow(dead_code, unused_imports, unused_variables) +)] + +use super::failed_write_error; +use super::pcm::validate_layout; +use crate::layout::channel_map; +use crate::{AudioBuffer, AudioError, ChannelLayout, EncodeFormat, EncodeOptions, RateControl, Result}; +use std::io::{Cursor, Seek, Write}; + +/// Frames handed to the encoder at a time, so its buffers and the packets waiting to be written +/// stay small however long a block is. Also the frames of an ALAC packet. +const CHUNK: usize = 4096; + +/// Bits per second per channel when no rate is asked for: AAC. +const AAC_DEFAULT_BPS_PER_CHANNEL: u32 = 64_000; +/// Kilobits per second per channel when no rate is asked for: MP3, and AC-3 and E-AC-3 (per +/// full-bandwidth channel). +const DEFAULT_KBPS_PER_CHANNEL: u32 = 96; + +/// The speakers of `mask` as labels: `"L R C LFE Ls Rs"`. +fn labels(mask: u32) -> String { + ChannelLayout::from_mask(mask).speakers().iter().map(|s| s.label()).collect::>().join(" ") +} + +/// [`channel_map`] for a format with one layout per channel count ([`EncodeFormat::layout_masks`]), +/// or why the audio does not fit. +fn fixed_map(format: EncodeFormat, name: &str, order: Option<&[usize]>, layout: &ChannelLayout) -> Result> { + let Some(&mask) = format.layout_masks(layout.channels()).first() else { + return Err(AudioError::Unsupported(format!("{name} cannot hold {} channels", layout.channels()))); + }; + channel_map(mask, order, layout).ok_or_else(|| { + AudioError::Unsupported(format!("{name} holds {} channels as {}; the audio is {}", layout.channels(), labels(mask), given(layout))) + }) +} + +/// The speakers of the caller's layout, for an error message. +fn given(layout: &ChannelLayout) -> String { + layout.mask().map_or_else(|| "without speaker assignment".into(), labels) +} + +/// The encoder's channel order over AAC's layout for the count: [`crate::aac::aac_order`], and for 8 channels +/// configuration 7's C, Lc Rc, L R, Ls Rs, LFE over the WAV order L R C LFE Ls Rs Lc Rc. +#[cfg(feature = "aac-encode")] +fn aac_order(channels: usize) -> Option<&'static [usize]> { + if channels == 8 { Some(&[2, 6, 7, 0, 1, 4, 5, 3]) } else { crate::aac::aac_order(channels) } +} + +/// The lowest AAC bit rate the writer uses, per channel: 6 kbit/s, and a quarter of the sample +/// rate (256 bits a frame). The encoder accepts less, and from about half of this down codes +/// silence; a file of silence is no answer to a low bit rate (or to kbit/s given as bit/s). +#[cfg(feature = "aac-encode")] +fn aac_floor_per_channel(sample_rate: u32) -> u32 { + (sample_rate / 4).max(6000) +} + +/// The AAC rate control for `rate`: a bit rate is kept within what the stream can carry, a +/// quality of 0–1 spans the encoder's VBR range. +#[cfg(feature = "aac-encode")] +fn aac_mode(rate: Option, sample_rate: u32, channels: usize) -> crate::aac::BitrateMode { + use crate::aac::BitrateMode; + let (lo, hi) = crate::aac::cbr_range(sample_rate, channels); + let lo = lo.max(aac_floor_per_channel(sample_rate).saturating_mul(channels as u32)).min(hi.max(lo)); + // An empty range (a rate or channel count the encoder refuses anyway) leaves the rate alone. + let fit = |bps: u32| if lo <= hi { bps.clamp(lo, hi) } else { bps }; + match rate { + None => BitrateMode::Cbr(fit(AAC_DEFAULT_BPS_PER_CHANNEL.saturating_mul(channels as u32))), + Some(RateControl::Bitrate(bps)) => BitrateMode::Cbr(fit(bps)), + // NaN stays NaN, which the encoder refuses. + Some(RateControl::Quality(q)) => BitrateMode::Vbr(1.0 + 4.0 * q.clamp(0.0, 1.0)), + } +} + +/// The bit rate in kbit/s for a codec with constant bit rates only. +fn constant_kbps(name: &str, rate: Option, default: u32) -> Result { + match rate { + None => Ok(default), + Some(RateControl::Bitrate(bps)) => Ok(bps / 1000), + Some(RateControl::Quality(_)) => { + Err(AudioError::Unsupported(format!("{name} is encoded at a constant bit rate; set a bit rate, not a quality"))) + } + } +} + +/// The AC-3 or E-AC-3 settings for audio with `layout`, and the channel map onto them: the coding +/// mode whose speakers the layout names, at the nearest bit rate the stream can carry. +#[cfg(any(feature = "ac3-encode", feature = "eac3-encode"))] +fn ac3_config( + codec: crate::ac3::encoder::Codec, + sample_rate: u32, + layout: &ChannelLayout, + rate: Option, +) -> Result<(crate::ac3::encoder::Config, Vec)> { + use crate::ac3::encoder::{Codec, CodingMode, Config}; + use CodingMode::*; + let name = if codec == Codec::Ac3 { "AC-3" } else { "E-AC-3" }; + let modes = [Mono, Stereo, ThreeFront, StereoSurround, ThreeFrontSurround, StereoTwoSurround, ThreeFrontTwoSurround, Surround71]; + let found = modes.iter().flat_map(|&mode| [false, true].map(|lfe| (mode, lfe))).find_map(|(mode, lfe)| { + // 7.1 is an E-AC-3 programme with an LFE; the encoder has no other. + if mode == Surround71 && (codec != Codec::Eac3 || !lfe) { + return None; + } + let config = Config { codec, sample_rate, bitrate_kbps: 0, mode, lfe }; + let map = channel_map(config.channel_layout().mask()?, None, layout)?; + Some((config, map)) + }); + let Some((mut config, map)) = found else { + return Err(AudioError::Unsupported(format!("{name} has no coding mode for {} channels that are {}", layout.channels(), given(layout)))); + }; + let full = config.mode.channels() as u32; + let want = constant_kbps(name, rate, DEFAULT_KBPS_PER_CHANNEL * full)?; + // What `Config::validate` asks of the rate: enough for the channels, and for AC-3 an entry of + // the frame size table, for E-AC-3 a syncframe its 11-bit word count can hold. + let least = if config.mode == Surround71 { 384 } else { 32 * full }; + config.bitrate_kbps = match codec { + Codec::Ac3 => { + let floor = crate::ac3::BITRATES.iter().copied().find(|&r| r >= least).unwrap_or(least); + crate::ac3::BITRATES.iter().rev().copied().find(|&r| r <= want).filter(|&r| r >= floor).unwrap_or(floor) + } + Codec::Eac3 => { + let most = if config.mode == Surround71 { 1024 } else { (2048 * u64::from(sample_rate) / 96_000).min(1024) as u32 }; + want.clamp(least, most.max(least)) + } + }; + Ok((config, map)) +} + +/// One encoded packet and the frames it decodes to. +struct Packet { + data: Vec, + frames: u32, +} + +/// A packet encoder of any of the codecs, taking planar `f32` in its own channel order. +enum Codec { + #[cfg(feature = "aac-encode")] + Aac(Box), + #[cfg(feature = "alac-encode")] + Alac { + enc: Box, + quantizers: Vec, + /// Quantized samples of the packet being filled, per channel. + pending: Vec>, + }, + #[cfg(feature = "mp3-encode")] + Mp3(Box), + #[cfg(any(feature = "ac3-encode", feature = "eac3-encode"))] + Ac3 { + enc: Box, + /// The block with every sample finite and within ±1, as the encoder requires. + planes: Vec>, + }, +} + +impl Codec { + /// Encodes `planes` (one slice per encoder channel, equally long, at most [`CHUNK`] frames) + /// and appends the packets completed. + fn push(&mut self, planes: &[&[f32]], out: &mut Vec) -> Result<()> { + match self { + #[cfg(feature = "aac-encode")] + Codec::Aac(enc) => out.extend(enc.encode(planes).into_iter().map(|data| Packet { data, frames: crate::aac::FRAME_LEN as u32 })), + #[cfg(feature = "alac-encode")] + Codec::Alac { enc, quantizers, pending } => { + let frames = planes.first().map_or(0, |p| p.len()); + let mut at = 0; + while at < frames { + let filled = pending.first().map_or(0, Vec::len); + let end = frames.min(at + CHUNK.saturating_sub(filled).max(1)); + for ((dst, q), src) in pending.iter_mut().zip(quantizers.iter_mut()).zip(planes) { + dst.extend(src.get(at..end).unwrap_or_default().iter().map(|&s| q.quantize(s))); + } + at = end; + if pending.first().is_some_and(|p| p.len() >= CHUNK) { + alac_packet(enc, pending, out)?; + } + } + } + #[cfg(feature = "mp3-encode")] + Codec::Mp3(enc) => { + let data = enc.encode(planes); + if !data.is_empty() { + out.push(Packet { data, frames: 0 }); + } + } + #[cfg(any(feature = "ac3-encode", feature = "eac3-encode"))] + Codec::Ac3 { enc, planes: clean } => { + for (dst, src) in clean.iter_mut().zip(planes) { + dst.clear(); + dst.extend(src.iter().map(|&s| if s.is_nan() { 0.0 } else { s.clamp(-1.0, 1.0) })); + } + let refs: Vec<&[f32]> = clean.iter().map(Vec::as_slice).collect(); + out.extend(enc.push(&refs)?.into_iter().map(|p| Packet { data: p.data, frames: p.samples })); + } + } + Ok(()) + } + + /// Encodes what the encoder still holds and appends the last packets. + fn finish(&mut self, out: &mut Vec) -> Result<()> { + match self { + #[cfg(feature = "aac-encode")] + Codec::Aac(enc) => out.extend(enc.flush().into_iter().map(|data| Packet { data, frames: crate::aac::FRAME_LEN as u32 })), + #[cfg(feature = "alac-encode")] + Codec::Alac { enc, pending, .. } => { + if pending.first().is_some_and(|p| !p.is_empty()) { + alac_packet(enc, pending, out)?; + } + } + #[cfg(feature = "mp3-encode")] + Codec::Mp3(enc) => { + let data = enc.flush(); + if !data.is_empty() { + out.push(Packet { data, frames: 0 }); + } + } + #[cfg(any(feature = "ac3-encode", feature = "eac3-encode"))] + Codec::Ac3 { enc, .. } => out.extend(enc.finish()?.packets.into_iter().map(|p| Packet { data: p.data, frames: p.samples })), + } + Ok(()) + } + + /// Decoded frames before the first frame of the audio. + fn priming(&self) -> u32 { + match self { + #[cfg(feature = "aac-encode")] + Codec::Aac(enc) => enc.priming_samples(), + #[allow(unreachable_patterns)] + _ => 0, + } + } + + /// The MP4 sample entry of the stream as encoded so far (`bytes` in packets of `frames` + /// frames in all): the configuration a decoder needs, with the average bit rate. + #[cfg(any(feature = "aac-encode", feature = "alac-encode"))] + fn sample_entry(&self, sample_rate: u32, channels: u32, bytes: u64, frames: u64) -> Option { + use craft_media::isobmff::{CodecConfig, FourCc, SampleEntry}; + match self { + #[cfg(feature = "aac-encode")] + Codec::Aac(enc) => { + let mut entry = SampleEntry::aac(enc.audio_specific_config(), channels, sample_rate); + if let CodecConfig::Aac(config) = &mut entry.codec + && frames > 0 + { + let bps = u128::from(bytes) * 8 * u128::from(sample_rate) / u128::from(frames); + config.avg_bitrate = u32::try_from(bps).unwrap_or(u32::MAX); + } + Some(entry) + } + #[cfg(feature = "alac-encode")] + Codec::Alac { enc, .. } => { + let bits = u16::from(enc.encoder_config().bit_depth); + Some(SampleEntry::audio(FourCc(*b"alac"), CodecConfig::Alac { cookie: enc.cookie() }, channels, f64::from(sample_rate), bits)) + } + #[allow(unreachable_patterns)] + _ => None, + } + } +} + +/// Encodes the pending samples as one ALAC packet and empties them. +#[cfg(feature = "alac-encode")] +fn alac_packet(enc: &mut crate::alac::Encoder, pending: &mut [Vec], out: &mut Vec) -> Result<()> { + let refs: Vec<&[i32]> = pending.iter().map(Vec::as_slice).collect(); + let frames = refs.first().map_or(0, |p| p.len()) as u32; + let data = enc.encode(&refs)?; + pending.iter_mut().for_each(Vec::clear); + out.push(Packet { data, frames }); + Ok(()) +} + +/// A muxer's error as this crate's: a failed write stays an I/O error. +#[cfg(any(feature = "aac-encode", feature = "alac-encode"))] +fn mux_error(e: craft_media::isobmff::Error) -> AudioError { + match e { + craft_media::isobmff::Error::Io(e) => AudioError::Io(e), + e => AudioError::Encode(format!("M4A: {e}")), + } +} + +/// An M4A file being written: one audio track whose samples are the packets. +/// +/// The last packet is held back until the length is known: its duration is cut to the audio's +/// last frame, so the sample durations add up to the priming plus the audio, and the edit that +/// skips the priming ends where the audio does. The movie timescale is the sample rate, which +/// keeps that edit exact. +#[cfg(any(feature = "aac-encode", feature = "alac-encode"))] +struct M4a { + mux: craft_media::isobmff::Mp4Writer, + /// The track, added with the first packet: a file with no audio gets a track without an edit. + track: Option, + entry: craft_media::isobmff::SampleEntry, + sample_rate: u32, + channels: u32, + priming: u32, + held: Option, + /// Frames and bytes of the packets written, the held one not counted. + frames: u64, + bytes: u64, +} + +#[cfg(any(feature = "aac-encode", feature = "alac-encode"))] +impl M4a { + fn new(out: W, codec: &Codec, sample_rate: u32, channels: u32) -> Result { + use craft_media::isobmff::{Brand, Mp4Writer, WriterOptions}; + let entry = codec.sample_entry(sample_rate, channels, 0, 0).ok_or_else(|| AudioError::Unsupported("M4A holds AAC and ALAC".into()))?; + let mut options = WriterOptions::new(Brand::Mp4); + options.movie_timescale = sample_rate; + let mux = Mp4Writer::new(out, options).map_err(mux_error)?; + Ok(M4a { mux, track: None, entry, sample_rate, channels, priming: codec.priming(), held: None, frames: 0, bytes: 0 }) + } + + fn track(&mut self, has_audio: bool) -> Result { + if let Some(track) = self.track { + return Ok(track); + } + let mut config = craft_media::isobmff::TrackConfig::new(self.entry.clone(), self.sample_rate); + if has_audio && self.priming > 0 { + config.media_start = Some(i64::from(self.priming)); + // AAC's overlap: a packet decodes right only after the one before it. Apple's + // decoders take the priming from the edit only when the track says so. + config.roll_distance = Some(-1); + } + let track = self.mux.add_track(config).map_err(mux_error)?; + self.track = Some(track); + Ok(track) + } + + fn sample(&mut self, packet: &Packet, duration: u32) -> Result<()> { + let track = self.track(true)?; + let sample = craft_media::isobmff::WriteSample { data: &packet.data, duration, composition_offset: 0, is_sync: true }; + self.mux.write_sample(track, sample).map_err(mux_error)?; + self.frames += u64::from(duration); + self.bytes += packet.data.len() as u64; + Ok(()) + } + + fn write(&mut self, packet: Packet) -> Result<()> { + if let Some(held) = self.held.replace(packet) { + self.sample(&held, held.frames)?; + } + Ok(()) + } + + /// Writes the last packet, cut to the end of `audio_frames` frames of audio, and the `moov`. + fn finish(mut self, codec: &Codec, audio_frames: u64) -> Result { + if let Some(held) = self.held.take() { + let rest = (u64::from(self.priming) + audio_frames).saturating_sub(self.frames); + self.sample(&held, rest.min(u64::from(held.frames)) as u32)?; + } + let track = self.track(false)?; + // The configuration is complete only now: the bit rate, and ALAC's largest packet. + if let Some(entry) = codec.sample_entry(self.sample_rate, self.channels, self.bytes, self.frames) { + self.mux.set_sample_entry(track, entry).map_err(mux_error)?; + } + self.mux.finish().map_err(mux_error) + } +} + +/// Where the packets go. +enum Sink { + #[cfg(any(feature = "aac-encode", feature = "alac-encode"))] + M4a(Box>), + /// The packets in a row are the file. + Raw(W), +} + +/// A file of one of the coded formats written block by block. +pub(crate) struct Writer { + codec: Codec, + sink: Sink, + /// For each encoder channel, the caller's channel that feeds it. + map: Vec, + /// The constant bit rate the encoder was set to, bits per second; `None` for ALAC and for + /// AAC at a constant quality. + bitrate: Option, + frames: u64, + packets: Vec, + /// Set when a write has failed: the encoder and the sink no longer agree, so the writer + /// refuses every later call. + failed: bool, +} + +impl Writer { + pub(crate) fn new(out: W, opts: &EncodeOptions, sample_rate: u32, channels: usize) -> Result { + validate_layout(channels, sample_rate)?; + let layout = opts.layout.unwrap_or_else(|| ChannelLayout::default_for(channels as u16)); + let (codec, map, bitrate) = match opts.format { + #[cfg(feature = "aac-encode")] + EncodeFormat::Aac => { + let map = fixed_map(EncodeFormat::Aac, "AAC", aac_order(channels), &layout)?; + let mode = aac_mode(opts.rate_control, sample_rate, channels); + let config = crate::aac::EncoderConfig { mode, ..crate::aac::EncoderConfig::cbr(sample_rate, channels, 0) }; + let bitrate = match mode { + crate::aac::BitrateMode::Cbr(bps) => Some(bps), + crate::aac::BitrateMode::Vbr(_) => None, + }; + (Codec::Aac(Box::new(crate::aac::Encoder::new(config)?)), map, bitrate) + } + #[cfg(feature = "alac-encode")] + EncodeFormat::Alac => { + let count = u8::try_from(channels).unwrap_or(u8::MAX); + let map = fixed_map(EncodeFormat::Alac, "ALAC", crate::alac::alac_order(count), &layout)?; + let bits = match opts.bit_depth { + crate::BitDepth::Int16 => 16, + crate::BitDepth::Int24 => 24, + crate::BitDepth::Int32 => 32, + crate::BitDepth::Float32 => return Err(AudioError::Unsupported("ALAC cannot store 32-bit float samples".into())), + }; + let enc = crate::alac::Encoder::new(crate::alac::EncoderConfig::new(sample_rate, count, bits).with_frame_length(CHUNK as u32))?; + let quantizers = super::pcm::Quantizer::per_channel(u32::from(bits), opts.dither, channels); + (Codec::Alac { enc: Box::new(enc), quantizers, pending: vec![Vec::new(); channels] }, map, None) + } + #[cfg(feature = "mp3-encode")] + EncodeFormat::Mp3 => { + let map = fixed_map(EncodeFormat::Mp3, "MP3", None, &layout)?; + let want = constant_kbps("MP3", opts.rate_control, DEFAULT_KBPS_PER_CHANNEL * channels as u32)?; + // A sample rate MP3 has not is the encoder's to refuse. + let kbps = crate::mp3::nearest_bitrate(sample_rate, want).unwrap_or(want); + let enc = crate::mp3::Encoder::new(crate::mp3::Config::cbr(sample_rate, channels, kbps))?; + (Codec::Mp3(Box::new(enc)), map, Some(kbps.saturating_mul(1000))) + } + #[cfg(any(feature = "ac3-encode", feature = "eac3-encode"))] + EncodeFormat::Ac3 | EncodeFormat::Eac3 => { + let codec = if opts.format == EncodeFormat::Ac3 { crate::ac3::encoder::Codec::Ac3 } else { crate::ac3::encoder::Codec::Eac3 }; + let (config, map) = ac3_config(codec, sample_rate, &layout, opts.rate_control)?; + let enc = crate::ac3::encoder::Encoder::new(config)?; + (Codec::Ac3 { enc: Box::new(enc), planes: vec![Vec::new(); channels] }, map, Some(config.bitrate_kbps.saturating_mul(1000))) + } + #[allow(unreachable_patterns)] + f => return Err(AudioError::Unsupported(format!("writing {f:?}"))), + }; + let sink = match opts.format { + #[cfg(any(feature = "aac-encode", feature = "alac-encode"))] + EncodeFormat::Aac | EncodeFormat::Alac => Sink::M4a(Box::new(M4a::new(out, &codec, sample_rate, channels as u32)?)), + _ => Sink::Raw(out), + }; + Ok(Writer { codec, sink, map, bitrate, frames: 0, packets: Vec::new(), failed: false }) + } + + pub(crate) fn bitrate(&self) -> Option { + self.bitrate + } + + /// Hands the packets waiting to the sink. + fn drain(&mut self) -> Result<()> { + for packet in self.packets.drain(..) { + match &mut self.sink { + #[cfg(any(feature = "aac-encode", feature = "alac-encode"))] + Sink::M4a(m4a) => m4a.write(packet)?, + Sink::Raw(out) => out.write_all(&packet.data)?, + } + } + Ok(()) + } + + /// Appends `frames` frames of `block`, which holds every channel at least that long. + pub(crate) fn write>(&mut self, block: &[C], frames: usize) -> Result<()> { + if self.failed { + return Err(failed_write_error()); + } + for at in (0..frames).step_by(CHUNK) { + let end = frames.min(at + CHUNK); + let planes: Vec<&[f32]> = self.map.iter().map(|&c| block.get(c).and_then(|p| p.as_ref().get(at..end)).unwrap_or_default()).collect(); + let written = self.codec.push(&planes, &mut self.packets).and_then(|()| self.drain()); + if let Err(e) = written { + self.failed = true; + return Err(e); + } + } + self.frames = self.frames.saturating_add(frames as u64); + Ok(()) + } + + pub(crate) fn finish(mut self) -> Result { + if self.failed { + return Err(failed_write_error()); + } + self.codec.finish(&mut self.packets)?; + self.drain()?; + match self.sink { + #[cfg(any(feature = "aac-encode", feature = "alac-encode"))] + Sink::M4a(m4a) => m4a.finish(&self.codec, self.frames), + Sink::Raw(mut out) => { + out.flush()?; + Ok(out) + } + } + } +} + +pub(crate) fn encode(buf: &AudioBuffer, opts: &EncodeOptions) -> Result> { + let mut w = Writer::new(Cursor::new(Vec::new()), opts, buf.sample_rate, buf.channels.len())?; + w.write(&buf.channels, buf.frames())?; + Ok(w.finish()?.into_inner()) +} + +#[cfg(test)] +mod tests; diff --git a/crates/audio/src/write/coded/tests.rs b/crates/audio/src/write/coded/tests.rs new file mode 100644 index 000000000..cbdbd3375 --- /dev/null +++ b/crates/audio/src/write/coded/tests.rs @@ -0,0 +1,167 @@ +use super::*; + +use crate::Speaker; + +fn map(mask: u32, order: Option<&[usize]>, layout: u32) -> Option> { + channel_map(mask, order, &ChannelLayout::from_mask(layout)) +} + +/// Channels are matched by speaker, not by position: a layout that lists the same speakers in +/// another WAV order is reordered, and one with other speakers is refused. +#[test] +fn channels_are_mapped_by_speaker() { + // L R C LFE Ls Rs onto itself, and onto an encoder order C L R Ls Rs LFE. + assert_eq!(map(0x60F, None, 0x60F), Some(vec![0, 1, 2, 3, 4, 5])); + assert_eq!(map(0x60F, Some(&[2, 0, 1, 4, 5, 3]), 0x60F), Some(vec![2, 0, 1, 4, 5, 3])); + // Back surrounds feed a format that has one surround pair, and the other way round. + assert_eq!(map(0x60F, None, 0x3F), Some(vec![0, 1, 2, 3, 4, 5])); + assert_eq!(map(0x3F, None, 0x60F), Some(vec![0, 1, 2, 3, 4, 5])); + // 6.1 with the back centre after the pair (L R C LFE Lb Rb Cb) onto L R C LFE Cb Ls Rs. + assert_eq!(map(0x70F, None, 0x13F), Some(vec![0, 1, 2, 3, 6, 4, 5])); + // Quad has no centre; 7.1 with sides and backs has no front pair left and right of centre. + assert_eq!(map(0x107, None, 0x33), None); + assert_eq!(map(0xFF, None, 0x63F), None); + // With both pairs in the format, one pair cannot feed both. + assert_eq!(map(0x63F, None, 0xFF), None); + assert_eq!(channel_map(0x3, None, &ChannelLayout::unordered(2)), None); + assert_eq!(map(0x3, None, 0x7), None); +} + +#[cfg(feature = "aac-encode")] +#[test] +fn every_aac_channel_count_has_a_layout_and_an_order() { + for channels in 1..=8 { + let mask = EncodeFormat::Aac.layout_masks(channels as u16)[0]; + assert_eq!(mask.count_ones() as usize, channels); + let mut order = map(mask, aac_order(channels), mask).unwrap(); + order.sort_unstable(); + assert_eq!(order, (0..channels).collect::>()); + } + assert!(EncodeFormat::Aac.layout_masks(9).is_empty() && EncodeFormat::Aac.layout_masks(0).is_empty()); + // The layouts this crate's decoder reports for what the encoder writes (`track::aac_layout`), + // and configuration 7's speakers for 8 channels. + let masks: Vec = (1..=8).map(|channels| EncodeFormat::Aac.layout_masks(channels)[0]).collect(); + assert_eq!(masks, [0x4, 0x3, 0x7, 0x107, 0x607, 0x60F, 0x70F, 0xFF]); + assert_eq!( + (EncodeFormat::Mp3.layout_masks(1), EncodeFormat::Mp3.layout_masks(2), EncodeFormat::Mp3.layout_masks(3)), + (&[0x4][..], &[0x3][..], &[][..]) + ); + // Configuration 7: C, Lc Rc, L R, Ls Rs, LFE. + let wide: Vec = ChannelLayout::from_mask(0xFF).speakers(); + let named: Vec = aac_order(8).unwrap().iter().map(|&c| wide[c]).collect(); + use Speaker::*; + assert_eq!(named, [FrontCenter, FrontLeftOfCenter, FrontRightOfCenter, FrontLeft, FrontRight, BackLeft, BackRight, Lfe]); +} + +/// The range the writer keeps a bit rate in is the one the encoder accepts. +#[cfg(feature = "aac-encode")] +#[test] +fn the_aac_bit_rate_range_is_the_encoders() { + use crate::aac::{Encoder, EncoderConfig, SAMPLE_RATES}; + for &rate in &SAMPLE_RATES { + for channels in 1..=8 { + let (lo, hi) = crate::aac::cbr_range(rate, channels); + assert!(lo <= hi, "{rate} Hz, {channels} channels"); + let accepts = |bps| Encoder::new(EncoderConfig::cbr(rate, channels, bps)).is_ok(); + assert!(accepts(lo) && accepts(hi), "{rate} Hz, {channels} channels: {lo}..={hi}"); + assert!(!accepts(lo - 1) && !accepts(hi + 1), "{rate} Hz, {channels} channels: {lo}..={hi}"); + for want in [None, Some(RateControl::Bitrate(0)), Some(RateControl::Bitrate(u32::MAX))] { + let mode = aac_mode(want, rate, channels); + assert!( + Encoder::new(EncoderConfig { mode, ..EncoderConfig::cbr(rate, channels, 0) }).is_ok(), + "{rate} Hz, {channels} channels, {want:?}" + ); + } + } + } +} + +/// Every layout the AC-3 encoders have is found from its speakers, at a bit rate the encoder +/// accepts whatever rate was asked for. +#[cfg(all(feature = "ac3-encode", feature = "eac3-encode"))] +#[test] +fn ac3_settings_follow_the_layout_and_validate() { + use crate::ac3::encoder::{Codec, CodingMode}; + let layouts = [ + (0x4, CodingMode::Mono, false), + (0x3, CodingMode::Stereo, false), + (0xB, CodingMode::Stereo, true), + (0x7, CodingMode::ThreeFront, false), + (0x103, CodingMode::StereoSurround, false), + (0x107, CodingMode::ThreeFrontSurround, false), + (0x33, CodingMode::StereoTwoSurround, false), + (0x603, CodingMode::StereoTwoSurround, false), + (0x37, CodingMode::ThreeFrontTwoSurround, false), + (0x3F, CodingMode::ThreeFrontTwoSurround, true), + (0x60F, CodingMode::ThreeFrontTwoSurround, true), + ]; + for codec in [Codec::Ac3, Codec::Eac3] { + for sample_rate in [32_000, 44_100, 48_000] { + for (mask, mode, lfe) in layouts { + for want in [None, Some(RateControl::Bitrate(0)), Some(RateControl::Bitrate(200_000)), Some(RateControl::Bitrate(u32::MAX))] { + let layout = ChannelLayout::from_mask(mask); + let (config, map) = ac3_config(codec, sample_rate, &layout, want).unwrap(); + assert_eq!((config.mode, config.lfe), (mode, lfe), "{mask:#x}"); + assert_eq!(map, (0..usize::from(layout.channels())).collect::>()); + assert_eq!(config.validate(), Ok(()), "{codec:?} {sample_rate} {mask:#x} {want:?}: {} kbit/s", config.bitrate_kbps); + } + } + let seven_one = ac3_config(codec, sample_rate, &ChannelLayout::SURROUND_7_1, Some(RateControl::Bitrate(u32::MAX))); + match codec { + Codec::Ac3 => assert!(matches!(seven_one, Err(AudioError::Unsupported(_)))), + Codec::Eac3 => assert_eq!(seven_one.unwrap().0.validate(), Ok(())), + } + } + } + // The highest table rate not above the one asked for; the default is 96 kbit/s per channel. + let rate = |codec, mask, want| ac3_config(codec, 48_000, &ChannelLayout::from_mask(mask), want).unwrap().0.bitrate_kbps; + assert_eq!(rate(Codec::Ac3, 0x3, Some(RateControl::Bitrate(200_000))), 192); + assert_eq!(rate(Codec::Ac3, 0x3, None), 192); + assert_eq!(rate(Codec::Ac3, 0x60F, None), 448); + assert_eq!(rate(Codec::Ac3, 0x60F, Some(RateControl::Bitrate(1))), 160); + assert_eq!(rate(Codec::Eac3, 0x3, Some(RateControl::Bitrate(200_500))), 200); + assert!(matches!(ac3_config(Codec::Ac3, 48_000, &ChannelLayout::STEREO, Some(RateControl::Quality(0.5))), Err(AudioError::Unsupported(_)))); + assert!(matches!(ac3_config(Codec::Ac3, 48_000, &ChannelLayout::unordered(2), None), Err(AudioError::Unsupported(_)))); +} + +/// The layouts `EncodeFormat` lists are the encoders' own: ALAC's layout per channel count, and +/// the speakers of every AC-3 and E-AC-3 coding mode the encoder has. +#[cfg(all(feature = "alac-encode", feature = "ac3-encode", feature = "eac3-encode"))] +#[test] +fn the_listed_layouts_are_the_encoders() { + use crate::ac3::encoder::{Codec, CodingMode::*, Config}; + for channels in 0..=10u16 { + let alac = crate::alac::channel_mask(channels as u8); + assert_eq!(EncodeFormat::Alac.layout_masks(channels), if alac == 0 { &[][..] } else { std::slice::from_ref(&alac) }, "ALAC, {channels}"); + for (format, codec) in [(EncodeFormat::Ac3, Codec::Ac3), (EncodeFormat::Eac3, Codec::Eac3)] { + let modes = [Mono, Stereo, ThreeFront, StereoSurround, ThreeFrontSurround, StereoTwoSurround, ThreeFrontTwoSurround, Surround71]; + let mut encoder: Vec = modes + .iter() + .flat_map(|&mode| [false, true].map(|lfe| Config { codec, sample_rate: 48_000, bitrate_kbps: 640, mode, lfe })) + .filter(|config| config.validate().is_ok() && config.channels() == usize::from(channels)) + .filter_map(|config| config.channel_layout().mask()) + .collect(); + let mut listed = format.layout_masks(channels).to_vec(); + encoder.sort_unstable(); + listed.sort_unstable(); + assert_eq!(listed, encoder, "{format:?}, {channels} channels"); + for &mask in &listed { + let layout = ChannelLayout::from_mask(mask); + assert_eq!(ac3_config(codec, 48_000, &layout, None).unwrap().0.channel_layout(), layout); + } + } + } +} + +/// The sample rates `EncodeFormat` lists are the encoders'. +#[cfg(all(feature = "aac-encode", feature = "mp3-encode"))] +#[test] +fn the_listed_sample_rates_are_the_encoders() { + let sorted = |rates: &[u32]| { + let mut rates = rates.to_vec(); + rates.sort_unstable(); + rates + }; + assert_eq!(EncodeFormat::Aac.sample_rates().unwrap(), sorted(&crate::aac::SAMPLE_RATES)); + assert_eq!(EncodeFormat::Mp3.sample_rates().unwrap(), sorted(&crate::mp3::SAMPLE_RATES)); +} diff --git a/crates/audio/src/write/mod.rs b/crates/audio/src/write/mod.rs index 2355a7b05..a2043d279 100644 --- a/crates/audio/src/write/mod.rs +++ b/crates/audio/src/write/mod.rs @@ -1,11 +1,18 @@ -//! Encoding: WAV/BWF/RF64 and AIFF/AIFF-C writers (`pcm-encode`) and the FLAC encoder -//! (`flac-encode`), whole-buffer and streaming. +//! Encoding: WAV/BWF/RF64 and AIFF/AIFF-C writers (`pcm-encode`), the FLAC encoder +//! (`flac-encode`) and the coded formats (AAC and ALAC in M4A, MP3, AC-3 and E-AC-3 streams, each +//! with its codec's feature), whole-buffer and streaming. #[cfg(feature = "pcm-encode")] mod aiff; +#[cfg(any(feature = "aac-encode", feature = "alac-encode", feature = "mp3-encode", feature = "ac3-encode", feature = "eac3-encode"))] +mod coded; #[cfg(feature = "flac-encode")] pub(crate) mod flac; mod pcm; +#[cfg(feature = "vorbis-encode")] +mod vorbis; +#[cfg(feature = "vorbis-encode")] +pub use vorbis::{OGG_BITRATES, OggVorbisOptions, encode as encode_ogg, encode_ogg_vorbis, quality_for_bitrate}; #[cfg(feature = "pcm-encode")] mod raw; #[cfg(feature = "pcm-encode")] @@ -14,11 +21,25 @@ mod wav; use crate::{AudioBuffer, AudioError, ChannelLayout, EncodeFormat, EncodeOptions, Result}; #[cfg(feature = "pcm-encode")] pub use raw::{pcm_aiff_file, pcm_aiff_header, pcm_wav_file, pcm_wav_header}; +#[cfg(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode" +))] use std::io::{Seek, Write}; -/// Encode `buf` as WAV/BWF, AIFF/AIFF-C or FLAC. Integer depths clamp samples to [-1, 1]; float +/// Encode `buf` as the file `opts.format` names. Integer depths clamp samples to [-1, 1]; float /// keeps values above full scale. `opts.layout`, when set, must have `buf`'s channel count. /// +/// The lossy formats take `opts.rate_control`. AAC and ALAC are written as M4A, which records the +/// encoder's priming and the exact length, so the file decodes to `buf`'s frames. MP3, AC-3 and +/// E-AC-3 are raw streams, which have no place for that: they decode with the codec's delay in +/// front and padded to whole frames. +/// /// ``` /// # #[cfg(feature = "pcm-encode")] { /// use craft_audio::{AudioBuffer, BitDepth, EncodeFormat, EncodeOptions, encode}; @@ -32,6 +53,7 @@ use std::io::{Seek, Write}; pub fn encode(buf: &AudioBuffer, opts: &EncodeOptions) -> Result> { check_built(opts.format)?; // Checked for every format; WAV and FLAC record it. + #[cfg_attr(not(any(feature = "pcm-encode", feature = "flac-encode")), allow(unused_variables))] let mask = opts.layout.map(|l| wav_mask(&l, buf.channels.len())).transpose()?; match opts.format { #[cfg(feature = "pcm-encode")] @@ -40,6 +62,27 @@ pub fn encode(buf: &AudioBuffer, opts: &EncodeOptions) -> Result> { EncodeFormat::Aiff => aiff::encode(buf, opts.bit_depth, opts.dither), #[cfg(feature = "flac-encode")] EncodeFormat::Flac => flac::encode(buf, opts.bit_depth, opts.dither, mask), + #[cfg(any(feature = "aac-encode", feature = "alac-encode", feature = "mp3-encode", feature = "ac3-encode", feature = "eac3-encode"))] + EncodeFormat::Aac | EncodeFormat::Alac | EncodeFormat::Mp3 | EncodeFormat::Ac3 | EncodeFormat::Eac3 => coded::encode(buf, opts), + #[cfg(feature = "vorbis-encode")] + EncodeFormat::Ogg => { + let count = u16::try_from(buf.channels.len()).unwrap_or(u16::MAX); + let layout = opts.layout.unwrap_or_else(|| ChannelLayout::default_for(count)); + if layout.channels() != count { + return Err(AudioError::Unsupported(format!("a {}-channel layout for {count} channels", layout.channels()))); + } + let map = vorbis::vorbis_map(&layout)?; + let quality = match opts.rate_control { + None => vorbis::quality_for_bitrate(192, buf.channels.len()), + Some(crate::RateControl::Bitrate(bps)) => { + vorbis::quality_for_bitrate((bps / 1000).clamp(OGG_BITRATES.0, OGG_BITRATES.1), buf.channels.len()) + } + Some(crate::RateControl::Quality(q)) if q.is_nan() => return Err(AudioError::Encode("the Vorbis quality is NaN".into())), + Some(crate::RateControl::Quality(q)) => -1.0 + 11.0 * q.clamp(0.0, 1.0), + }; + let planar: Vec<&[f32]> = map.iter().filter_map(|&i| buf.channels.get(i).map(Vec::as_slice)).collect(); + vorbis::encode_planar(buf.sample_rate, &planar, &OggVorbisOptions::new(quality)) + } #[allow(unreachable_patterns)] f => Err(AudioError::Unsupported(format!("writing {f:?}"))), } @@ -47,31 +90,49 @@ pub fn encode(buf: &AudioBuffer, opts: &EncodeOptions) -> Result> { /// Refuses a format whose writer is not built, before anything else about the request is checked. fn check_built(format: EncodeFormat) -> Result<()> { - let (built, feature) = - if format == EncodeFormat::Flac { (cfg!(feature = "flac-encode"), "flac-encode") } else { (cfg!(feature = "pcm-encode"), "pcm-encode") }; - if built { Ok(()) } else { Err(AudioError::Unsupported(format!("writing {format:?} needs the `{feature}` feature"))) } + if format.is_built() { Ok(()) } else { Err(AudioError::Unsupported(format!("writing {format:?} needs the `{}` feature", format.feature()))) } } -/// The file extensions of the formats [`encode`] can write (those whose features are on). -#[cfg(all(feature = "pcm-encode", feature = "flac-encode"))] -pub const ENCODE_EXTENSIONS: &[&str] = &["wav", "wave", "bwf", "aif", "aiff", "aifc", "flac"]; -/// The file extensions of the formats [`encode`] can write (those whose features are on). -#[cfg(all(feature = "pcm-encode", not(feature = "flac-encode")))] -pub const ENCODE_EXTENSIONS: &[&str] = &["wav", "wave", "bwf", "aif", "aiff", "aifc"]; -/// The file extensions of the formats [`encode`] can write (those whose features are on). -#[cfg(all(feature = "flac-encode", not(feature = "pcm-encode")))] -pub const ENCODE_EXTENSIONS: &[&str] = &["flac"]; +/// Every extension [`ENCODE_EXTENSIONS`] can list, and how many of them the features built give. +const fn extensions() -> ([&'static str; 13], usize) { + let mut all = [""; 13]; + let mut n = 0; + macro_rules! add { + ($built:expr, $($extension:literal),+) => { + if $built { + $( + all[n] = $extension; + n += 1; + )+ + } + }; + } + add!(cfg!(feature = "pcm-encode"), "wav", "wave", "bwf", "aif", "aiff", "aifc"); + add!(cfg!(feature = "flac-encode"), "flac"); + add!(cfg!(any(feature = "aac-encode", feature = "alac-encode")), "m4a"); + add!(cfg!(feature = "mp3-encode"), "mp3"); + add!(cfg!(feature = "ac3-encode"), "ac3"); + add!(cfg!(feature = "eac3-encode"), "eac3", "ec3"); + add!(cfg!(feature = "vorbis-encode"), "ogg"); + (all, n) +} -/// The format [`encode`] writes for a file extension or format name (`wav`, `aiff`, `flac`, -/// case-insensitive, with or without the dot), or `None` when it cannot write it. `aifc` is -/// AIFF: float AIFF is written as AIFF-C. +const EXTENSIONS: ([&str; 13], usize) = extensions(); + +/// The file extensions of the formats [`encode`] can write (those whose features are on). +pub const ENCODE_EXTENSIONS: &[&str] = EXTENSIONS.0.split_at(EXTENSIONS.1).0; + +/// The format [`encode`] writes for a file extension or format name (`wav`, `aiff`, `flac`, `m4a`, +/// `mp3`, `ac3`, `eac3`, case-insensitive, with or without the dot), or `None` when it cannot +/// write it. `aifc` is AIFF: float AIFF is written as AIFF-C. `m4a` is AAC when the AAC encoder +/// is built and ALAC otherwise; `alac` always names ALAC. /// /// ``` /// # #[cfg(feature = "pcm-encode")] { /// use craft_audio::{EncodeFormat, encode_format_for}; /// /// assert_eq!(encode_format_for(".AIFF"), Some(EncodeFormat::Aiff)); -/// assert_eq!(encode_format_for("mp3"), None); +/// assert_eq!(encode_format_for("unknown"), None); /// # } /// ``` pub fn encode_format_for(name: &str) -> Option { @@ -82,6 +143,20 @@ pub fn encode_format_for(name: &str) -> Option { "aif" | "aiff" | "aifc" => Some(EncodeFormat::Aiff), #[cfg(feature = "flac-encode")] "flac" => Some(EncodeFormat::Flac), + #[cfg(feature = "aac-encode")] + "m4a" => Some(EncodeFormat::Aac), + #[cfg(all(feature = "alac-encode", not(feature = "aac-encode")))] + "m4a" => Some(EncodeFormat::Alac), + #[cfg(feature = "alac-encode")] + "alac" => Some(EncodeFormat::Alac), + #[cfg(feature = "mp3-encode")] + "mp3" => Some(EncodeFormat::Mp3), + #[cfg(feature = "ac3-encode")] + "ac3" => Some(EncodeFormat::Ac3), + #[cfg(feature = "eac3-encode")] + "eac3" | "ec3" => Some(EncodeFormat::Eac3), + #[cfg(feature = "vorbis-encode")] + "ogg" => Some(EncodeFormat::Ogg), _ => None, } } @@ -98,15 +173,24 @@ fn wav_mask(layout: &ChannelLayout, channels: usize) -> Result { /// The error a stream encoder returns for every call after one of its writes has failed: the /// sink may already hold a prefix of the failed block, so the encoder's counters can no longer /// describe the bytes on the sink, and later writes or `finish` must not claim success. +#[cfg(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode" +))] fn failed_write_error() -> AudioError { AudioError::Encode("the stream encoder already failed on an earlier write; its output is incomplete".into()) } -/// Writes WAV/BWF (RF64 above 4 GiB), AIFF / AIFF-C `fl32` or FLAC block by block, producing -/// the same bytes [`encode`] would for the whole buffer: a file of any length never -/// has to fit in memory. The file starts at the sink's position, and nothing may follow that -/// position (a new file, or one truncated there): the encoder cannot truncate a `Write + Seek`, -/// so the bytes of a longer file already there would stay after the new one. +/// Writes a file of any [`EncodeFormat`] block by block, producing the same bytes [`encode`] +/// would for the whole buffer: a file of any length never has to fit in memory. The file starts +/// at the sink's position, and nothing may follow that position (a new file, or one truncated +/// there): the encoder cannot truncate a `Write + Seek`, so the bytes of a longer file already +/// there would stay after the new one. /// /// ``` /// # #[cfg(feature = "flac-encode")] { @@ -123,40 +207,76 @@ fn failed_write_error() -> AudioError { /// assert_eq!(&flac[..4], b"fLaC"); /// # } /// ``` +#[cfg(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode" +))] pub struct FileEncoder { writer: Writer, + format: EncodeFormat, channels: usize, } +#[cfg(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode" +))] impl std::fmt::Debug for FileEncoder { fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { - let format = match &self.writer { - #[cfg(feature = "pcm-encode")] - Writer::Wav(_) => EncodeFormat::Wav, - #[cfg(feature = "pcm-encode")] - Writer::Aiff(_) => EncodeFormat::Aiff, - #[cfg(feature = "flac-encode")] - Writer::Flac(_) => EncodeFormat::Flac, - }; - f.debug_struct("FileEncoder").field("format", &format).finish_non_exhaustive() + f.debug_struct("FileEncoder").field("format", &self.format).finish_non_exhaustive() } } -enum Writer { +// The writers differ in size with the features built; the coded one is boxed for its encoder. +#[allow(clippy::large_enum_variant)] +#[cfg(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode" +))] +enum Writer { #[cfg(feature = "pcm-encode")] Wav(wav::Writer), #[cfg(feature = "pcm-encode")] Aiff(aiff::Writer), #[cfg(feature = "flac-encode")] Flac(flac::Writer), + #[cfg(any(feature = "aac-encode", feature = "alac-encode", feature = "mp3-encode", feature = "ac3-encode", feature = "eac3-encode"))] + Coded(Box>), } +#[cfg(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode" +))] impl FileEncoder { /// Starts a file of `channels` channels at `sample_rate` at the current position of `out`. /// `opts.layout`, when set, must have `channels` channels. pub fn new(out: W, opts: &EncodeOptions, sample_rate: u32, channels: u16) -> Result { check_built(opts.format)?; - let (depth, dither, channels) = (opts.bit_depth, opts.dither, usize::from(channels)); + let channels = usize::from(channels); + #[cfg_attr(not(any(feature = "pcm-encode", feature = "flac-encode")), allow(unused_variables))] + let (depth, dither) = (opts.bit_depth, opts.dither); + #[cfg_attr(not(any(feature = "pcm-encode", feature = "flac-encode")), allow(unused_variables))] let mask = opts.layout.map(|l| wav_mask(&l, channels)).transpose()?; let writer = match opts.format { #[cfg(feature = "pcm-encode")] @@ -172,24 +292,45 @@ impl FileEncoder { EncodeFormat::Aiff => Writer::Aiff(aiff::Writer::new(out, channels, sample_rate, depth, dither)?), #[cfg(feature = "flac-encode")] EncodeFormat::Flac => Writer::Flac(flac::Writer::new(out, channels, sample_rate, depth, dither, mask)?), + #[cfg(any(feature = "aac-encode", feature = "alac-encode", feature = "mp3-encode", feature = "ac3-encode", feature = "eac3-encode"))] + EncodeFormat::Aac | EncodeFormat::Alac | EncodeFormat::Mp3 | EncodeFormat::Ac3 | EncodeFormat::Eac3 => { + Writer::Coded(Box::new(coded::Writer::new(out, opts, sample_rate, channels)?)) + } #[allow(unreachable_patterns)] f => return Err(AudioError::Unsupported(format!("writing {f:?}"))), }; - Ok(FileEncoder { writer, channels }) + Ok(FileEncoder { writer, format: opts.format, channels }) + } + + /// The constant bit rate the file is written at, in bits per second for all channels + /// together: the rate asked for or the format's default, moved to one the stream + /// can carry as [`RateControl::Bitrate`](crate::RateControl::Bitrate) describes. `None` for the lossless + /// formats and for AAC at a constant quality. + pub fn bitrate(&self) -> Option { + match &self.writer { + #[cfg(any(feature = "aac-encode", feature = "alac-encode", feature = "mp3-encode", feature = "ac3-encode", feature = "eac3-encode"))] + Writer::Coded(w) => w.bitrate(), + #[allow(unreachable_patterns)] + _ => None, + } } /// Fails, without writing anything, when `frames` more frames would not fit the format (an - /// AIFF holds 4 GiB, a FLAC 2^36 - 1 frames). + /// AIFF holds 4 GiB, a FLAC 2^36 - 1 frames). The coded formats have no limit a file reaches. pub fn check_room(&self, frames: u64) -> Result<()> { match &self.writer { #[cfg(feature = "pcm-encode")] - Writer::Wav(w) => w.check_room(frames), + Writer::Wav(w) => w.check_room(frames).map(drop), #[cfg(feature = "pcm-encode")] - Writer::Aiff(w) => w.check_room(frames), + Writer::Aiff(w) => w.check_room(frames).map(drop), #[cfg(feature = "flac-encode")] - Writer::Flac(w) => w.check_room(frames), + Writer::Flac(w) => w.check_room(frames).map(drop), + #[cfg(any(feature = "aac-encode", feature = "alac-encode", feature = "mp3-encode", feature = "ac3-encode", feature = "eac3-encode"))] + Writer::Coded(_) => { + let _ = frames; + Ok(()) + } } - .map(drop) } /// Appends the first `frames` frames of `block`, one slice per channel (`&[Vec]`, @@ -212,10 +353,13 @@ impl FileEncoder { Writer::Aiff(w) => w.write(block, frames), #[cfg(feature = "flac-encode")] Writer::Flac(w) => w.write(block, frames), + #[cfg(any(feature = "aac-encode", feature = "alac-encode", feature = "mp3-encode", feature = "ac3-encode", feature = "eac3-encode"))] + Writer::Coded(w) => w.write(block, frames), } } - /// Completes the headers and hands back `out`, positioned at the end of the file. + /// Encodes what is still buffered, completes the headers and hands back `out`, positioned at + /// the end of the file. pub fn finish(self) -> Result { match self.writer { #[cfg(feature = "pcm-encode")] @@ -224,6 +368,67 @@ impl FileEncoder { Writer::Aiff(w) => w.finish(), #[cfg(feature = "flac-encode")] Writer::Flac(w) => w.finish(), + #[cfg(any(feature = "aac-encode", feature = "alac-encode", feature = "mp3-encode", feature = "ac3-encode", feature = "eac3-encode"))] + Writer::Coded(w) => w.finish(), + } + } +} + +impl EncodeOptions { + /// Whether a file of `channels` channels at `sample_rate` can be written with these options: + /// `Ok`, or the error [`FileEncoder::new`] and [`encode`] would return, with nothing written. + /// For choosing a format before a file is created; [`EncodeFormat::sample_rates`], + /// [`EncodeFormat::max_channels`] and [`EncodeFormat::supports_layout`] say what a format takes. + /// + /// ``` + /// # #[cfg(feature = "mp3-encode")] { + /// use craft_audio::{EncodeFormat, EncodeOptions}; + /// + /// let mp3 = EncodeOptions::new(EncodeFormat::Mp3).with_bitrate(200_000); + /// assert!(mp3.check(48_000, 2).is_ok()); + /// assert!(mp3.check(96_000, 2).is_err()); // not an MP3 sample rate + /// assert_eq!(mp3.bitrate(48_000, 2).unwrap(), Some(192_000)); // the table rate below 200 kbit/s + /// # } + /// ``` + pub fn check(&self, sample_rate: u32, channels: u16) -> Result<()> { + self.bitrate(sample_rate, channels).map(drop) + } + + /// The constant bit rate such a file would be written at ([`FileEncoder::bitrate`]), or the + /// error [`EncodeOptions::check`] returns. + pub fn bitrate(&self, sample_rate: u32, channels: u16) -> Result> { + #[cfg(feature = "vorbis-encode")] + if self.format == EncodeFormat::Ogg { + // Vorbis is whole-buffer only and VBR has no constant output bitrate. + encode(&AudioBuffer::new(sample_rate, channels, 0), self)?; + return Ok(None); + } + // The encoder a file would get, on a sink that is thrown away: what it accepts and the + // rate it settles on are then the writer's by construction. + #[cfg(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode" + ))] + { + Ok(FileEncoder::new(std::io::Cursor::new(Vec::new()), self, sample_rate, channels)?.bitrate()) + } + #[cfg(not(any( + feature = "pcm-encode", + feature = "flac-encode", + feature = "aac-encode", + feature = "alac-encode", + feature = "mp3-encode", + feature = "ac3-encode", + feature = "eac3-encode" + )))] + { + let _ = (sample_rate, channels); + Err(AudioError::Unsupported("streaming encoders are not built".into())) } } } @@ -252,7 +457,7 @@ pub fn wav_float_header(sample_rate: u32, channels: u16, frames: u64) -> Result< // With only one of `pcm-encode` and `flac-encode`, the other's formats are refused, not silently // written as something else. -#[cfg(all(test, not(all(feature = "pcm-encode", feature = "flac-encode"))))] +#[cfg(all(test, any(feature = "pcm-encode", feature = "flac-encode"), not(all(feature = "pcm-encode", feature = "flac-encode"))))] mod tests { use super::*; @@ -265,6 +470,7 @@ mod tests { let enc = FileEncoder::new(std::io::Cursor::new(Vec::new()), &opts, 48_000, 1); assert!(matches!(enc, Err(AudioError::Unsupported(_)))); assert_eq!(encode_format_for(name), None); + assert!(!format.is_built() && EncodeFormat::ALL.iter().any(|f| f.is_built())); // Also with a layout it could not use. let opts = opts.with_layout(ChannelLayout::STEREO); assert!(matches!(encode(&buf, &opts), Err(AudioError::Unsupported(_)))); diff --git a/crates/audio/src/write/pcm.rs b/crates/audio/src/write/pcm.rs index 614d5a4a7..402b9bebc 100644 --- a/crates/audio/src/write/pcm.rs +++ b/crates/audio/src/write/pcm.rs @@ -1,6 +1,9 @@ //! PCM helpers shared by the WAV, AIFF and FLAC writers: quantization (with optional TPDF //! dither), interleaving and layout checks. +// The coded formats use the layout check, and of those only ALAC the quantizer. +#![cfg_attr(not(any(feature = "pcm-encode", feature = "flac-encode")), allow(dead_code))] + use crate::{AudioBuffer, AudioError, Result}; #[cfg(feature = "pcm-encode")] use {crate::BitDepth, std::io::Write}; diff --git a/crates/audio/src/write/vorbis.rs b/crates/audio/src/write/vorbis.rs new file mode 100644 index 000000000..6c19ada77 --- /dev/null +++ b/crates/audio/src/write/vorbis.rs @@ -0,0 +1,267 @@ +//! Ogg Vorbis files (Vorbis I §A, RFC 5334) from the pure-Rust [`crate::vorbis::encoder`] and +//! craft-media's Ogg page writer: 1 to 8 channels in Vorbis channel order, a quality from -1 to +//! 10 (variable bit rate), and the tags in the comment header. Builds for every target. + +use craft_media::ogg::{PageWriter, vorbis_comment}; + +use crate::vorbis::encoder::{DEFAULT_QUALITY, Encoder, EncoderConfig, Error as VorbisError}; +use crate::{AudioBuffer, AudioError, ChannelLayout, Result}; + +/// The bit rates [`encode_ogg`](crate::encode_ogg) takes, in kbit/s. Vorbis is variable: the bit rate picks a +/// quality ([`quality_for_bitrate`]), and the file comes out near it for typical music. +pub const OGG_BITRATES: (u32, u32) = (64, 500); + +/// The serial number of the stream when the options do not set one (`LCRF`): one stream per file, +/// so a constant keeps files byte-for-byte reproducible. +const SERIAL: u32 = 0x4C43_5246; + +/// Samples per channel fed to the encoder at a time. +const CHUNK: usize = 4096; + +/// What [`encode_ogg_vorbis`] writes besides the audio. +/// +/// `#[non_exhaustive]`: build it with `OggVorbisOptions::default()` or +/// [`OggVorbisOptions::new`] and the `with_*` methods. +#[non_exhaustive] +#[derive(Debug, Clone, PartialEq)] +pub struct OggVorbisOptions { + /// -1 (smallest) to 10 (best); clamped, and NaN or infinite reads as + /// [`DEFAULT_QUALITY`] (6). + pub quality: f32, + /// The comment header's vendor string. + pub vendor: String, + /// `NAME=value` comments (Vorbis I §5), in order. A name that is empty, or holds `=` or a byte + /// outside 0x20–0x7D, is left out. + pub tags: Vec<(String, String)>, + /// The logical stream's serial number. + pub serial: u32, +} + +impl Default for OggVorbisOptions { + fn default() -> Self { + OggVorbisOptions { quality: DEFAULT_QUALITY, vendor: "craft-audio".into(), tags: Vec::new(), serial: SERIAL } + } +} + +impl OggVorbisOptions { + /// The defaults at `quality`. + pub fn new(quality: f32) -> Self { + OggVorbisOptions { quality, ..Self::default() } + } + + /// Sets [`quality`](Self::quality). + pub fn with_quality(mut self, quality: f32) -> Self { + self.quality = quality; + self + } + + /// Sets [`vendor`](Self::vendor). + pub fn with_vendor(mut self, vendor: impl Into) -> Self { + self.vendor = vendor.into(); + self + } + + /// Adds a comment ([`tags`](Self::tags)). + pub fn with_tag(mut self, name: impl Into, value: impl Into) -> Self { + self.tags.push((name.into(), value.into())); + self + } + + /// Sets [`serial`](Self::serial). + pub fn with_serial(mut self, serial: u32) -> Self { + self.serial = serial; + self + } +} + +fn encode_error(e: VorbisError) -> AudioError { + AudioError::Encode(format!("Vorbis: {e}")) +} + +/// Encode `buf` as an Ogg Vorbis file with `opts`. The channels are in Vorbis channel order +/// (Vorbis I §4.3.9: L C R for three, L C R Ls Rs LFE for 5.1, L C R Ls Rs Rl Rr LFE for 7.1). +/// The last granule position is the buffer's length, so a decoder returns exactly its frames. +/// +/// ``` +/// use craft_audio::{AudioBuffer, OggVorbisOptions, encode_ogg_vorbis}; +/// +/// let buf = AudioBuffer::new(48_000, 6, 4_800); // 0.1 s of 5.1 silence +/// let opts = OggVorbisOptions::new(4.0).with_tag("TITLE", "Silence"); +/// let ogg = encode_ogg_vorbis(&buf, &opts).unwrap(); +/// assert!(ogg.starts_with(b"OggS")); +/// ``` +pub fn encode_ogg_vorbis(buf: &AudioBuffer, opts: &OggVorbisOptions) -> Result> { + let planar: Vec<&[f32]> = buf.channels.iter().map(Vec::as_slice).collect(); + encode_planar(buf.sample_rate, &planar, opts) +} + +/// [`encode_ogg_vorbis`] over channels already in Vorbis order. +pub(crate) fn encode_planar(sample_rate: u32, planar: &[&[f32]], opts: &OggVorbisOptions) -> Result> { + if !(1..=8).contains(&planar.len()) { + return Err(AudioError::Unsupported(format!("Ogg Vorbis holds 1 to 8 channels, not {}", planar.len()))); + } + let frames = planar.first().map_or(0, |c| c.len()); + if planar.iter().any(|c| c.len() != frames || c.iter().any(|s| !s.is_finite())) { + return Err(AudioError::Encode("Vorbis samples must be finite and channels equally long".into())); + } + if sample_rate == 0 { + return Err(AudioError::Encode("there is no audio to encode".into())); + } + let mut enc = Encoder::new(EncoderConfig::new(sample_rate, planar.len(), opts.quality)).map_err(encode_error)?; + let mut pages = PageWriter::new(opts.serial); + let tags: Vec<(&str, &str)> = opts.tags.iter().map(|(k, v)| (k.as_str(), v.as_str())).collect(); + let mut out = pages.header(&enc.identification()); + out.extend(pages.header(&vorbis_comment(&opts.vendor, &tags))); + out.extend(pages.header(&enc.setup())); + // The first audio packet (which outputs nothing) gets a page of its own: a stream whose first + // audio page is also its last would leave decoders to choose between start and end trimming + // (Vorbis I §A.2), and ffmpeg then plays the padding. + let mut first = true; + let mut add = |packets: Vec<(Vec, i64)>, out: &mut Vec| { + for (packet, granule) in packets { + out.extend(pages.packet(&packet, granule)); + if std::mem::take(&mut first) { + out.extend(pages.flush()); + } + } + }; + let mut at = 0; + loop { + let end = (at + CHUNK).min(frames); + let chunk: Vec<&[f32]> = planar.iter().map(|c| c.get(at..end).unwrap_or_default()).collect(); + add(enc.encode(&chunk).map_err(encode_error)?, &mut out); + at = end; + if at >= frames { + break; + } + } + add(enc.finish().map_err(encode_error)?, &mut out); + out.extend(pages.finish(None)); + Ok(out) +} + +/// Nominal kbit/s per channel at each whole quality from -1 to 10: this encoder's rate on +/// music-like test signals at 48 kHz (harmonic chords with noise), averaged. +const KBPS_PER_CHANNEL: [f32; 12] = [18.0, 23.0, 31.0, 43.0, 58.0, 75.0, 89.0, 100.0, 111.0, 121.0, 130.0, 140.0]; + +/// The quality whose nominal bit rate is `kbps` kbit/s for `channels` channels (clamped to -1..=10). +/// Vorbis is variable: the bit rate of a file depends on the audio. +pub fn quality_for_bitrate(kbps: u32, channels: usize) -> f32 { + let per = kbps as f32 / channels.max(1) as f32; + let mut q = -1.0; + for pair in KBPS_PER_CHANNEL.windows(2) { + if let [lo, hi] = *pair + && per <= hi + { + return q + ((per - lo) / (hi - lo)).clamp(0.0, 1.0); + } + q += 1.0; + } + 10.0 +} + +/// Encode `buf` as an Ogg Vorbis stream near `kbps` kbit/s ([`OGG_BITRATES`], clamped): the +/// quality [`quality_for_bitrate`] gives, no tags. 1 to 8 channels in Vorbis channel order. +/// `Err` says why there is nothing. +pub fn encode(buf: &AudioBuffer, kbps: u32) -> Result> { + let kbps = kbps.clamp(OGG_BITRATES.0, OGG_BITRATES.1); + encode_ogg_vorbis(buf, &OggVorbisOptions::new(quality_for_bitrate(kbps, buf.channels.len()))) +} + +/// WAV-order speakers ([`crate::EncodeFormat::layout_masks`]) to Vorbis order, per channel count. +const VORBIS_ORDER: [&[usize]; 8] = + [&[0], &[0, 1], &[0, 2, 1], &[0, 1, 2, 3], &[0, 2, 1, 3, 4], &[0, 2, 1, 4, 5, 3], &[0, 2, 1, 5, 6, 4, 3], &[0, 2, 1, 6, 7, 4, 5, 3]]; + +/// For each Vorbis channel, the channel of audio in `layout` it comes from. +pub(crate) fn vorbis_map(layout: &ChannelLayout) -> Result> { + let n = layout.channels(); + let mask = crate::EncodeFormat::Ogg.layout_masks(n).first().copied(); + let order = usize::from(n).checked_sub(1).and_then(|i| VORBIS_ORDER.get(i)).copied(); + match (mask, order) { + (Some(mask), Some(order)) => crate::layout::channel_map(mask, Some(order), layout) + .ok_or_else(|| AudioError::Unsupported(format!("Vorbis has no layout of {n} channels with these speakers"))), + _ => Err(AudioError::Unsupported(format!("Ogg Vorbis holds 1 to 8 channels, not {n}"))), + } +} + +#[cfg(test)] +mod tests { + #![allow(clippy::unwrap_used)] + use super::{OGG_BITRATES, OggVorbisOptions, encode, encode_ogg_vorbis, quality_for_bitrate, vorbis_map}; + use crate::{AudioBuffer, AudioError, ChannelLayout}; + + #[test] + fn rejects_unsupported_or_malformed_channels_without_dropping_audio() { + for channels in [0, 9] { + assert!(matches!(encode(&AudioBuffer::new(48_000, channels, 16), 192), Err(AudioError::Unsupported(_)))); + } + let ragged = AudioBuffer { sample_rate: 48_000, channels: vec![vec![0.0; 16], vec![0.0; 15]] }; + assert!(encode(&ragged, 192).is_err()); + assert!(encode(&AudioBuffer::new(0, 1, 16), 192).is_err()); + let invalid = AudioBuffer { sample_rate: 48_000, channels: vec![vec![f32::NAN]] }; + assert!(encode(&invalid, 192).is_err()); + let infinite = AudioBuffer { sample_rate: 48_000, channels: vec![vec![f32::INFINITY]] }; + assert!(encode(&infinite, 192).is_err()); + } + + #[test] + fn generic_and_direct_encoders_write_one_to_eight_channels() { + for channels in 1..=8 { + let audio = AudioBuffer::new(48_000, channels, 512); + let opts = crate::EncodeOptions::new(crate::EncodeFormat::Ogg); + assert!(opts.check(48_000, channels).is_ok()); + assert_eq!(opts.bitrate(48_000, channels).unwrap(), None); + let direct = encode(&audio, 192).unwrap(); + assert!(direct.starts_with(b"OggS")); + // silence: the channel order changes nothing, so the generic path writes the same file + let generic = crate::encode(&audio, &opts.with_bitrate(192_000)).unwrap(); + assert_eq!(direct, generic, "{channels} channels"); + #[cfg(feature = "decode")] + { + let (info, decoded) = crate::decode(&generic, Some("ogg")).unwrap(); + assert_eq!(info.channels, channels); + assert_eq!(decoded.frames(), audio.frames()); + } + } + } + + #[test] + fn the_quality_follows_the_bit_rate() { + assert_eq!(quality_for_bitrate(0, 2), -1.0); + assert_eq!(quality_for_bitrate(10_000, 2), 10.0); + assert!((quality_for_bitrate(200, 2) - 6.0).abs() < 1e-6); + assert!((quality_for_bitrate(600, 6) - 6.0).abs() < 1e-6); + let mut last = -2.0; + for kbps in OGG_BITRATES.0..=OGG_BITRATES.1 { + let q = quality_for_bitrate(kbps, 2); + assert!(q >= last && (-1.0..=10.0).contains(&q)); + last = q; + } + assert!(quality_for_bitrate(192, 0).is_finite()); + } + + #[test] + fn layouts_map_to_vorbis_order() { + // 5.1 in WAV order (L R C LFE Ls Rs) is written L C R Ls Rs LFE + assert_eq!(vorbis_map(&ChannelLayout::SURROUND_5_1).unwrap(), [0, 2, 1, 4, 5, 3]); + assert_eq!(vorbis_map(&ChannelLayout::SURROUND_5_1_SIDE).unwrap(), [0, 2, 1, 4, 5, 3]); + assert_eq!(vorbis_map(&ChannelLayout::from_mask(0x63F)).unwrap(), [0, 2, 1, 6, 7, 4, 5, 3]); + assert_eq!(vorbis_map(&ChannelLayout::from_mask(0x7)).unwrap(), [0, 2, 1]); + assert_eq!(vorbis_map(&ChannelLayout::STEREO).unwrap(), [0, 1]); + assert!(vorbis_map(&ChannelLayout::from_mask(0xC)).is_err()); + assert!(vorbis_map(&ChannelLayout::unordered(9)).is_err()); + assert!(vorbis_map(&ChannelLayout::unordered(0)).is_err()); + } + + #[test] + fn options_build_and_tags_reach_the_comment_header() { + let opts = OggVorbisOptions::default().with_quality(2.0).with_vendor("v").with_tag("TITLE", "t").with_tag("bad=name", "x").with_serial(9); + assert_eq!((opts.quality, opts.vendor.as_str(), opts.serial, opts.tags.len()), (2.0, "v", 9, 2)); + let ogg = encode_ogg_vorbis(&AudioBuffer::new(44_100, 1, 100), &opts).unwrap(); + let file = craft_media::ogg::open(&ogg).unwrap(); + let s = &file.streams[0]; + assert_eq!((s.serial, s.saw_eos, s.headers.len()), (9, true, 3)); + assert_eq!(s.headers[1], craft_media::ogg::vorbis_comment("v", &[("TITLE", "t")])); + assert_eq!(s.last_granule, Some(100)); + } +} diff --git a/crates/audio/tests/audio/decode_extensions.rs b/crates/audio/tests/audio/decode_extensions.rs index e8fe22d58..fc961cc39 100644 --- a/crates/audio/tests/audio/decode_extensions.rs +++ b/crates/audio/tests/audio/decode_extensions.rs @@ -62,11 +62,21 @@ fn every_decode_extension_probes_and_decodes() { fn every_encode_format_has_an_extension_it_is_written_and_read_under() { for &format in EncodeFormat::ALL { let ext = format.extension(); - assert_eq!(craft_audio::encode_format_for(ext), Some(format), "{ext}"); - assert!(craft_audio::ENCODE_EXTENSIONS.contains(&ext), "{ext}"); + // The extension names the format, or for `m4a`, which two formats share, AAC before ALAC. + // A format whose feature is off has no name. + match craft_audio::encode_format_for(ext) { + Some(named) => { + let shares_m4a = |f| matches!(f, EncodeFormat::Aac | EncodeFormat::Alac); + assert!(named == format || (shares_m4a(named) && shares_m4a(format)), "{ext}"); + assert!(named.is_built(), "{ext}"); + assert!(craft_audio::ENCODE_EXTENSIONS.contains(&ext), "{ext}"); + } + None => assert!(!matches!(format, EncodeFormat::Wav | EncodeFormat::Aiff | EncodeFormat::Flac), "{ext}"), + } assert!(DECODE_EXTENSIONS.contains(&ext), "{ext}: what is written reads back"); assert!(format.mime_type().starts_with("audio/"), "{format:?}"); } - assert_eq!(EncodeFormat::ALL.iter().map(|f| f.extension()).collect::>(), ["wav", "aif", "flac"]); + let extensions: Vec<_> = EncodeFormat::ALL.iter().map(|f| f.extension()).collect(); + assert_eq!(extensions, ["wav", "aif", "flac", "m4a", "m4a", "mp3", "ac3", "eac3", "ogg"]); assert_eq!(EncodeFormat::Aiff.mime_type(), "audio/aiff"); } diff --git a/crates/audio/tests/audio/encode_format.rs b/crates/audio/tests/audio/encode_format.rs index 30b992685..72749df04 100644 --- a/crates/audio/tests/audio/encode_format.rs +++ b/crates/audio/tests/audio/encode_format.rs @@ -14,7 +14,15 @@ fn only_formats_the_encoder_writes_are_named() { for name in craft_audio::ENCODE_EXTENSIONS { assert!(encode_format_for(name).is_some(), "{name}"); } - for other in ["mp3", "ogg", "m4a", "xyz", ""] { + for other in ["opus", "aac", "mp4", "xyz", ""] { assert_eq!(encode_format_for(other), None, "{other}"); } } + +#[test] +fn ogg_names_follow_the_feature() { + let built = cfg!(feature = "vorbis-encode"); + assert_eq!(EncodeFormat::Ogg.is_built(), built); + assert_eq!(encode_format_for("ogg"), built.then_some(EncodeFormat::Ogg)); + assert_eq!(craft_audio::ENCODE_EXTENSIONS.contains(&"ogg"), built); +} diff --git a/crates/audio/tests/audio/main.rs b/crates/audio/tests/audio/main.rs index 8c37256d6..52e1d17f4 100644 --- a/crates/audio/tests/audio/main.rs +++ b/crates/audio/tests/audio/main.rs @@ -10,3 +10,4 @@ mod raw_pcm; mod roundtrip; mod stream; mod wav_empty_data; +mod wav_ixml; diff --git a/crates/audio/tests/audio/wav_ixml.rs b/crates/audio/tests/audio/wav_ixml.rs new file mode 100644 index 000000000..88e0cfcbf --- /dev/null +++ b/crates/audio/tests/audio/wav_ixml.rs @@ -0,0 +1,87 @@ +//! iXML timestamp parity with FilmCraft's streaming WAV reader, including hostile metadata. +use std::io::Cursor; + +use craft_audio::{DecodeOptions, Decoder}; +use craft_media::ReadSeekSource; + +fn chunk(id: &[u8; 4], body: &[u8]) -> Vec { + let mut out = id.to_vec(); + out.extend_from_slice(&(body.len() as u32).to_le_bytes()); + out.extend_from_slice(body); + if !body.len().is_multiple_of(2) { + out.push(0); + } + out +} + +fn wav(metadata: &[Vec]) -> Vec { + let mut out = b"RIFF\0\0\0\0WAVE".to_vec(); + out.extend(chunk(b"fmt ", &[1, 0, 1, 0, 0x80, 0xbb, 0, 0, 0, 0x77, 1, 0, 2, 0, 16, 0])); + for m in metadata { + out.extend(m); + } + out.extend(chunk(b"data", &[0, 0, 1, 0])); + let len = (out.len() - 8) as u32; + out[4..8].copy_from_slice(&len.to_le_bytes()); + out +} + +fn xml(low: &str, high: Option<&str>) -> Vec { + let high = high.map(|v| format!("{v}")).unwrap_or_default(); + chunk(b"iXML", format!("{low}{high}").as_bytes()) +} + +fn check(file: Vec, expected: Option) { + let options = DecodeOptions::default(); + let mut bytes = Decoder::from_bytes(&file, Some("wav"), &options).unwrap(); + assert_eq!(bytes.info().bwf.as_ref().map(|b| b.time_reference), expected); + let source = ReadSeekSource::new(Cursor::new(file.clone())).unwrap(); + let mut stream = Decoder::open(source, Some("wav"), &options).unwrap(); + assert_eq!(stream.info().bwf.as_ref().map(|b| b.time_reference), expected); + assert_eq!(bytes.info().frames, 2); + assert_eq!(stream.next_block().unwrap().unwrap().frames(), 2); + assert_eq!(bytes.next_block().unwrap().unwrap().frames(), 2); +} + +#[test] +fn ixml_fallback_preserves_large_timestamps_and_optional_high_word() { + check(wav(&[xml(" 5 ", Some("1"))]), Some((1 << 32) + 5)); + check(wav(&[xml("0", None)]), Some(0)); + check(wav(&[xml("4294967295", Some("4294967295"))]), Some(u64::MAX)); +} + +#[test] +fn bext_wins_in_either_chunk_order_including_timestamp_zero() { + let bext = chunk(b"bext", &[0; 602]); + let ixml = xml("5", Some("1")); + check(wav(&[bext.clone(), ixml.clone()]), Some(0)); + check(wav(&[ixml.clone(), bext.clone()]), Some(0)); + check(wav(&[bext, chunk(b"bext", &[0; 12]), ixml.clone()]), Some(0)); + check(wav(&[chunk(b"bext", &[0; 12]), ixml]), Some((1 << 32) + 5)); +} + +#[test] +fn invalid_metadata_never_fabricates_a_timestamp_or_blocks_audio() { + for low in ["4294967296", "-1", "+1", "", "1.5", "5"] { + check(wav(&[xml(low, Some("0"))]), None); + } + for high in ["4294967296", "-1", "", "oops"] { + check(wav(&[xml("5", Some(high))]), None); + } + for text in [ + "5", + "", + "56", + "5", + ] { + check(wav(&[chunk(b"iXML", text.as_bytes())]), None); + } + check(wav(&[chunk(b"iXML", &vec![b' '; (1 << 20) + 1])]), None); + let deep = format!("{}{}", "".repeat(40), "".repeat(40)); + check(wav(&[chunk(b"iXML", deep.as_bytes())]), None); +} + +#[test] +fn only_first_ixml_document_is_considered() { + check(wav(&[xml("5", None), xml("6", None)]), Some(5)); +} diff --git a/crates/audio/tests/codecs/aac/he_aac_oracle.rs b/crates/audio/tests/codecs/aac/he_aac_oracle.rs new file mode 100644 index 000000000..0eeba69c8 --- /dev/null +++ b/crates/audio/tests/codecs/aac/he_aac_oracle.rs @@ -0,0 +1,169 @@ +//! HE-AAC (SBR) decoding against ffmpeg's decoder, on streams from ffmpeg's AudioToolbox encoder +//! (oracle only; skipped without ffmpeg or without `aac_at`, which only macOS builds have). + +use std::process::Command; + +use craft_audio::aac::{Decoder, split_adts}; + +use super::common::*; + +/// ffmpeg, if it has the AudioToolbox AAC encoder. +fn ffmpeg_with_aac_at() -> Option { + let ff = ffmpeg()?; + let out = Command::new(&ff).args(["-hide_banner", "-encoders"]).output().ok()?; + if !String::from_utf8_lossy(&out.stdout).contains("aac_at") { + eprintln!("SKIPPED: ffmpeg has no aac_at encoder"); + return None; + } + Some(ff) +} + +fn input(rate: u32, channels: usize, secs: f32) -> Vec> { + let n = (rate as f32 * secs) as usize; + (0..channels) + .map(|c| { + let a = chords(rate, n, 0.004 * c as f32); + let s = speech(rate, n, 11 + c as u64); + let z = noise(n, 0.05, 3 + c as u64); + let k = clicks(rate, n); + a.iter().zip(&s).zip(&z).zip(&k).map(|(((a, s), z), k)| 0.4 * a + 0.4 * s * if c == 1 { -0.7 } else { 1.0 } + z + 0.3 * k).collect() + }) + .collect() +} + +/// ffmpeg's decode of a stream of `channels` channels. ffmpeg decodes mono HE-AAC as stereo (it +/// allows for parametric stereo the stream may hold), with the same signal in both channels when +/// there is none: of those, the first is taken (a downmix would scale it by -3 dB). +fn ffmpeg_decode_as(ff: &std::path::Path, file: &std::path::Path, channels: usize) -> Vec> { + let mut planar = ffmpeg_decode_raw(ff, file, channels.max(2)); + planar.truncate(channels); + planar +} + +fn ffmpeg_decode_raw(ff: &std::path::Path, file: &std::path::Path, channels: usize) -> Vec> { + let out = Command::new(ff).args(["-v", "error", "-i"]).arg(file).args(["-ac", &channels.to_string(), "-f", "f32le", "-"]).output().unwrap(); + assert!(out.status.success() && out.stderr.is_empty(), "{}", String::from_utf8_lossy(&out.stderr)); + let samples: Vec = out.stdout.as_chunks::<4>().0.iter().map(|&b| f32::from_le_bytes(b)).collect(); + (0..channels).map(|c| samples.iter().skip(c).step_by(channels).copied().collect()).collect() +} + +/// Our decode of an ADTS stream, SBR on. +fn decode_sbr(data: &[u8]) -> (Vec>, u32) { + let frames = split_adts(data).unwrap(); + let mut dec = Decoder::from_adts(&frames[0].0).unwrap(); + dec.set_sbr(true); + let mut out: Vec> = Vec::new(); + for (_, au) in frames { + let pcm = dec.decode(au).unwrap(); + out.resize(pcm.len(), Vec::new()); + for (o, p) in out.iter_mut().zip(pcm) { + o.extend(p); + } + } + (out, dec.sample_rate()) +} + +/// The lag of `b` behind `a` (within ±`range`) that correlates best. +fn best_lag(a: &[f32], b: &[f32], range: i64) -> i64 { + let n = a.len().min(b.len()) as i64; + let (from, to) = (range, n - range); + (-range..=range) + .max_by(|&x, &y| { + let c = |lag: i64| (from..to).step_by(3).map(|i| a[i as usize] as f64 * b[(i + lag) as usize] as f64).sum::(); + c(x).total_cmp(&c(y)) + }) + .unwrap() +} + +/// Energy in dB of `x[from..from + 2048]` within `lo..hi` Hz (Hann-windowed DFT). +fn band_db(x: &[f32], from: usize, rate: f32, lo: f32, hi: f32) -> f64 { + let n = 2048; + let seg = &x[from..from + n]; + let (b0, b1) = ((lo / rate * n as f32) as usize, (hi / rate * n as f32) as usize); + let mut e = 0f64; + for b in b0..b1.min(n / 2) { + let (mut re, mut im) = (0f64, 0f64); + for (i, &s) in seg.iter().enumerate() { + let w = 0.5 - 0.5 * (2.0 * std::f64::consts::PI * i as f64 / n as f64).cos(); + let ph = 2.0 * std::f64::consts::PI * (b * i) as f64 / n as f64; + re += s as f64 * w * ph.cos(); + im += s as f64 * w * ph.sin(); + } + e += re * re + im * im; + } + 10.0 * (e + 1e-12).log10() +} + +struct Case { + name: &'static str, + rate: u32, + channels: usize, + bitrate: &'static str, +} + +/// HE-AAC streams of 1 to 6 channels at 22.05 to 48 kHz decode as ffmpeg decodes them: the same +/// timing, the low band to 55 dB and the rebuilt high band's level to 0.1 dB. +#[test] +fn matches_ffmpeg_he_aac_decoder() { + let Some(ff) = ffmpeg_with_aac_at() else { return }; + let cases = [ + Case { name: "he_stereo_44k_48k", rate: 44_100, channels: 2, bitrate: "48k" }, + Case { name: "he_mono_44k_32k", rate: 44_100, channels: 1, bitrate: "32k" }, + Case { name: "he_stereo_48k_64k", rate: 48_000, channels: 2, bitrate: "64k" }, + Case { name: "he_stereo_32k_32k", rate: 32_000, channels: 2, bitrate: "32k" }, + Case { name: "he_stereo_44k_24k", rate: 44_100, channels: 2, bitrate: "24k" }, + Case { name: "he_stereo_22k_24k", rate: 22_050, channels: 2, bitrate: "24k" }, + Case { name: "he_51_48k_160k", rate: 48_000, channels: 6, bitrate: "160k" }, + ]; + for case in &cases { + let path = fixtures().join(format!("{}.aac", case.name)); + ffmpeg_encode(&ff, &input(case.rate, case.channels, 3.0), case.rate, &["-c:a", "aac_at", "-profile:a", "4", "-b:a", case.bitrate], &path); + let data = std::fs::read(&path).unwrap(); + let refr = ffmpeg_decode_as(&ff, &path, case.channels); + let (ours, rate) = decode_sbr(&data); + // ffmpeg puts surround in WAV order, the decoder in AAC element order + let ours: Vec> = match (case.channels == 6).then(|| craft_audio::aac::wav_order(6)).flatten() { + Some(order) => order.iter().map(|&c| ours[c].clone()).collect(), + None => ours, + }; + assert_eq!(rate, case.rate, "{}", case.name); + assert_eq!(ours.len(), case.channels); + // the same delay as ffmpeg's decoder: 962 samples + assert_eq!(best_lag(&refr[0], &ours[0], 2100), 0, "{}", case.name); + let core_nyquist = case.rate as f32 / 4.0; + for c in 0..case.channels { + let n = refr[c].len().min(ours[c].len()) - 4096; + let a: Vec = refr[c][2048..n].to_vec(); + let b: Vec = ours[c][2048..n].to_vec(); + let full = snr(&a, &b); + let (la, lb) = (lowpass(&a, core_nyquist * 0.8, case.rate as f32), lowpass(&b, core_nyquist * 0.8, case.rate as f32)); + let low = snr(&la, &lb); + let mut diffs = Vec::new(); + for from in (4096..a.len() - 4096).step_by(4096) { + for (lo, hi) in + [(core_nyquist * 1.05, core_nyquist * 1.3), (core_nyquist * 1.3, core_nyquist * 1.7), (core_nyquist * 1.7, core_nyquist * 2.0)] + { + let (ea, eb) = (band_db(&a, from, case.rate as f32, lo, hi), band_db(&b, from, case.rate as f32, lo, hi)); + if ea > 0.0 { + diffs.push(eb - ea); + } + } + } + // the LFE (WAV channel 3 of 5.1) has no SBR, so no high band; every other channel must + // have one to compare + let lfe = case.channels == 6 && c == 3; + assert!(lfe || !diffs.is_empty(), "{} ch{c}: no high band to compare", case.name); + let mean = diffs.iter().sum::() / diffs.len() as f64; + let mad = diffs.iter().map(|d| d.abs()).sum::() / diffs.len() as f64; + println!( + "{} ch{c}: full SNR {full:.1} dB, low band SNR {low:.1} dB, high band level {mean:+.2} dB (mean |Δ| {mad:.2} dB over {})", + case.name, + diffs.len() + ); + // The low band differs by the core decoders' rounding; the high band's noise comes + // from the same table, so whole frames match too. + assert!(full > 45.0 && low > 55.0, "{} ch{c}: SNR {full:.1} / {low:.1} dB", case.name); + assert!(lfe || mad < 0.1, "{} ch{c}: high band off by {mad:.2} dB", case.name); + } + } +} diff --git a/crates/audio/tests/codecs/aac/mod.rs b/crates/audio/tests/codecs/aac/mod.rs index 45e627242..7d55a7364 100644 --- a/crates/audio/tests/codecs/aac/mod.rs +++ b/crates/audio/tests/codecs/aac/mod.rs @@ -3,4 +3,5 @@ mod common; mod decoder_oracle; mod encoder_oracle; +mod he_aac_oracle; mod robustness; diff --git a/crates/audio/tests/codecs/file_formats.rs b/crates/audio/tests/codecs/file_formats.rs new file mode 100644 index 000000000..82edf72aa --- /dev/null +++ b/crates/audio/tests/codecs/file_formats.rs @@ -0,0 +1,591 @@ +//! The coded formats of `encode` and `FileEncoder` (AAC and ALAC in M4A; MP3, AC-3 and E-AC-3 +//! streams), read back with this crate's decoder and, where ffmpeg is installed, with ffmpeg. + +use std::io::{Cursor, Seek, SeekFrom, Write}; +use std::process::Command; + +use craft_audio::{ + AudioBuffer, AudioError, BitDepth, ChannelLayout, CodecId, Container, ENCODE_EXTENSIONS, EncodeFormat, EncodeOptions, FileEncoder, RateControl, + decode, encode, encode_format_for, +}; + +const CODED: [EncodeFormat; 5] = [EncodeFormat::Aac, EncodeFormat::Alac, EncodeFormat::Mp3, EncodeFormat::Ac3, EncodeFormat::Eac3]; + +/// `channels` channels of `frames` frames, each channel a tone of its own frequency, on the +/// 16-bit grid (so a lossless 16-bit format returns it exactly). +fn tones(sample_rate: u32, channels: usize, frames: usize) -> AudioBuffer { + let channel = |c: usize| { + let step = (60.0 + 17.0 * c as f32) * std::f32::consts::TAU / sample_rate as f32; + (0..frames).map(|i| ((i as f32 * step).sin() * 0.4 * 32768.0).round() / 32768.0).collect() + }; + AudioBuffer { sample_rate, channels: (0..channels).map(channel).collect() } +} + +/// Signal-to-noise ratio in dB of `got` against `want`, over the frames both have. +fn snr(want: &[f32], got: &[f32]) -> f64 { + let (signal, noise) = + want.iter().zip(got).fold((0f64, 0f64), |(s, n), (&w, &g)| (s + f64::from(w).powi(2), n + (f64::from(w) - f64::from(g)).powi(2))); + 10.0 * (signal / noise.max(1e-30)).log10() +} + +fn is_unsupported(r: Result) -> bool { + matches!(r, Err(AudioError::Unsupported(_))) +} + +/// An AAC file decodes to the frames that were written, in place: the edit list skips the +/// encoder's priming and cuts the padding of the last packet. +#[test] +fn aac_decodes_to_the_frames_written() { + for frames in [1, 1000, 1024, 1025, 5000, 48_017] { + let buf = tones(48_000, 2, frames); + let file = encode(&buf, &EncodeOptions::new(EncodeFormat::Aac)).unwrap(); + let (info, out) = decode(&file, Some("m4a")).unwrap(); + assert_eq!((info.container, info.codec, info.sample_rate, info.damaged), (Container::Mp4, CodecId::Aac, 48_000, 0)); + assert_eq!((info.frames, out.frames(), info.layout), (frames as u64, frames, ChannelLayout::STEREO), "{frames} frames"); + if frames >= 5000 { + for (want, got) in buf.channels.iter().zip(&out.channels) { + assert!(snr(want, got) > 30.0, "{frames} frames: {} dB", snr(want, got)); + } + } + } +} + +/// The M4A track's durations add up to the priming and the audio, and its one edit skips the +/// priming and lasts as long as the audio, in a movie timescale that is the sample rate. +#[test] +fn aac_timing_is_in_the_sample_table_and_the_edit_list() { + use craft_audio::craft_media::isobmff; + let file = encode(&tones(44_100, 1, 3000), &EncodeOptions::new(EncodeFormat::Aac)).unwrap(); + let mp4 = isobmff::open(file.as_slice()).unwrap(); + let track = &mp4.tracks[0]; + assert_eq!((mp4.timescale, track.timescale), (44_100, 44_100)); + assert_eq!(track.samples.iter().map(|s| u64::from(s.duration)).sum::(), 1024 + 3000); + assert_eq!(track.edits.len(), 1); + assert_eq!((track.edits[0].media_time, track.edits[0].segment_duration), (1024, 3000)); + // Every packet is in a `roll` group of distance -1, which Apple's decoders need to apply the edit. + let groups = &track.sample_groups; + assert_eq!((groups.len(), groups[0].grouping_type), (1, isobmff::FourCc(*b"roll"))); + assert_eq!((groups[0].runs.as_slice(), &groups[0].descriptions), ([(track.samples.len() as u32, 1)].as_slice(), &vec![vec![0xFF, 0xFF]])); +} + +/// The sample entry is completed when the file is: AAC's average bit rate, and the largest packet +/// and average bit rate in ALAC's cookie. +#[test] +fn the_sample_entry_describes_the_finished_stream() { + use craft_audio::craft_media::isobmff::{self, CodecConfig}; + let buf = tones(48_000, 2, 48_000); + let file = encode(&buf, &EncodeOptions::new(EncodeFormat::Aac).with_bitrate(96_000)).unwrap(); + let mp4 = isobmff::open(file.as_slice()).unwrap(); + let CodecConfig::Aac(aac) = &mp4.tracks[0].entries[0].codec else { panic!("not AAC") }; + assert!((90_000..100_000).contains(&aac.avg_bitrate), "{} bit/s", aac.avg_bitrate); + + let file = encode(&buf, &EncodeOptions::new(EncodeFormat::Alac).with_bit_depth(BitDepth::Int16)).unwrap(); + let mp4 = isobmff::open(file.as_slice()).unwrap(); + let CodecConfig::Alac { cookie } = &mp4.tracks[0].entries[0].codec else { panic!("not ALAC") }; + let config = craft_audio::alac::SpecificConfig::parse(cookie).unwrap(); + let largest = mp4.tracks[0].samples.iter().map(|s| s.size).max().unwrap(); + assert_eq!((config.max_frame_bytes, config.bit_depth, config.frame_length), (largest, 16, 4096)); + assert!(config.avg_bit_rate > 0); +} + +/// Dither reaches ALAC's quantizer: a level below one 16-bit step is silence without it. +#[test] +fn alac_dithers_when_asked() { + let buf = AudioBuffer { sample_rate: 48_000, channels: vec![vec![0.3 / 32768.0; 4096]] }; + let opts = EncodeOptions::new(EncodeFormat::Alac).with_bit_depth(BitDepth::Int16); + let (_, plain) = decode(&encode(&buf, &opts).unwrap(), Some("m4a")).unwrap(); + let (_, dithered) = decode(&encode(&buf, &opts.with_dither(true)).unwrap(), Some("m4a")).unwrap(); + assert!(plain.channels[0].iter().all(|&s| s == 0.0)); + assert!(dithered.channels[0].iter().any(|&s| s != 0.0)); +} + +/// ALAC returns the quantized samples exactly, for every channel count, with each channel where +/// its speaker is. +#[test] +fn alac_is_lossless_in_every_layout() { + for channels in 1..=8u8 { + let layout = ChannelLayout::from_mask(craft_audio::alac::channel_mask(channels)); + let buf = tones(44_100, usize::from(channels), 10_000); + let opts = EncodeOptions::new(EncodeFormat::Alac).with_bit_depth(BitDepth::Int16).with_layout(layout); + let file = encode(&buf, &opts).unwrap(); + let (info, out) = decode(&file, Some("m4a")).unwrap(); + assert_eq!((info.container, info.codec, info.frames, info.layout), (Container::Mp4, CodecId::Alac, 10_000, layout), "{channels} channels"); + assert_eq!(out, buf, "{channels} channels"); + } + // 24 bits, and a length that is no whole number of packets. + let buf = tones(96_000, 2, 4096 * 2 + 1); + let (_, out) = decode(&encode(&buf, &EncodeOptions::new(EncodeFormat::Alac)).unwrap(), Some("m4a")).unwrap(); + assert_eq!(out, buf); +} + +/// Channels go to the encoder by speaker. ALAC's 6.1 has the back centre last and AAC's has it +/// before the surround pair; 5.1 with side surrounds is the format's one 5.1. +#[test] +fn channels_reach_their_speakers() { + // WAV 6.1 with a back pair (L R C LFE Lb Rb Cb), the usual 7-channel layout, into ALAC: the same. + let buf = tones(48_000, 7, 8000); + let (info, out) = decode(&encode(&buf, &EncodeOptions::new(EncodeFormat::Alac).with_bit_depth(BitDepth::Int16)).unwrap(), Some("m4a")).unwrap(); + assert_eq!(info.layout, ChannelLayout::from_mask(0x13F)); + assert_eq!(out, buf); + // 5.1 with side surrounds decodes from ALAC as 5.1 with back surrounds, channel for channel. + let buf = tones(48_000, 6, 8000); + let opts = EncodeOptions::new(EncodeFormat::Alac).with_bit_depth(BitDepth::Int16).with_layout(ChannelLayout::SURROUND_5_1_SIDE); + assert_eq!(decode(&encode(&buf, &opts).unwrap(), Some("m4a")).unwrap().1, buf); + + // AAC: the decoded channel each input channel comes back as. 5.1 decodes in WAV order; the + // 6.1 and 7.1 streams have no speaker assignment and decode in AAC order (C L R Ls Rs Cs LFE + // and C Lc Rc L R Ls Rs LFE). + let wide = ChannelLayout::from_mask(0xFF); + let cases: [(usize, Option, &[usize]); 3] = + [(6, None, &[0, 1, 2, 3, 4, 5]), (7, None, &[2, 0, 1, 4, 5, 6, 3]), (8, Some(wide), &[2, 6, 7, 0, 1, 4, 5, 3])]; + for (channels, layout, decoded_from) in cases { + let buf = tones(48_000, channels, 48_000); + let mut opts = EncodeOptions::new(EncodeFormat::Aac); + opts.layout = layout; + let (info, out) = decode(&encode(&buf, &opts).unwrap(), Some("m4a")).unwrap(); + assert_eq!((usize::from(info.channels), out.frames()), (channels, 48_000)); + for (decoded, &input) in decoded_from.iter().enumerate() { + let db = snr(&buf.channels[input], &out.channels[decoded]); + assert!(db > 20.0, "{channels} channels: decoded {decoded} is not input {input} ({db} dB)"); + } + } +} + +/// MP3, AC-3 and E-AC-3 are raw streams: the audio comes back behind the codec's delay. +#[test] +fn raw_streams_decode_behind_the_codec_delay() { + let buf = tones(48_000, 2, 48_000); + for (format, codec, container, delay, floor) in [ + (EncodeFormat::Mp3, CodecId::Mp3, Container::MpegAudio, craft_audio::mp3::DELAY as usize, 25.0), + (EncodeFormat::Ac3, CodecId::Ac3, Container::Ac3, 256, 15.0), + (EncodeFormat::Eac3, CodecId::Eac3, Container::Ac3, 256, 15.0), + ] { + let file = encode(&buf, &EncodeOptions::new(format)).unwrap(); + let (info, out) = decode(&file, Some(format.extension())).unwrap(); + assert_eq!((info.codec, info.container, info.sample_rate, info.layout, info.damaged), (codec, container, 48_000, ChannelLayout::STEREO, 0)); + assert_eq!((format.codec(), format.container()), (codec, container)); + assert!(out.frames() >= 48_000 + delay, "{format:?}: {} frames", out.frames()); + for (want, got) in buf.channels.iter().zip(&out.channels) { + let db = snr(want, &got[delay..]); + assert!(db > floor, "{format:?}: {db} dB"); + } + } + // 5.1 with back surrounds is AC-3's 3/2 mode with an LFE, channel for channel. + let buf = tones(48_000, 6, 48_000); + let (info, out) = decode(&encode(&buf, &EncodeOptions::new(EncodeFormat::Ac3)).unwrap(), Some("ac3")).unwrap(); + assert_eq!(info.layout, ChannelLayout::SURROUND_5_1_SIDE); + for (c, (want, got)) in buf.channels.iter().zip(&out.channels).enumerate() { + // The encoder low-pass filters the LFE, which shifts its phase: it is compared by level. + if c == 3 { + let rms = |x: &[f32]| (x.iter().map(|&s| f64::from(s).powi(2)).sum::() / x.len() as f64).sqrt(); + let level = rms(&got[4800..]) / rms(want); + assert!((0.5..1.5).contains(&level), "LFE at {level} of its level"); + } else { + assert!(snr(want, &got[256..]) > 12.0, "channel {c}: {} dB", snr(want, &got[256..])); + } + } + // 7.1 is an E-AC-3 programme. + let buf = tones(48_000, 8, 4800); + assert!(is_unsupported(encode(&buf, &EncodeOptions::new(EncodeFormat::Ac3)))); + // (Read back by ffmpeg in `ffmpeg_reads_the_files`: this crate's file reader plays the 5.1 + // independent substream of a raw 7.1 stream.) + assert!(!encode(&buf, &EncodeOptions::new(EncodeFormat::Eac3)).unwrap().is_empty()); +} + +/// Blocks of any sizes give the bytes `encode` gives. +#[test] +fn streaming_writes_the_bytes_of_encode() { + let buf = tones(44_100, 2, 30_000); + for format in CODED { + let opts = EncodeOptions::new(format); + let whole = encode(&buf, &opts).unwrap(); + let mut enc = FileEncoder::new(Cursor::new(Vec::new()), &opts, 44_100, 2).unwrap(); + assert!(format!("{enc:?}").contains(&format!("{format:?}"))); + let mut at = 0; + for n in [1, 4095, 4097, 0, 9000, 12_807] { + let block: Vec<&[f32]> = buf.channels.iter().map(|c| &c[at..]).collect(); + enc.check_room(n as u64).unwrap(); + enc.write(&block, n).unwrap(); + at += n; + } + assert_eq!(at, buf.frames()); + assert!(enc.finish().unwrap().into_inner() == whole, "{format:?}"); + } +} + +/// A file starts at the sink's position. +#[test] +fn a_file_starts_at_the_sinks_position() { + let buf = tones(48_000, 1, 6000); + for format in CODED { + let opts = EncodeOptions::new(format); + let mut sink = Cursor::new(b"junk".to_vec()); + sink.seek(SeekFrom::End(0)).unwrap(); + let mut enc = FileEncoder::new(sink, &opts, 48_000, 1).unwrap(); + enc.write(&buf.channels, 6000).unwrap(); + let mut sink = enc.finish().unwrap(); + let end = sink.stream_position().unwrap(); + let bytes = sink.into_inner(); + assert_eq!(end, bytes.len() as u64, "{format:?}"); + assert!(bytes[4..] == encode(&buf, &opts).unwrap(), "{format:?}"); + } +} + +/// A file with no audio is still a file of its format. +#[test] +fn no_audio_gives_an_empty_file() { + let buf = AudioBuffer::new(48_000, 2, 0); + for format in CODED { + let file = encode(&buf, &EncodeOptions::new(format)).unwrap(); + if format.container() == Container::Mp4 { + let mp4 = craft_audio::craft_media::isobmff::open(file.as_slice()).unwrap(); + assert_eq!((mp4.tracks.len(), mp4.tracks[0].samples.len(), mp4.tracks[0].edits.len()), (1, 0, 0), "{format:?}"); + } else { + assert!(file.is_empty(), "{format:?}"); + } + } +} + +/// The bit rate asked for is the one written, moved to the nearest the format has. +#[test] +fn the_bit_rate_is_followed() { + let buf = tones(48_000, 2, 48_000); + let size = |opts: &EncodeOptions| encode(&buf, opts).unwrap().len(); + // MP3 at 48 kHz: 24 bytes per frame and kbit/s, 43 frames for a second and the delay. + assert_eq!(size(&EncodeOptions::new(EncodeFormat::Mp3)), 43 * 192 * 3); + assert_eq!(size(&EncodeOptions::new(EncodeFormat::Mp3).with_bitrate(130_000)), 43 * 128 * 3); + assert_eq!(size(&EncodeOptions::new(EncodeFormat::Mp3).with_bitrate(1)), 43 * 32 * 3); + assert_eq!(size(&EncodeOptions::new(EncodeFormat::Mp3).with_bitrate(u32::MAX)), 43 * 320 * 3); + // AC-3: 32 packets of 1536 frames, 4 bytes per packet and kbit/s. + assert_eq!(size(&EncodeOptions::new(EncodeFormat::Ac3)), 32 * 192 * 4); + assert_eq!(size(&EncodeOptions::new(EncodeFormat::Ac3).with_bitrate(100_000)), 32 * 96 * 4); + assert_eq!(size(&EncodeOptions::new(EncodeFormat::Eac3).with_bitrate(100_000)), 32 * 100 * 4); + // AAC: a second of audio at about the rate asked for, around the 128 kbit/s default. + let (low, default, high) = ( + size(&EncodeOptions::new(EncodeFormat::Aac).with_bitrate(64_000)), + size(&EncodeOptions::new(EncodeFormat::Aac)), + size(&EncodeOptions::new(EncodeFormat::Aac).with_bitrate(256_000)), + ); + assert!((7000..10_000).contains(&low) && (15_000..18_500).contains(&default) && (30_000..35_000).contains(&high), "{low} {default} {high}"); + // A rate no stream can carry is moved into range, not refused. + encode(&buf, &EncodeOptions::new(EncodeFormat::Aac).with_bitrate(u32::MAX)).unwrap(); + // The lowest rate still codes the audio: the encoder accepts rates that leave only silence + // (128 for 128 kbit/s is one), and the writer does not go that low. + for (sample_rate, channels) in [(8000, 1), (16_000, 2), (44_100, 2), (48_000, 6), (96_000, 2)] { + let buf = tones(sample_rate, channels, sample_rate as usize); + let (_, out) = decode(&encode(&buf, &EncodeOptions::new(EncodeFormat::Aac).with_bitrate(128)).unwrap(), Some("m4a")).unwrap(); + let rms = |x: &[f32]| (x.iter().map(|&s| f64::from(s).powi(2)).sum::() / x.len() as f64).sqrt(); + let level = rms(&out.channels[0]) / rms(&buf.channels[0]); + assert!(level > 0.5, "{sample_rate} Hz, {channels} channels: {level} of the level"); + } + // Quality is AAC's; ALAC has no rate to control. + let (small, large) = + (size(&EncodeOptions::new(EncodeFormat::Aac).with_quality(0.0)), size(&EncodeOptions::new(EncodeFormat::Aac).with_quality(1.0))); + assert!(small < large, "{small} {large}"); + assert!(matches!(encode(&buf, &EncodeOptions::new(EncodeFormat::Aac).with_quality(f32::NAN)), Err(AudioError::Encode(_)))); + for format in [EncodeFormat::Mp3, EncodeFormat::Ac3, EncodeFormat::Eac3] { + assert!(is_unsupported(encode(&buf, &EncodeOptions::new(format).with_quality(0.5))), "{format:?}"); + } + let alac = EncodeOptions::new(EncodeFormat::Alac); + assert_eq!(size(&alac), size(&alac.clone().with_bitrate(64_000))); + assert_eq!(alac.rate_control, None); + assert_eq!(alac.with_quality(0.5).rate_control, Some(RateControl::Quality(0.5))); +} + +/// What a format cannot hold is refused before anything is written. +#[test] +fn what_a_format_cannot_hold_is_refused() { + let refused = |format, sample_rate, channels: u16, layout: Option| { + let mut opts = EncodeOptions::new(format); + opts.layout = layout; + let mut sink = Cursor::new(Vec::new()); + let r = FileEncoder::new(&mut sink, &opts, sample_rate, channels).map(drop); + assert!(r.is_ok() || sink.into_inner().is_empty(), "{format:?}: bytes written for a refused file"); + assert_eq!(r.is_err(), encode(&AudioBuffer::new(sample_rate, channels, 100), &opts).is_err()); + r + }; + // Quad has no centre, which AAC's and ALAC's four channels have. + assert!(is_unsupported(refused(EncodeFormat::Aac, 48_000, 4, None))); + assert!(is_unsupported(refused(EncodeFormat::Alac, 48_000, 4, None))); + refused(EncodeFormat::Aac, 48_000, 4, Some(ChannelLayout::from_mask(0x107))).unwrap(); + // WAV 7.1 (backs and sides) is not the 7.1 of AAC and ALAC (a front pair left and right of centre). + assert!(is_unsupported(refused(EncodeFormat::Aac, 48_000, 8, None))); + assert!(is_unsupported(refused(EncodeFormat::Alac, 48_000, 8, Some(ChannelLayout::SURROUND_7_1)))); + assert!(is_unsupported(refused(EncodeFormat::Aac, 48_000, 2, Some(ChannelLayout::unordered(2))))); + for format in CODED { + assert!(is_unsupported(refused(format, 48_000, 9, None)), "{format:?}"); + assert!(matches!(refused(format, 48_000, 2, Some(ChannelLayout::MONO)), Err(AudioError::Encode(_))), "{format:?}"); + assert!(matches!(refused(format, 0, 2, None), Err(AudioError::Encode(_))), "{format:?}"); + assert!(matches!(refused(format, 48_000, 0, None), Err(AudioError::Encode(_))), "{format:?}"); + } + assert!(is_unsupported(refused(EncodeFormat::Mp3, 48_000, 3, None))); + assert!(is_unsupported(refused(EncodeFormat::Ac3, 48_000, 7, None))); + // Sample rates the codec has not. + for format in [EncodeFormat::Aac, EncodeFormat::Mp3, EncodeFormat::Ac3, EncodeFormat::Eac3] { + assert!(matches!(refused(format, 44_000, 2, None), Err(AudioError::Encode(_))), "{format:?}"); + } + refused(EncodeFormat::Alac, 44_000, 2, None).unwrap(); + let float = EncodeOptions::new(EncodeFormat::Alac).with_bit_depth(BitDepth::Float32); + assert!(is_unsupported(encode(&AudioBuffer::new(48_000, 2, 100), &float))); +} + +/// Samples no encoder should see: NaN is silence, and values past full scale do not fail a +/// write (AC-3, whose encoder refuses them, gets them clamped). +#[test] +fn non_finite_and_loud_samples_are_written() { + let mut buf = tones(48_000, 2, 6000); + buf.channels[0][100] = f32::NAN; + buf.channels[0][200] = f32::INFINITY; + buf.channels[1][300] = -7.5; + for format in CODED { + let file = encode(&buf, &EncodeOptions::new(format)).unwrap(); + let (_, out) = decode(&file, Some(format.extension())).unwrap(); + assert!(out.channels.iter().flatten().all(|s| s.is_finite()), "{format:?}"); + } +} + +/// A sink that fails once `room` bytes are written. +struct Full { + inner: Cursor>, + room: usize, +} + +impl Write for Full { + fn write(&mut self, data: &[u8]) -> std::io::Result { + if self.inner.get_ref().len() + data.len() > self.room { + return Err(std::io::Error::other("disk full")); + } + self.inner.write(data) + } + fn flush(&mut self) -> std::io::Result<()> { + Ok(()) + } +} + +impl Seek for Full { + fn seek(&mut self, pos: SeekFrom) -> std::io::Result { + self.inner.seek(pos) + } +} + +/// After a failed write the encoder is unusable: later writes and `finish` fail too. +#[test] +fn a_failed_write_ends_the_encoder() { + let buf = tones(48_000, 2, 48_000); + for format in CODED { + let mut enc = FileEncoder::new(Full { inner: Cursor::new(Vec::new()), room: 2000 }, &EncodeOptions::new(format), 48_000, 2).unwrap(); + assert!(matches!(enc.write(&buf.channels, 48_000), Err(AudioError::Io(_))), "{format:?}"); + assert!(matches!(enc.write(&buf.channels, 1), Err(AudioError::Encode(_))), "{format:?}"); + assert!(matches!(enc.finish(), Err(AudioError::Encode(_))), "{format:?}"); + } +} + +#[test] +fn the_formats_are_named() { + assert_eq!(encode_format_for("m4a"), Some(EncodeFormat::Aac)); + assert_eq!(encode_format_for(".ALAC"), Some(EncodeFormat::Alac)); + assert_eq!(encode_format_for("mp3"), Some(EncodeFormat::Mp3)); + assert_eq!(encode_format_for("ac3"), Some(EncodeFormat::Ac3)); + assert_eq!(encode_format_for("eac3"), Some(EncodeFormat::Eac3)); + assert_eq!(encode_format_for("ec3"), Some(EncodeFormat::Eac3)); + for &format in EncodeFormat::ALL { + assert!(ENCODE_EXTENSIONS.contains(&format.extension()), "{format:?}"); + let named = encode_format_for(format.extension()).unwrap(); + assert_eq!((named.extension(), named.mime_type()), (format.extension(), format.mime_type())); + assert_eq!(format.is_lossy(), !matches!(format.codec(), CodecId::Pcm | CodecId::Flac | CodecId::Alac)); + } + for name in ENCODE_EXTENSIONS { + assert!(encode_format_for(name).is_some(), "{name}"); + } + assert_eq!(ENCODE_EXTENSIONS.len(), 13); + for format in CODED { + assert!(EncodeFormat::ALL.contains(&format) && format.is_built()); + } + let features: Vec<_> = EncodeFormat::ALL.iter().map(|f| f.feature()).collect(); + assert_eq!( + features, + ["pcm-encode", "pcm-encode", "flac-encode", "alac-encode", "aac-encode", "mp3-encode", "ac3-encode", "eac3-encode", "vorbis-encode"] + ); +} + +/// ffmpeg's decode of `file` as interleaved `f32`, split into `channels` channels. +fn ffmpeg_decode(ff: &std::path::Path, file: &[u8], name: &str, channels: usize) -> Vec> { + let path = crate::testkit::fixtures_dir("file_formats").join(name); + std::fs::write(&path, file).unwrap(); + let out = Command::new(ff).args(["-v", "error", "-i"]).arg(&path).args(["-f", "f32le", "-"]).output().unwrap(); + assert!(out.status.success() && out.stderr.is_empty(), "{name}: {}", String::from_utf8_lossy(&out.stderr)); + let mut planar = vec![Vec::new(); channels]; + for (i, s) in out.stdout.as_chunks::<4>().0.iter().enumerate() { + planar[i % channels].push(f32::from_le_bytes(*s)); + } + planar +} + +/// ffmpeg reads every format: the M4A files without their priming (it applies the edit list) and, +/// from ffmpeg 9, at the length written; ALAC exactly; the raw streams behind their delay. +#[test] +fn ffmpeg_reads_the_files() { + let ff = require_ffmpeg!(); + for (channels, layout) in [(1, None), (2, None), (6, None), (8, Some(ChannelLayout::from_mask(0xFF)))] { + let buf = tones(48_000, channels, 20_000); + let mut opts = EncodeOptions::new(EncodeFormat::Aac); + opts.layout = layout; + let out = ffmpeg_decode(&ff, &encode(&buf, &opts).unwrap(), &format!("aac_{channels}.m4a"), channels); + // The priming is skipped by every ffmpeg (the channels below line up). The padding of the + // last packet is cut by ffmpeg 9; ffmpeg 8 plays it, to the end of the 20th packet. + assert!([20_000, 20_480].contains(&out[0].len()), "AAC, {channels} channels: {} frames", out[0].len()); + if channels <= 6 { + // ffmpeg also returns 5.1 in WAV order. + for (c, (want, got)) in buf.channels.iter().zip(&out).enumerate() { + assert!(snr(want, got) > 20.0, "AAC, {channels} channels, channel {c}: {} dB", snr(want, got)); + } + } + let opts = opts.with_format(EncodeFormat::Alac).with_bit_depth(BitDepth::Int16); + let opts = if channels == 6 { opts.with_layout(ChannelLayout::SURROUND_5_1) } else { opts }; + let out = ffmpeg_decode(&ff, &encode(&buf, &opts).unwrap(), &format!("alac_{channels}.m4a"), channels); + assert!(out == buf.channels, "ALAC, {channels} channels"); + } + let buf = tones(44_100, 2, 20_000); + for (format, delay, floor) in + [(EncodeFormat::Mp3, craft_audio::mp3::DELAY as usize, 25.0), (EncodeFormat::Ac3, 256, 15.0), (EncodeFormat::Eac3, 256, 15.0)] + { + let name = format!("stream.{}", format.extension()); + let out = ffmpeg_decode(&ff, &encode(&buf, &EncodeOptions::new(format)).unwrap(), &name, 2); + for (want, got) in buf.channels.iter().zip(&out) { + assert!( + got.len() >= 20_000 + delay && snr(want, &got[delay..]) > floor, + "{format:?}: {} frames, {} dB", + got.len(), + snr(want, &got[delay..]) + ); + } + } + // E-AC-3 7.1: the front channels, which the 5.1 downmix leaves as they are. + let buf = tones(48_000, 8, 20_000); + let out = ffmpeg_decode(&ff, &encode(&buf, &EncodeOptions::new(EncodeFormat::Eac3)).unwrap(), "stream_7_1.eac3", 8); + assert!(out[7].len() >= 20_256); + for c in 0..3 { + assert!(snr(&buf.channels[c], &out[c][256..]) > 12.0, "E-AC-3 7.1, channel {c}: {} dB", snr(&buf.channels[c], &out[c][256..])); + } +} + +/// Apple's decoder (macOS `afinfo`; skipped elsewhere) takes the priming and the length of an AAC +/// file from the edit list, which it does only for a track with a `roll` sample group. +#[test] +fn apple_reads_the_aac_priming_and_length() { + let Some(afinfo) = crate::testkit::oracle::find_tool("afinfo") else { return }; + for (frames, remainder) in [(20_000, 480), (1, 1023), (4096, 0)] { + let path = crate::testkit::fixtures_dir("file_formats").join(format!("apple_{frames}.m4a")); + std::fs::write(&path, encode(&tones(48_000, 2, frames), &EncodeOptions::new(EncodeFormat::Aac)).unwrap()).unwrap(); + let out = Command::new(&afinfo).arg(&path).output().unwrap(); + let text = String::from_utf8_lossy(&out.stdout); + assert!(text.contains(&format!("{frames} valid frames + 1024 priming + {remainder} remainder")), "{frames} frames: {text}"); + } +} + +/// What `EncodeFormat` says a format takes is what the writer accepts: for every format, a file +/// can be begun exactly when the sample rate, the channel count and the layout are listed. +#[test] +fn the_format_queries_agree_with_the_writer() { + // Every rate a format lists, each with its neighbours, and some that none lists. + let mut rates: Vec = EncodeFormat::ALL.iter().filter_map(|f| f.sample_rates()).flatten().flat_map(|&r| [r - 1, r, r + 1]).collect(); + rates.extend([44_000, 176_400, 192_000]); + rates.sort_unstable(); + rates.dedup(); + let masks = [ + 0x4, 0x1, 0x3, 0xC, 0x600, 0x7, 0xB, 0x103, 0x33, 0x107, 0x603, 0xF, 0x10B, 0x37, 0x607, 0x10F, 0x60B, 0x3F, 0x60F, 0x13F, 0x70F, 0x63F, 0xFF, + ]; + let layouts = masks.iter().map(|&mask| ChannelLayout::from_mask(mask)).chain((1..=9).map(ChannelLayout::unordered)); + for layout in layouts { + for &format in EncodeFormat::ALL { + // Vorbis is whole-buffer-only; its direct/generic API and options are tested separately. + if format == EncodeFormat::Ogg { + continue; + } + // 24 bits is a depth every format has. + let opts = EncodeOptions::new(format).with_layout(layout); + for &sample_rate in &rates { + let listed = format.sample_rates().is_none_or(|r| r.contains(&sample_rate)) + && layout.channels() <= format.max_channels() + && format.supports_layout(&layout); + let checked = opts.check(sample_rate, layout.channels()); + assert_eq!(checked.is_ok(), listed, "{format:?}, {sample_rate} Hz, {layout:?}: {checked:?}"); + let begun = FileEncoder::new(Cursor::new(Vec::new()), &opts, sample_rate, layout.channels()); + assert_eq!(begun.is_ok(), listed, "{format:?}, {sample_rate} Hz, {layout:?}"); + } + if let Some(fixed) = format.fixed_layouts(layout.channels()) { + assert!(fixed.iter().all(|l| format.supports_layout(l) && l.channels() == layout.channels())); + assert_eq!(format.supports_layout(&layout), fixed.iter().any(|l| same_speakers(l, &layout)), "{format:?}, {layout:?}"); + } + } + } + for &format in EncodeFormat::ALL { + assert!(!format.supports_layout(&ChannelLayout::unordered(0))); + assert_eq!(format.supports_quality(), EncodeOptions::new(format).with_quality(0.5).check(48_000, 2).is_ok() && format.is_lossy()); + assert_eq!(format.fixed_layouts(2).is_none(), matches!(format, EncodeFormat::Wav | EncodeFormat::Aiff | EncodeFormat::Flac)); + } + // The most channels is the most: one more is refused, by the layouts of a format that has + // a fixed set and by the writer of one that has not. + let most: Vec = EncodeFormat::ALL.iter().map(|f| f.max_channels()).collect(); + assert_eq!(most, [32_767, 32_767, 8, 8, 8, 2, 6, 8, 8]); + for &format in EncodeFormat::ALL { + let most = format.max_channels(); + match format.fixed_layouts(most) { + Some(layouts) => assert!(!layouts.is_empty() && format.fixed_layouts(most + 1).unwrap().is_empty(), "{format:?}"), + None => { + let opts = EncodeOptions::new(format).with_bit_depth(BitDepth::Int16); + assert!(opts.check(48_000, most).is_ok() && opts.check(48_000, most + 1).is_err(), "{format:?}"); + assert!(format.supports_layout(&ChannelLayout::unordered(most)) && !format.supports_layout(&ChannelLayout::unordered(most + 1))); + } + } + } + assert_eq!(EncodeFormat::Ac3.fixed_layouts(3).unwrap().len(), 3); + assert!(EncodeFormat::Aac.fixed_layouts(9).unwrap().is_empty()); +} + +/// Whether two layouts name the same speakers, a surround pair being either sides or backs. +fn same_speakers(a: &ChannelLayout, b: &ChannelLayout) -> bool { + let one_pair = |mask: u32| if mask & 0x630 == 0x30 { mask & !0x30 | 0x600 } else { mask }; + matches!((a.mask(), b.mask()), (Some(a), Some(b)) if one_pair(a) == one_pair(b)) +} + +/// `bitrate` is the rate the file is written at, for the options and for the encoder they give. +#[test] +fn the_bit_rate_used_can_be_asked_for() { + let rate = |opts: EncodeOptions, sample_rate, channels| { + let asked = opts.bitrate(sample_rate, channels).unwrap(); + assert_eq!(FileEncoder::new(Cursor::new(Vec::new()), &opts, sample_rate, channels).unwrap().bitrate(), asked); + asked + }; + assert_eq!(rate(EncodeOptions::new(EncodeFormat::Mp3), 48_000, 2), Some(192_000)); + assert_eq!(rate(EncodeOptions::new(EncodeFormat::Mp3).with_bitrate(130_000), 48_000, 2), Some(128_000)); + assert_eq!(rate(EncodeOptions::new(EncodeFormat::Mp3).with_bitrate(u32::MAX), 22_050, 1), Some(160_000)); + assert_eq!(rate(EncodeOptions::new(EncodeFormat::Ac3).with_bitrate(191_999), 48_000, 2), Some(160_000)); + assert_eq!(rate(EncodeOptions::new(EncodeFormat::Ac3), 48_000, 6), Some(448_000)); + assert_eq!(rate(EncodeOptions::new(EncodeFormat::Eac3).with_bitrate(100_999), 48_000, 2), Some(100_000)); + assert_eq!(rate(EncodeOptions::new(EncodeFormat::Aac), 48_000, 2), Some(128_000)); + assert_eq!(rate(EncodeOptions::new(EncodeFormat::Aac).with_bitrate(128), 48_000, 2), Some(24_000)); + assert_eq!(rate(EncodeOptions::new(EncodeFormat::Aac).with_bitrate(u32::MAX), 8000, 1), Some(48_000)); + assert_eq!(rate(EncodeOptions::new(EncodeFormat::Aac).with_quality(0.5), 48_000, 2), None); + for format in [EncodeFormat::Wav, EncodeFormat::Aiff, EncodeFormat::Flac, EncodeFormat::Alac] { + assert_eq!(rate(EncodeOptions::new(format).with_bitrate(96_000), 48_000, 2), None, "{format:?}"); + } + // The size of a second of MP3 is the rate reported. + let buf = tones(48_000, 2, 48_000); + let opts = EncodeOptions::new(EncodeFormat::Mp3).with_bitrate(130_000); + assert_eq!(encode(&buf, &opts).unwrap().len() as u32, 43 * 1152 / 8 * opts.bitrate(48_000, 2).unwrap().unwrap() / 48_000); + assert!(is_unsupported(EncodeOptions::new(EncodeFormat::Mp3).bitrate(48_000, 6))); +} + +#[test] +fn vorbis_is_whole_buffer_only_and_validates_layouts() { + let opts = EncodeOptions::new(EncodeFormat::Ogg).with_layout(ChannelLayout::STEREO); + assert!(opts.check(48_000, 2).is_ok()); + assert_eq!(opts.bitrate(48_000, 2).unwrap(), None); + assert!(FileEncoder::new(Cursor::new(Vec::new()), &opts, 48_000, 2).is_err()); + let buf = craft_audio::AudioBuffer::new(48_000, 2, 512); + assert!(craft_audio::encode(&buf, &opts).unwrap().starts_with(b"OggS")); + let wrong = opts.with_layout(ChannelLayout::from_mask(0xC)); + assert!(wrong.check(48_000, 2).is_err()); + assert!(craft_audio::encode(&buf, &wrong).is_err()); +} diff --git a/crates/audio/tests/codecs/main.rs b/crates/audio/tests/codecs/main.rs index 5db6af4b0..41829be5c 100644 --- a/crates/audio/tests/codecs/main.rs +++ b/crates/audio/tests/codecs/main.rs @@ -1,5 +1,5 @@ -//! Tests of the codecs ported from FilmCraft, built as one test binary: each file is a module, as -//! in their original crates. +//! Tests of the codecs ported from FilmCraft and EffectCraft, built as one test binary: each file +//! is a module, as in their original crates. // Index loops follow the specifications, as in FilmCraft (which allows this lint workspace-wide). #![allow(clippy::needless_range_loop)] @@ -12,9 +12,13 @@ mod ac3_encoder_oracle; mod ac3_oracle; mod alac; mod concealment; +mod file_formats; mod mp3_decoder_oracle; mod mp3_encoder_oracle; mod mpa_oracle; mod opus; +mod opusenc; mod routing_oracle; mod track_decoder; +#[cfg(feature = "vorbis-encode")] +mod vorbis_encoder_oracle; diff --git a/crates/audio/tests/codecs/opusenc/common.rs b/crates/audio/tests/codecs/opusenc/common.rs new file mode 100644 index 000000000..571c9826c --- /dev/null +++ b/crates/audio/tests/codecs/opusenc/common.rs @@ -0,0 +1,286 @@ +//! Shared helpers for the SILK/hybrid round-trip tests: synthetic signals, encoding, decoding +//! with FilmCraft's spec-derived decoder, delay-searched SNR and an Ogg Opus writer for ffmpeg. +#![allow(dead_code)] + +use craft_audio::opus::encoder::OpusEncoder; + +pub const SR: f64 = 48_000.0; +pub const FRAME: usize = OpusEncoder::FRAME_SIZE; + +/// Two-pole resonator state. +struct Reson { + y1: f64, + y2: f64, +} + +impl Reson { + fn run(&mut self, x: f64, f: f64, bw: f64) -> f64 { + let r = (-std::f64::consts::PI * bw / SR).exp(); + let a1 = 2.0 * r * (2.0 * std::f64::consts::PI * f / SR).cos(); + let a2 = -r * r; + let g = 1.0 - r; + let y = g * x + a1 * self.y1 + a2 * self.y2; + self.y2 = self.y1; + self.y1 = y; + y + } +} + +/// Speech-like test signal: a glottal pulse train (100–200 Hz, gliding pitch) through three +/// formant resonators whose frequencies change per syllable, with syllable envelopes, short +/// pauses and a little aspiration noise. +pub fn speech(len: usize, seed: u32) -> Vec { + let vowels: [[f64; 3]; 5] = + [[730.0, 1090.0, 2440.0], [270.0, 2290.0, 3010.0], [530.0, 1840.0, 2480.0], [570.0, 840.0, 2410.0], [300.0, 870.0, 2240.0]]; + let syl = (0.24 * SR) as usize; + let mut out = Vec::with_capacity(len); + let mut res = [Reson { y1: 0.0, y2: 0.0 }, Reson { y1: 0.0, y2: 0.0 }, Reson { y1: 0.0, y2: 0.0 }]; + let mut phase = 0.0f64; + let mut s = seed; + let mut prev_glottal = 0.0; + for n in 0..len { + let t = n as f64 / SR; + let k = n / syl; + let pos = (n % syl) as f64 / syl as f64; + // Every fourth syllable is a pause. + let pause = k % 4 == 3; + let env = if pause { 0.0 } else { (std::f64::consts::PI * pos).sin().powf(0.7) }; + let f0 = 100.0 + 50.0 * (1.0 + (2.0 * std::f64::consts::PI * 0.9 * t).sin()) + 15.0 * ((k * 7) % 5) as f64 - 30.0 * pos; + phase += f0 / SR; + if phase >= 1.0 { + phase -= 1.0; + } + // Rosenberg-like glottal flow; its derivative excites the vocal tract. + let g = if phase < 0.4 { + 0.5 * (1.0 - (std::f64::consts::PI * phase / 0.4).cos()) + } else if phase < 0.56 { + (std::f64::consts::FRAC_PI_2 * (phase - 0.4) / 0.16).cos() + } else { + 0.0 + }; + let dg = (g - prev_glottal) * 40.0; + prev_glottal = g; + s = s.wrapping_mul(1_664_525).wrapping_add(1_013_904_223); + let noise = ((s >> 8) as f64 / (1u32 << 24) as f64 - 0.5) * 0.02; + let v = vowels[(k * 3 + 1) % 5]; + let w = vowels[(k * 3 + 4) % 5]; + let mix = pos; + let mut x = dg + noise; + let mut y = 0.0; + for (i, r) in res.iter_mut().enumerate() { + let f = v[i] * (1.0 - mix) + w[i] * mix; + y += r.run(x, f, 60.0 + 40.0 * i as f64) * [1.0, 0.6, 0.3][i]; + } + x = y * env * 1.2; + out.push(x as f32); + } + let peak = out.iter().fold(0f32, |m, v| m.max(v.abs())); + let g = 0.5 / peak.max(1e-9); + out.iter().map(|v| v * g).collect() +} + +/// Music-like test signal: three-note chords of harmonic tones (six partials each), changing +/// every half second with soft attacks and releases. +pub fn music(len: usize) -> Vec { + let chords: [[f64; 3]; 4] = [[261.63, 329.63, 392.0], [220.0, 277.18, 329.63], [174.61, 220.0, 261.63], [196.0, 246.94, 293.66]]; + let seg = (0.5 * SR) as usize; + (0..len) + .map(|n| { + let t = n as f64 / SR; + let k = n / seg; + let pos = (n % seg) as f64 / seg as f64; + let env = (pos * 20.0).min(1.0) * ((1.0 - pos) * 10.0).min(1.0); + let mut x = 0.0; + for (j, &f) in chords[k % 4].iter().enumerate() { + for h in 1..=6 { + x += (2.0 * std::f64::consts::PI * f * h as f64 * t + j as f64).sin() / h as f64; + } + } + (0.12 * env * x) as f32 + }) + .collect() +} + +pub fn interleave(planar: &[Vec]) -> Vec { + let c = planar.len(); + let len = planar[0].len(); + let mut pcm = vec![0f32; len * c]; + for (ch, p) in planar.iter().enumerate() { + for (i, &v) in p.iter().enumerate() { + pcm[i * c + ch] = v; + } + } + pcm +} + +/// Encodes planar input (padding the tail so the encoder delay is flushed). +pub fn encode(mut enc: OpusEncoder, planar: &[Vec]) -> (OpusEncoder, Vec>) { + let c = planar.len(); + let len = planar[0].len(); + let total = len + enc.pre_skip() as usize + 16; + let frames = total.div_ceil(FRAME); + let mut pcm = interleave(planar); + pcm.resize(frames * FRAME * c, 0.0); + let packets = pcm.chunks(FRAME * c).map(|f| enc.encode_float(f).unwrap()).collect(); + (enc, packets) +} + +/// Decodes with filmcraft-opus, drops the pre-skip and returns planar channels (full length). +pub fn decode(head: &[u8], packets: &[Vec]) -> Vec> { + let mut dec = craft_audio::opus::Decoder::new(head).expect("OpusHead"); + dec.set_trim_pre_skip(true); + let mut out: Vec> = vec![Vec::new(); dec.channels()]; + for p in packets { + let planes = dec.decode(Some(p)).expect("decode"); + for (o, p) in out.iter_mut().zip(planes) { + o.extend_from_slice(&p); + } + } + out +} + +/// Waveform SNR (dB) of `dec[i + shift]` against `orig[i]`, excluding the first and last 20 ms. +pub fn snr_at(orig: &[f32], dec: &[f32], shift: isize) -> f64 { + let (mut s, mut e) = (0f64, 0f64); + for i in FRAME..orig.len() - FRAME { + let j = i as isize + shift; + let d = if j >= 0 && (j as usize) < dec.len() { dec[j as usize] as f64 } else { 0.0 }; + s += (orig[i] as f64).powi(2); + e += (orig[i] as f64 - d).powi(2); + } + 10.0 * (s / e.max(1e-30)).log10() +} + +/// Best SNR over residual delays of ±`range` samples: (SNR dB, delay). +pub fn snr_search(orig: &[f32], dec: &[f32], range: isize) -> (f64, isize) { + (-range..=range).map(|d| (snr_at(orig, dec, d), d)).fold((f64::MIN, 0), |a, b| if b.0 > a.0 { b } else { a }) +} + +/// Segmental SNR (dB): mean over 20 ms segments holding signal (each clamped to [-10, 40]). +pub fn seg_snr(orig: &[f32], dec: &[f32], shift: isize) -> f64 { + let mut acc = 0f64; + let mut n = 0; + let mut i = FRAME; + while i + FRAME <= orig.len() - FRAME { + let (mut s, mut e) = (0f64, 0f64); + for k in i..i + FRAME { + let j = k as isize + shift; + let d = if j >= 0 && (j as usize) < dec.len() { dec[j as usize] as f64 } else { 0.0 }; + s += (orig[k] as f64).powi(2); + e += (orig[k] as f64 - d).powi(2); + } + if s / FRAME as f64 > 1e-5 { + acc += (10.0 * (s / e.max(1e-30)).log10()).clamp(-10.0, 40.0); + n += 1; + } + i += FRAME; + } + acc / n.max(1) as f64 +} + +/// Average bitrate (bits per second) of a packet sequence of 20 ms packets. +pub fn bitrate(packets: &[Vec]) -> f64 { + packets.iter().map(|p| p.len() * 8).sum::() as f64 / (packets.len() as f64 * 0.02) +} + +/// Low-pass filtered copy (FFT brick wall) used to compare band-limited modes fairly. +pub fn lowpass(x: &[f32], fc: f64) -> Vec { + let n = x.len().next_power_of_two(); + let mut re: Vec = x.iter().map(|&v| v as f64).collect(); + re.resize(n, 0.0); + let mut im = vec![0f64; n]; + fft(&mut re, &mut im, false); + for k in 0..n { + let f = (k.min(n - k)) as f64 * SR / n as f64; + if f > fc { + re[k] = 0.0; + im[k] = 0.0; + } + } + fft(&mut re, &mut im, true); + re[..x.len()].iter().map(|&v| (v / n as f64) as f32).collect() +} + +pub fn fft(re: &mut [f64], im: &mut [f64], inverse: bool) { + let n = re.len(); + let mut j = 0; + for i in 1..n { + let mut bit = n >> 1; + while j & bit != 0 { + j ^= bit; + bit >>= 1; + } + j |= bit; + if i < j { + re.swap(i, j); + im.swap(i, j); + } + } + let mut len = 2; + while len <= n { + let ang = if inverse { 2.0 } else { -2.0 } * std::f64::consts::PI / len as f64; + for s in (0..n).step_by(len) { + for k in 0..len / 2 { + let (wr, wi) = ((ang * k as f64).cos(), (ang * k as f64).sin()); + let (a, b) = (s + k, s + k + len / 2); + let tr = re[b] * wr - im[b] * wi; + let ti = re[b] * wi + im[b] * wr; + re[b] = re[a] - tr; + im[b] = im[a] - ti; + re[a] += tr; + im[a] += ti; + } + } + len <<= 1; + } +} + +// ---- Ogg Opus (RFC 3533 + RFC 7845) for the ffmpeg oracle ---- + +fn ogg_crc(data: &[u8]) -> u32 { + let mut crc = 0u32; + for &b in data { + crc ^= (b as u32) << 24; + for _ in 0..8 { + crc = if crc & 0x8000_0000 != 0 { (crc << 1) ^ 0x04C1_1DB7 } else { crc << 1 }; + } + } + crc +} + +fn ogg_page(out: &mut Vec, packet: &[u8], header_type: u8, granule: u64, seq: u32) { + let mut lacing = vec![255u8; packet.len() / 255]; + lacing.push((packet.len() % 255) as u8); + let start = out.len(); + out.extend_from_slice(b"OggS"); + out.push(0); + out.push(header_type); + out.extend_from_slice(&granule.to_le_bytes()); + out.extend_from_slice(&0x4543_4F50u32.to_le_bytes()); + out.extend_from_slice(&seq.to_le_bytes()); + out.extend_from_slice(&0u32.to_le_bytes()); + out.push(lacing.len() as u8); + out.extend_from_slice(&lacing); + out.extend_from_slice(packet); + let crc = ogg_crc(&out[start..]); + out[start + 22..start + 26].copy_from_slice(&crc.to_le_bytes()); +} + +/// An Ogg Opus file holding `packets` for `len` samples of audio. +pub fn ogg_opus(enc: &OpusEncoder, packets: &[Vec], len: usize) -> Vec { + let mut out = Vec::new(); + ogg_page(&mut out, &enc.opus_head(), 0x02, 0, 0); + let vendor = b"effectcraft-opusenc"; + let mut tags = b"OpusTags".to_vec(); + tags.extend_from_slice(&(vendor.len() as u32).to_le_bytes()); + tags.extend_from_slice(vendor); + tags.extend_from_slice(&0u32.to_le_bytes()); + ogg_page(&mut out, &tags, 0, 0, 1); + let pre_skip = enc.pre_skip() as u64; + for (i, p) in packets.iter().enumerate() { + let last = i + 1 == packets.len(); + let granule = if last { pre_skip + len as u64 } else { ((i + 1) * FRAME) as u64 }; + ogg_page(&mut out, p, if last { 0x04 } else { 0 }, granule, i as u32 + 2); + } + out +} diff --git a/crates/audio/tests/codecs/opusenc/ffmpeg.rs b/crates/audio/tests/codecs/opusenc/ffmpeg.rs new file mode 100644 index 000000000..41965be20 --- /dev/null +++ b/crates/audio/tests/codecs/opusenc/ffmpeg.rs @@ -0,0 +1,83 @@ +//! ffmpeg as a second, external oracle: every mode/bandwidth written as Ogg Opus must decode +//! without errors with both ffmpeg's native `opus` decoder and `libopus`, honour the pre-skip +//! and end granule, and reproduce the input. Skipped when ffmpeg is not installed. + +use super::common; + +use std::process::Command; + +use common::*; +use craft_audio::opus::encoder::{Bandwidth, Mode, OpusEncoder}; + +/// ffmpeg, found and skipped by craft-media's oracle rules (EffectCraft looked in Homebrew's +/// directory, then `PATH`). +fn ffmpeg() -> Option { + crate::testkit::ffmpeg_or_skip("opus encoder ffmpeg oracle") +} + +/// Decodes an Ogg Opus file with ffmpeg's decoder `codec`; returns interleaved f32 output. +fn ffmpeg_decode(exe: &std::path::Path, codec: &str, file: &[u8], channels: usize, tag: &str) -> Option> { + let has = match Command::new(exe).args(["-hide_banner", "-decoders"]).output() { + Ok(out) => out, + Err(e) => return crate::common::skip(&format!("ffmpeg -decoders failed: {e}")), + }; + if !String::from_utf8_lossy(&has.stdout).lines().any(|l| l.split_whitespace().nth(1) == Some(codec)) { + return crate::common::skip(&format!("ffmpeg has no {codec} decoder")); + } + let path = std::env::temp_dir().join(format!("effectcraft-opusenc-{}-{tag}.opus", std::process::id())); + std::fs::write(&path, file).unwrap(); + let out = Command::new(exe) + .args(["-v", "error", "-c:a", codec, "-i"]) + .arg(&path) + .args(["-f", "f32le", "-ac", &channels.to_string(), "-ar", "48000", "-"]) + .output() + .expect("run ffmpeg"); + let _ = std::fs::remove_file(&path); + let stderr = String::from_utf8_lossy(&out.stderr); + assert!(out.status.success(), "ffmpeg ({codec}) failed: {stderr}"); + // ffmpeg's native decoder also makes its raw-PCM muxer print a timestamp complaint for + // narrowband SILK streams written by libopus itself; that line is about ffmpeg's own + // timestamping, not the bitstream, so it is the one message tolerated. + let errors: Vec<&str> = stderr.lines().filter(|l| !l.trim().is_empty() && !l.contains("non monotonically increasing dts to muxer")).collect(); + assert!(errors.is_empty(), "ffmpeg ({codec}) reported errors: {errors:?}"); + Some(out.stdout.as_chunks::<4>().0.iter().map(|b| f32::from_le_bytes([b[0], b[1], b[2], b[3]])).collect()) +} + +#[test] +fn ffmpeg_decodes_every_mode() { + let Some(exe) = ffmpeg() else { + eprintln!("ffmpeg not installed; skipping the oracle test"); + return; + }; + let len = 48_000; + let sp = speech(len + 30_000, 3); + let left = sp[..len].to_vec(); + let right: Vec = sp[30_000..].iter().map(|v| v * 0.7).collect(); + let cases = [ + (1usize, Mode::Silk, Bandwidth::Narrowband, 8_000u32, 3_400.0, 8.0), + (1, Mode::Silk, Bandwidth::Mediumband, 11_000, 5_000.0, 10.0), + (1, Mode::Silk, Bandwidth::Wideband, 16_000, 7_000.0, 12.0), + (2, Mode::Silk, Bandwidth::Wideband, 32_000, 7_000.0, 12.0), + (1, Mode::Hybrid, Bandwidth::SuperWideband, 24_000, 24_000.0, 10.0), + (1, Mode::Hybrid, Bandwidth::Fullband, 32_000, 24_000.0, 10.0), + (2, Mode::Hybrid, Bandwidth::Fullband, 64_000, 24_000.0, 10.0), + (2, Mode::Celt, Bandwidth::Fullband, 96_000, 24_000.0, 10.0), + ]; + for (c, mode, bw, rate, fc, min_snr) in cases { + let planar: Vec> = if c == 1 { vec![left.clone()] } else { vec![left.clone(), right.clone()] }; + let (enc, packets) = encode(OpusEncoder::with_mode(c as u8, rate, mode, bw).unwrap(), &planar); + let file = ogg_opus(&enc, &packets, len); + for codec in ["opus", "libopus"] { + let Some(dec) = ffmpeg_decode(&exe, codec, &file, c, &format!("{mode:?}{bw:?}{c}{codec}")) else { + continue; + }; + assert_eq!(dec.len(), len * c, "{codec}: pre-skip and end granule must be honoured"); + for (ch, x) in planar.iter().enumerate() { + let d: Vec = (0..len).map(|i| dec[i * c + ch]).collect(); + let (snr, delay) = snr_search(&lowpass(x, fc), &d, 6); + eprintln!("ffmpeg {codec:7} {mode:?}/{bw:?} {c} ch @ {} kb/s, channel {ch}: SNR {snr:.1} dB (residual delay {delay})", rate / 1000); + assert!(snr > min_snr, "{codec} {mode:?}/{bw:?}: SNR {snr:.1} dB"); + } + } + } +} diff --git a/crates/audio/tests/codecs/opusenc/mod.rs b/crates/audio/tests/codecs/opusenc/mod.rs new file mode 100644 index 000000000..f2a886463 --- /dev/null +++ b/crates/audio/tests/codecs/opusenc/mod.rs @@ -0,0 +1,5 @@ +//! EffectCraft's `effectcraft-opusenc` integration tests. + +mod common; +mod ffmpeg; +mod modes; diff --git a/crates/audio/tests/codecs/opusenc/modes.rs b/crates/audio/tests/codecs/opusenc/modes.rs new file mode 100644 index 000000000..263858560 --- /dev/null +++ b/crates/audio/tests/codecs/opusenc/modes.rs @@ -0,0 +1,292 @@ +//! SILK, hybrid and CELT round trips through FilmCraft's spec-derived Opus decoder: every +//! mode/bandwidth decodes, waveform quality on speech-like and music-like signals, bitrate +//! tracking, stereo, the application tables and delay alignment. + +use super::common; + +use common::*; +use craft_audio::opus::encoder::{Application, Bandwidth, Error, Mode, OpusEncoder}; + +struct Run { + snr: f64, + seg: f64, + delay: isize, + kbps: f64, + packets: Vec>, +} + +/// Encodes `planar`, decodes, and measures each channel against `reference` (searching a +/// residual delay of up to ±6 samples). +fn run(enc: OpusEncoder, planar: &[Vec], reference: &[Vec]) -> Vec { + let (enc, packets) = encode(enc, planar); + let dec = decode(&enc.opus_head(), &packets); + let kbps = bitrate(&packets) / 1000.0; + reference + .iter() + .zip(&dec) + .map(|(r, d)| { + assert!(d.len() >= r.len(), "decoded {} < {}", d.len(), r.len()); + assert!(d.iter().all(|v| v.is_finite())); + let (snr, delay) = snr_search(r, d, 6); + Run { snr, seg: seg_snr(r, d, delay), delay, kbps, packets: packets.clone() } + }) + .collect() +} + +fn mono(mode: Mode, bw: Bandwidth, rate: u32, x: &[f32], fc: f64) -> Run { + let reference = if fc < 20_000.0 { lowpass(x, fc) } else { x.to_vec() }; + run(OpusEncoder::with_mode(1, rate, mode, bw).unwrap(), &[x.to_vec()], &[reference]).remove(0) +} + +/// Upper edge of the band a SILK bandwidth keeps (for band-limited references). +fn band_edge(bw: Bandwidth) -> f64 { + match bw { + Bandwidth::Narrowband => 3_400.0, + Bandwidth::Mediumband => 5_000.0, + Bandwidth::Wideband => 7_000.0, + _ => 24_000.0, + } +} + +const LEN: usize = 96_000; + +#[test] +fn every_mode_and_bandwidth_decodes() { + let sp = speech(LEN + 40_000, 11); + let left = sp[..LEN].to_vec(); + let right: Vec = sp[40_000..].iter().map(|v| v * 0.6).collect(); + for (mode, bw, config, rate) in [ + (Mode::Silk, Bandwidth::Narrowband, 1u8, 8_000u32), + (Mode::Silk, Bandwidth::Mediumband, 5, 11_000), + (Mode::Silk, Bandwidth::Wideband, 9, 16_000), + (Mode::Hybrid, Bandwidth::SuperWideband, 13, 24_000), + (Mode::Hybrid, Bandwidth::Fullband, 15, 32_000), + (Mode::Celt, Bandwidth::Fullband, 31, 64_000), + ] { + for c in [1usize, 2] { + let planar = if c == 1 { vec![left.clone()] } else { vec![left.clone(), right.clone()] }; + let reference: Vec> = planar.iter().map(|x| lowpass(x, band_edge(bw))).collect(); + let enc = OpusEncoder::with_mode(c as u8, rate * c as u32, mode, bw).unwrap(); + assert_eq!((enc.mode(), enc.bandwidth()), (mode, bw)); + let res = run(enc, &planar, &reference); + for p in &res[0].packets { + assert_eq!(p[0], (config << 3) | if c == 2 { 4 } else { 0 }, "TOC of {mode:?}/{bw:?}"); + assert!(p.len() >= 3 && p.len() <= 1276); + } + for (ch, r) in res.iter().enumerate() { + eprintln!( + "{mode:?}/{bw:?} {c} ch @ {} kb/s: {:.1} kb/s, channel {ch} SNR {:.1} dB, segSNR {:.1} dB, residual delay {}", + rate * c as u32 / 1000, + r.kbps, + r.snr, + r.seg, + r.delay + ); + assert!(r.snr > 8.0, "{mode:?}/{bw:?} {c} ch channel {ch}: SNR {:.1} dB", r.snr); + assert!(r.delay.abs() <= 1, "{mode:?}/{bw:?}: residual delay {} (pre-skip mismatch)", r.delay); + } + } + } +} + +#[test] +fn silk_speech_quality() { + let x = speech(LEN, 7); + for (bw, rate, min_snr) in [ + (Bandwidth::Narrowband, 8_000u32, 10.0), + (Bandwidth::Mediumband, 11_000, 14.0), + (Bandwidth::Wideband, 12_000, 14.0), + (Bandwidth::Wideband, 16_000, 18.0), + (Bandwidth::Wideband, 24_000, 24.0), + ] { + let r = mono(Mode::Silk, bw, rate, &x, band_edge(bw)); + eprintln!("SILK {bw:?} speech @ {} kb/s: {:.1} kb/s, SNR {:.1} dB, segSNR {:.1} dB", rate / 1000, r.kbps, r.snr, r.seg); + assert!(r.snr > min_snr && r.seg > min_snr - 3.0, "SILK {bw:?} @ {rate}: SNR {:.1} / seg {:.1}", r.snr, r.seg); + } +} + +#[test] +fn silk_beats_celt_on_speech_at_low_rate() { + let x = speech(LEN, 5); + // Both measured against the full-band input. + let silk = mono(Mode::Silk, Bandwidth::Wideband, 12_000, &x, 24_000.0); + let celt = mono(Mode::Celt, Bandwidth::Fullband, 12_000, &x, 24_000.0); + eprintln!( + "12 kb/s mono speech: SILK WB {:.1} dB (seg {:.1}) at {:.1} kb/s vs CELT {:.1} dB (seg {:.1}) at {:.1} kb/s", + silk.snr, silk.seg, silk.kbps, celt.snr, celt.seg, celt.kbps + ); + assert!(silk.snr > celt.snr + 3.0, "SILK {:.1} dB vs CELT {:.1} dB", silk.snr, celt.snr); + assert!(silk.seg > celt.seg + 3.0, "SILK seg {:.1} dB vs CELT seg {:.1} dB", silk.seg, celt.seg); +} + +#[test] +fn hybrid_quality() { + let sp = speech(LEN, 9); + let mu = music(LEN); + for (name, x, bw, rate, min_snr) in [ + ("speech", &sp, Bandwidth::SuperWideband, 24_000u32, 18.0), + ("speech", &sp, Bandwidth::Fullband, 32_000, 20.0), + ("music", &mu, Bandwidth::SuperWideband, 24_000, 9.0), + ("music", &mu, Bandwidth::Fullband, 32_000, 10.0), + ] { + let r = mono(Mode::Hybrid, bw, rate, x, 24_000.0); + eprintln!("hybrid {bw:?} {name} @ {} kb/s: {:.1} kb/s, SNR {:.1} dB, segSNR {:.1} dB", rate / 1000, r.kbps, r.snr, r.seg); + assert!(r.snr > min_snr, "hybrid {bw:?} {name}: SNR {:.1} dB", r.snr); + // Constant packet size. + assert!(r.packets.iter().all(|p| p.len() == rate as usize / 400)); + } +} + +#[test] +fn silk_music_quality() { + let x = music(LEN); + for (rate, min_snr) in [(12_000u32, 5.0), (16_000, 8.0), (24_000, 13.0)] { + let r = mono(Mode::Silk, Bandwidth::Wideband, rate, &x, 7_000.0); + eprintln!("SILK WB music @ {} kb/s: {:.1} kb/s, SNR {:.1} dB", rate / 1000, r.kbps, r.snr); + assert!(r.snr > min_snr, "SILK music @ {rate}: SNR {:.1} dB", r.snr); + } +} + +#[test] +fn bitrate_tracking() { + let sp = speech(LEN, 13); + let mu = music(LEN); + for (name, x) in [("speech", &sp), ("music", &mu)] { + for (mode, bw, rate) in [ + (Mode::Silk, Bandwidth::Narrowband, 6_000u32), + (Mode::Silk, Bandwidth::Narrowband, 9_000), + (Mode::Silk, Bandwidth::Mediumband, 11_000), + (Mode::Silk, Bandwidth::Wideband, 12_000), + (Mode::Silk, Bandwidth::Wideband, 20_000), + (Mode::Silk, Bandwidth::Wideband, 32_000), + (Mode::Hybrid, Bandwidth::SuperWideband, 20_000), + (Mode::Hybrid, Bandwidth::Fullband, 36_000), + ] { + let r = mono(mode, bw, rate, x, band_edge(bw)); + let ratio = r.kbps * 1000.0 / rate as f64; + eprintln!( + "{name} {mode:?}/{bw:?} target {} kb/s: actual {:.2} kb/s ({:+.0} %), SNR {:.1} dB", + rate / 1000, + r.kbps, + (ratio - 1.0) * 100.0, + r.snr + ); + assert!((0.75..=1.25).contains(&ratio), "{name} {mode:?}/{bw:?} @ {rate}: {:.2} kb/s", r.kbps); + } + } +} + +#[test] +fn stereo_silk_and_hybrid() { + let sp = speech(LEN + 40_000, 21); + let a = sp[..LEN].to_vec(); + let b: Vec = sp[40_000..].iter().map(|v| v * 0.7).collect(); + let mu = music(LEN); + let cases: Vec<(&str, Vec, Vec)> = vec![ + ("dual mono", a.clone(), a.clone()), + ("panned", a.iter().map(|v| v * 0.9).collect(), a.iter().map(|v| v * 0.35).collect()), + ("independent", a.clone(), b), + ("speech+music", a.iter().zip(&mu).map(|(x, y)| 0.8 * x + 0.3 * y).collect(), a.iter().zip(&mu).map(|(x, y)| 0.3 * x + 0.8 * y).collect()), + ]; + for (name, l, r) in &cases { + for (mode, bw, rate, min_snr) in [(Mode::Silk, Bandwidth::Wideband, 32_000u32, 9.0), (Mode::Hybrid, Bandwidth::Fullband, 64_000, 9.0)] { + let fc = band_edge(bw); + let res = run(OpusEncoder::with_mode(2, rate, mode, bw).unwrap(), &[l.clone(), r.clone()], &[lowpass(l, fc), lowpass(r, fc)]); + eprintln!( + "stereo {name} {mode:?}/{bw:?} @ {} kb/s: {:.1} kb/s, L {:.1} dB, R {:.1} dB", + rate / 1000, + res[0].kbps, + res[0].snr, + res[1].snr + ); + assert!(res[0].snr > min_snr && res[1].snr > min_snr, "stereo {name} {mode:?}: L {:.1} R {:.1}", res[0].snr, res[1].snr); + } + } +} + +#[test] +fn application_tables_and_pre_skip() { + let cases = [ + (Application::Voip, 1u8, 8_000u32, Mode::Silk, Bandwidth::Narrowband), + (Application::Voip, 1, 11_000, Mode::Silk, Bandwidth::Mediumband), + (Application::Voip, 1, 16_000, Mode::Silk, Bandwidth::Wideband), + (Application::Voip, 1, 24_000, Mode::Hybrid, Bandwidth::SuperWideband), + (Application::Voip, 1, 32_000, Mode::Hybrid, Bandwidth::Fullband), + (Application::Voip, 1, 48_000, Mode::Celt, Bandwidth::Fullband), + (Application::Voip, 2, 32_000, Mode::Silk, Bandwidth::Wideband), + (Application::Audio, 1, 12_000, Mode::Silk, Bandwidth::Wideband), + (Application::Audio, 1, 24_000, Mode::Hybrid, Bandwidth::Fullband), + (Application::Audio, 2, 48_000, Mode::Hybrid, Bandwidth::Fullband), + (Application::Audio, 1, 64_000, Mode::Celt, Bandwidth::Fullband), + (Application::Audio, 2, 128_000, Mode::Celt, Bandwidth::Fullband), + ]; + for (app, c, rate, mode, bw) in cases { + let e = OpusEncoder::with_application(c, rate, app).unwrap(); + assert_eq!((e.mode(), e.bandwidth(), e.application()), (mode, bw, app), "{app:?} {c} ch @ {rate}"); + assert_eq!(e.pre_skip(), 120, "every mode is aligned to a 120-sample delay"); + } + assert_eq!(OpusEncoder::new(2, 128_000).unwrap().mode(), Mode::Celt); + assert_eq!(OpusEncoder::new(1, 12_000).unwrap().application(), Application::Audio); +} + +/// EffectCraft's encoder panicked here; craft-audio returns an error (never crash). +#[test] +fn invalid_mode_bandwidth_is_an_error() { + assert_eq!(OpusEncoder::with_mode(1, 16_000, Mode::Silk, Bandwidth::Fullband).err(), Some(Error::ModeBandwidth(Mode::Silk, Bandwidth::Fullband))); + assert_eq!(OpusEncoder::new(3, 128_000).err(), Some(Error::Channels(3))); + assert_eq!(OpusEncoder::new(0, 128_000).err(), Some(Error::Channels(0))); + // From 86 channels the per-channel minimum bitrate passes the 510 kb/s cap. + for channels in [86, u8::MAX] { + assert_eq!(OpusEncoder::new(channels, 128_000).err(), Some(Error::Channels(channels))); + let app = OpusEncoder::with_application(channels, 128_000, Application::Voip).err(); + assert_eq!(app, Some(Error::Channels(channels))); + } +} + +#[test] +fn extreme_inputs_stay_decodable() { + let len = 24_000; + let mut s = 1u32; + let noise: Vec = (0..len) + .map(|_| { + s = s.wrapping_mul(1_664_525).wrapping_add(1_013_904_223); + (s >> 8) as f32 / (1u32 << 23) as f32 - 1.0 + }) + .collect(); + let square: Vec = (0..len).map(|i| if (i / 60) % 2 == 0 { 0.999 } else { -0.999 }).collect(); + let clicks: Vec = (0..len).map(|i| if i % 4_801 == 0 { 1.0 } else { 0.0 }).collect(); + let dc = vec![0.7f32; len]; + let nan: Vec = (0..len).map(|i| if i % 1000 == 0 { f32::NAN } else { 0.1 * (i as f32 * 0.01).sin() }).collect(); + for (name, x) in [("noise", &noise), ("square", &square), ("clicks", &clicks), ("dc", &dc), ("nan", &nan)] { + for (mode, bw, rate) in + [(Mode::Silk, Bandwidth::Narrowband, 6_000u32), (Mode::Silk, Bandwidth::Wideband, 40_000), (Mode::Hybrid, Bandwidth::Fullband, 24_000)] + { + for c in [1usize, 2] { + let planar: Vec> = if c == 1 { vec![x.clone()] } else { vec![x.clone(), x.iter().map(|v| -0.5 * v).collect()] }; + let (enc, packets) = encode(OpusEncoder::with_mode(c as u8, rate * c as u32, mode, bw).unwrap(), &planar); + assert!(packets.iter().all(|p| p.len() >= 3 && p.len() <= 1276), "{name} {mode:?}: packet sizes"); + let dec = decode(&enc.opus_head(), &packets); + let peak = dec.iter().flatten().fold(0f32, |m, v| m.max(v.abs())); + let snr = if name == "nan" { f64::NAN } else { snr_search(&planar[0], &dec[0], 2).0 }; + eprintln!("{name} {mode:?}/{bw:?} {c} ch: peak {peak:.2}, SNR {snr:.1} dB, {:.1} kb/s", bitrate(&packets) / 1000.0); + assert!(dec.iter().all(|ch| ch.iter().all(|v| v.is_finite() && v.abs() <= 4.0)), "{name} {mode:?}/{bw:?} {c} ch: bad output"); + } + } + } +} + +#[test] +fn silk_and_hybrid_silence() { + for (mode, bw, rate) in [(Mode::Silk, Bandwidth::Wideband, 16_000u32), (Mode::Hybrid, Bandwidth::Fullband, 32_000)] { + let x = vec![0f32; 48_000]; + let (enc, packets) = encode(OpusEncoder::with_mode(1, rate, mode, bw).unwrap(), std::slice::from_ref(&x)); + let dec = decode(&enc.opus_head(), &packets); + let peak = dec[0].iter().fold(0f32, |m, v| m.max(v.abs())); + let kbps = bitrate(&packets) / 1000.0; + eprintln!("{mode:?} silence: {kbps:.2} kb/s, decoded peak {peak:.2e}"); + assert!(peak < 1e-3, "{mode:?} silence decodes to peak {peak}"); + if mode == Mode::Silk { + assert!(kbps < 4.0, "SILK silence costs {kbps:.2} kb/s"); + } + } +} diff --git a/crates/audio/tests/codecs/track_decoder.rs b/crates/audio/tests/codecs/track_decoder.rs index 0607db4ac..fee8f1f97 100644 --- a/crates/audio/tests/codecs/track_decoder.rs +++ b/crates/audio/tests/codecs/track_decoder.rs @@ -1,7 +1,8 @@ //! `TrackDecoder` on streams an app's own demuxer hands over without codec configuration: -//! ADTS-framed AAC (MPEG-TS). +//! ADTS-framed AAC (MPEG-TS) and Opus without an `OpusHead` (RTP). use craft_audio::aac::{Encoder as AacEncoder, EncoderConfig}; +use craft_audio::opus::Encoder as OpusEncoder; use craft_audio::{CodecId, CodecParams, TrackDecoder}; fn tone(n: usize) -> Vec { @@ -32,3 +33,20 @@ fn adts_aac_without_extradata() { let raw = &units[0][7..]; assert!(TrackDecoder::new(&CodecParams::new(CodecId::Aac), &[raw]).is_err()); } + +#[test] +fn opus_without_opus_head() { + let mut enc = OpusEncoder::new(1, 64_000).unwrap(); + let pcm = tone(OpusEncoder::FRAME_SIZE * 10); + let packets: Vec> = pcm.chunks(OpusEncoder::FRAME_SIZE).map(|f| enc.encode_float(f).unwrap()).collect(); + + let params = CodecParams::new(CodecId::Opus).with_channels(1).with_sample_rate(48_000); + let mut dec = TrackDecoder::new(¶ms, &[]).unwrap(); + assert_eq!((dec.sample_rate(), dec.channels()), (48_000, 1)); + for p in &packets { + assert_eq!(dec.decode(p).unwrap().frames(), OpusEncoder::FRAME_SIZE); + } + + // More than two channels need the mapping an OpusHead carries. + assert!(TrackDecoder::new(&CodecParams::new(CodecId::Opus).with_channels(6), &[]).is_err()); +} diff --git a/crates/audio/tests/codecs/vorbis_encoder_oracle.rs b/crates/audio/tests/codecs/vorbis_encoder_oracle.rs new file mode 100644 index 000000000..342f4fd9a --- /dev/null +++ b/crates/audio/tests/codecs/vorbis_encoder_oracle.rs @@ -0,0 +1,131 @@ +//! ffmpeg as an external oracle for the pure-Rust Vorbis encoder: both of its decoders (its own +//! and libvorbis) read every file exactly as long as the input, surround comes out on the right +//! speakers, and the tags reach ffprobe. Skipped when ffmpeg is not installed. + +use std::path::Path; +use std::process::Command; + +use craft_audio::{AudioBuffer, ChannelLayout, EncodeFormat, EncodeOptions, OggVorbisOptions, encode, encode_ogg_vorbis}; + +/// A different tone on every channel, so each output channel can be traced to its input. +fn tones(rate: u32, channels: usize, frames: usize) -> AudioBuffer { + let channels = (0..channels) + .map(|c| { + let f = 220.0 * (c + 1) as f32; + (0..frames).map(|i| (std::f32::consts::TAU * f * i as f32 / rate as f32).sin() * 0.3).collect() + }) + .collect(); + AudioBuffer { sample_rate: rate, channels } +} + +/// Decodes `file` with ffmpeg's `decoder`; planar output in ffmpeg's (WAV) channel order. +fn ffmpeg_decode(ff: &Path, decoder: &str, file: &[u8], channels: usize, tag: &str) -> Vec> { + let path = crate::testkit::fixtures_dir("vorbis-encoder").join(format!("{tag}.ogg")); + std::fs::write(&path, file).unwrap(); + let out = Command::new(ff) + .args(["-v", "error", "-xerror", "-c:a", decoder, "-i"]) + .arg(&path) + .args(["-f", "f32le", "-c:a", "pcm_f32le", "-"]) + .output() + .unwrap(); + assert!(out.status.success() && out.stderr.is_empty(), "{tag} {decoder}: {}", String::from_utf8_lossy(&out.stderr)); + let pcm: Vec = out.stdout.as_chunks::<4>().0.iter().map(|&b| f32::from_le_bytes(b)).collect(); + (0..channels).map(|c| pcm.iter().skip(c).step_by(channels).copied().collect()).collect() +} + +/// ffmpeg's decoders to check: its own `vorbis` always, and `libvorbis` when ffmpeg has it. +fn decoders(ff: &Path) -> Vec<&'static str> { + let list = + Command::new(ff).args(["-hide_banner", "-decoders"]).output().map(|o| String::from_utf8_lossy(&o.stdout).into_owned()).unwrap_or_default(); + if list.lines().any(|l| l.split_whitespace().nth(1) == Some("libvorbis")) { + vec!["vorbis", "libvorbis"] + } else { + eprintln!("SKIPPED (libvorbis decoder): this ffmpeg has no libvorbis; checking its own decoder only"); + vec!["vorbis"] + } +} + +fn correlation(a: &[f32], b: &[f32]) -> f64 { + let (mut ab, mut aa, mut bb) = (0f64, 0f64, 0f64); + for (&x, &y) in a.iter().zip(b) { + ab += f64::from(x) * f64::from(y); + aa += f64::from(x) * f64::from(x); + bb += f64::from(y) * f64::from(y); + } + ab / (aa * bb).sqrt().max(1e-20) +} + +/// Vorbis order (§4.3.9) of the WAV-order layouts the test writes, per channel count. +const VORBIS_ORDER: [&[usize]; 8] = + [&[0], &[0, 1], &[0, 2, 1], &[0, 1, 2, 3], &[0, 2, 1, 3, 4], &[0, 2, 1, 4, 5, 3], &[0, 2, 1, 5, 6, 4, 3], &[0, 2, 1, 6, 7, 4, 5, 3]]; + +/// 1 to 8 channels through `encode` with a WAV-order layout, at exactly the input's length out of +/// both decoders: ffmpeg's own puts every channel back on its WAV speaker, and libvorbis (through +/// ffmpeg, which passes its order through) gives the file's Vorbis order. +#[test] +fn every_channel_count_decodes_on_the_right_speakers() { + let ff = require_ffmpeg!(); + for channels in 1..=8u16 { + let layout = if channels == 7 { ChannelLayout::from_mask(0x70F) } else { ChannelLayout::default_for(channels) }; + let buf = tones(48_000, channels.into(), 24_000 + 77); + let opts = EncodeOptions::new(EncodeFormat::Ogg).with_layout(layout).with_quality(0.6); + let file = encode(&buf, &opts).unwrap(); + for decoder in decoders(&ff) { + let out = ffmpeg_decode(&ff, decoder, &file, channels.into(), &format!("tones{channels}-{decoder}")); + for (j, got) in out.iter().enumerate() { + let c = if decoder == "libvorbis" { VORBIS_ORDER[usize::from(channels) - 1][j] } else { j }; + assert_eq!(got.len(), buf.frames(), "{channels} ch {decoder}: channel {j} length"); + let best = (0..buf.channels.len()).max_by(|&a, &b| { + correlation(&buf.channels[a][2048..], &got[2048..]).total_cmp(&correlation(&buf.channels[b][2048..], &got[2048..])) + }); + assert_eq!(best, Some(c), "{channels} ch {decoder}: output channel {c} carries input {best:?}"); + let r = correlation(&buf.channels[c][2048..], &got[2048..]); + assert!(r > 0.95, "{channels} ch {decoder}: channel {c} correlation {r:.3}"); + } + } + } +} + +/// The comment header's tags, UTF-8 values included, are what ffprobe reports; a name that is not +/// a valid field name is left out. +#[test] +fn tags_reach_ffprobe() { + let Some(probe) = crate::common::tool("ffprobe").or_else(|| crate::common::skip("ffprobe not found")) else { return }; + let opts = OggVorbisOptions::new(3.0) + .with_tag("TITLE", "SmörgĂ¥sbord ✓") + .with_tag("ARTIST", "Craft") + .with_tag("bad=name", "dropped") + .with_vendor("craft-audio test"); + let file = encode_ogg_vorbis(&tones(44_100, 2, 44_100), &opts).unwrap(); + let path = crate::testkit::fixtures_dir("vorbis-encoder").join("tags.ogg"); + std::fs::write(&path, &file).unwrap(); + let out = Command::new(probe).args(["-v", "error", "-of", "json", "-show_streams", "-show_format"]).arg(&path).output().unwrap(); + assert!(out.status.success(), "{}", String::from_utf8_lossy(&out.stderr)); + let j: serde_json::Value = serde_json::from_slice(&out.stdout).unwrap(); + let s = &j["streams"][0]; + assert_eq!((s["codec_name"].as_str(), s["channels"].as_u64(), s["sample_rate"].as_str()), (Some("vorbis"), Some(2), Some("44100"))); + let tags = &s["tags"]; + assert_eq!((tags["TITLE"].as_str(), tags["ARTIST"].as_str()), (Some("SmörgĂ¥sbord ✓"), Some("Craft"))); + assert!(tags.as_object().unwrap().keys().all(|k| !k.contains("bad")), "{tags}"); + assert_eq!(j["format"]["duration"].as_str().map(|d| d.parse::().unwrap()), Some(1.0)); +} + +/// The lowest and highest qualities and the bit-rate path decode at the exact length, mono at +/// 44.1 kHz and stereo at 22.05 kHz. +#[test] +fn quality_extremes_and_rates_decode_exactly() { + let ff = require_ffmpeg!(); + for (rate, channels, quality) in [(44_100u32, 1usize, -1.0f32), (44_100, 1, 10.0), (22_050, 2, -1.0), (22_050, 2, 10.0), (96_000, 2, 5.0)] { + let buf = tones(rate, channels, rate as usize / 2 + 13); + let file = encode_ogg_vorbis(&buf, &OggVorbisOptions::new(quality)).unwrap(); + for decoder in decoders(&ff) { + let out = ffmpeg_decode(&ff, decoder, &file, channels, &format!("q{quality}-{rate}-{channels}-{decoder}")); + assert!(out.iter().all(|c| c.len() == buf.frames()), "q{quality} {rate} Hz {decoder}: {}", out[0].len()); + let r = correlation(&buf.channels[0][2048..], &out[0][2048..]); + assert!(r > 0.9, "q{quality} {rate} Hz {decoder}: correlation {r:.3}"); + } + } + let low = craft_audio::encode_ogg(&tones(48_000, 2, 48_000), craft_audio::OGG_BITRATES.0).unwrap(); + let high = craft_audio::encode_ogg(&tones(48_000, 2, 48_000), craft_audio::OGG_BITRATES.1).unwrap(); + assert!(low.len() < high.len(), "{} vs {} bytes", low.len(), high.len()); +} diff --git a/crates/audio/tests/flac_decode.rs b/crates/audio/tests/flac_decode.rs new file mode 100644 index 000000000..a1895789b --- /dev/null +++ b/crates/audio/tests/flac_decode.rs @@ -0,0 +1,211 @@ +//! The native FLAC decoder: round trips through this crate's encoder over bit depths, channel +//! counts, block sizes and effort levels; hostile input; and ffmpeg as the reference (oracle; +//! skipped when ffmpeg is absent, a failure with `CRAFT_MEDIA_REQUIRE_ORACLES=1`) for files +//! ffmpeg's encoder writes at every compression level. + +use craft_audio::flac::{DecodeError, Decoder, Encoder, EncoderConfig, STREAMINFO_OFFSET, StreamInfo, decode}; +use std::process::Command; + +fn signal(channels: usize, frames: usize, bits: u32, seed: u32) -> Vec> { + let max = if bits == 32 { i32::MAX as f64 } else { ((1i64 << (bits - 1)) - 1) as f64 }; + let mut s = seed; + (0..channels) + .map(|c| { + (0..frames) + .map(|i| { + s ^= s << 13; + s ^= s >> 17; + s ^= s << 5; + let t = i as f64; + let v = match (i / 3000) % 4 { + 0 => (t * (0.01 + 0.007 * c as f64)).sin() * 0.6 + (t * 0.31).sin() * 0.1, + 1 => (f64::from(s) / f64::from(u32::MAX) - 0.5) * 1.8, + 2 => 0.0, + _ => { + if (i / 40) % 2 == 0 { + 0.99 + } else { + -0.99 + } + } + }; + (v * max).round() as i32 + }) + .collect() + }) + .collect() +} + +fn encode(channels: &[Vec], rate: u32, bits: u8, block: u16, level: u8) -> Vec { + let mut cfg = EncoderConfig::new(rate, channels.len() as u8, bits); + cfg.block_size = block; + cfg.level = level; + let mut enc = Encoder::new(cfg).expect("encoder"); + let mut file = enc.header(&[]); + let refs: Vec<&[i32]> = channels.iter().map(Vec::as_slice).collect(); + file.extend(enc.encode(&refs).expect("encode")); + file.extend(enc.finish()); + let at = STREAMINFO_OFFSET as usize; + file[at..at + 34].copy_from_slice(&enc.streaminfo()); + file +} + +#[test] +fn round_trips_through_the_encoder() { + for (channels, bits) in [(1usize, 8u8), (2, 8), (1, 12), (2, 16), (2, 20), (3, 24), (6, 24), (8, 16), (2, 32), (1, 32)] { + for (block, level) in [(4096u16, 5u8), (16, 0), (1000, 8), (65535, 3), (576, 1)] { + let frames = 9000 + usize::from(block % 97); + let src = signal(channels, frames, u32::from(bits), 0x1234_5679 + u32::from(bits)); + let file = encode(&src, 44_100, bits, block, level); + let got = decode(&file).unwrap_or_else(|e| panic!("{channels}ch {bits}-bit block {block} level {level}: {e}")); + assert_eq!((got.info.channels as usize, got.info.bits_per_sample, got.info.sample_rate), (channels, bits, 44_100)); + assert_eq!(got.channels, src, "{channels}ch {bits}-bit block {block} level {level}"); + } + } +} + +#[test] +fn frames_decode_one_at_a_time() { + let src = signal(2, 10_000, 16, 7); + let file = encode(&src, 48_000, 16, 4096, 5); + let info = StreamInfo::parse(&file[STREAMINFO_OFFSET as usize..]).expect("streaminfo"); + let dec = Decoder::new(info); + // The first frame follows the metadata blocks; find the first position a frame decodes at. + let at = (42..file.len()).find(|&p| dec.decode_frame(&file[p..]).is_ok()).expect("a frame"); + let first = dec.decode_frame(&file[at..]).expect("frame"); + assert_eq!((first.block_size, first.number, first.sample_rate, first.bits_per_sample), (4096, 0, 48_000, 16)); + assert_eq!(first.channels[0], src[0][..4096]); + // Without the STREAMINFO a frame that spells out its own rate and depth still decodes. + let _ = Decoder::without_streaminfo().decode_frame(&file[at..]); +} + +#[test] +fn hostile_input_is_an_error_not_a_panic() { + let src = signal(2, 6000, 16, 11); + let file = encode(&src, 44_100, 16, 1152, 5); + for n in 0..file.len().min(3000) { + let _ = decode(&file[..n]); + } + // A cut file is truncated, not silently shorter. + assert!(decode(&file[..file.len() - 5]).is_err()); + // Every bit flip in the first frames is caught by a CRC or another check, never a panic. + for i in 0..file.len().min(2500) { + for bit in 0..8 { + let mut c = file.clone(); + c[i] ^= 1 << bit; + let _ = decode(&c); + } + } + // A sample limit stops a stream that claims too much. + assert_eq!(craft_audio::flac::decode_with_limit(&file, 1000), Err(DecodeError::TooLarge)); + assert!(decode(b"nope").is_err()); + assert!(decode(b"fLaC").is_err()); +} + +// ---- ffmpeg ---- + +/// Little-endian 16- or 32-bit samples of raw PCM. +fn pcm_values(raw: &[u8], bytes: usize) -> Vec { + if bytes == 2 { + raw.as_chunks::<2>().0.iter().map(|b| i32::from(i16::from_le_bytes(*b))).collect() + } else { + raw.as_chunks::<4>().0.iter().map(|b| i32::from_le_bytes(*b)).collect() + } +} + +/// Planar channels as interleaved samples. +fn interleave(channels: &[Vec]) -> Vec { + let frames = channels.first().map_or(0, Vec::len); + (0..frames).flat_map(|i| channels.iter().map(move |c| c[i])).collect() +} + +fn ffmpeg_run(ff: &std::path::Path, args: &[&str]) -> Vec { + let out = Command::new(ff).args(["-v", "error", "-nostdin", "-y"]).args(args).output().expect("run ffmpeg"); + assert!(out.status.success(), "ffmpeg {args:?}: {}", String::from_utf8_lossy(&out.stderr)); + out.stdout +} + +#[test] +fn ffmpeg_files_decode_to_ffmpegs_samples() { + let Some(ff) = craft_media_testkit::ffmpeg() else { return }; + let dir = craft_media_testkit::fixture_subdir("flac_decode"); + let tones = "aevalsrc=exprs=0.4*sin(2*PI*441*t)+0.2*sin(2*PI*3100*t)|0.3*sin(2*PI*660*t)*sin(2*PI*3*t):s=44100:d=1.3"; + let noise = "anoisesrc=d=1.3:c=white:r=44100:a=0.5:s=3"; + for (src, name) in [(tones, "tones"), (noise, "noise")] { + for channels in [1usize, 2, 6] { + for (fmt, bytes, bits) in [("s16", 2usize, 16u32), ("s32", 4, 24)] { + for level in [0, 3, 5, 8, 12] { + let file = dir.join(format!("ffmpeg_{name}_{channels}ch_{bits}_l{level}.flac")); + let f = file.to_str().expect("path"); + let (ch, lv) = (channels.to_string(), level.to_string()); + ffmpeg_run(&ff, &["-f", "lavfi", "-i", src, "-ac", &ch, "-sample_fmt", fmt, "-compression_level", &lv, "-c:a", "flac", f]); + let theirs = ffmpeg_run(&ff, &["-i", f, "-f", if bits == 16 { "s16le" } else { "s32le" }, "-"]); + let got = + decode(&std::fs::read(&file).expect("read")).unwrap_or_else(|e| panic!("{name} {channels}ch {bits}-bit level {level}: {e}")); + assert_eq!(got.info.channels as usize, channels); + let frames = got.channels[0].len(); + assert_eq!(theirs.len(), frames * channels * bytes, "{name} {channels}ch {bits} l{level}: length"); + // ffmpeg returns 24-bit audio left-aligned in 32 bits. + let shift = if bits == 24 { 8 } else { 0 }; + let mine: Vec = interleave(&got.channels).into_iter().map(|v| v << shift).collect(); + let ok = mine == pcm_values(&theirs, bytes); + assert!(ok, "{name} {channels}ch {bits}-bit level {level}: samples differ from ffmpeg's decode"); + } + } + } + } +} + +#[test] +fn ffmpeg_decodes_our_files_to_our_samples() { + let Some(ff) = craft_media_testkit::ffmpeg() else { return }; + let dir = craft_media_testkit::fixture_subdir("flac_decode"); + for (channels, bits) in [(1usize, 16u8), (2, 16), (2, 24), (6, 16)] { + let src = signal(channels, 20_000, u32::from(bits), 99); + let file = dir.join(format!("ours_{channels}ch_{bits}.flac")); + std::fs::write(&file, encode(&src, 44_100, bits, 4096, 5)).expect("write"); + let theirs = ffmpeg_run(&ff, &["-i", file.to_str().expect("path"), "-f", if bits == 16 { "s16le" } else { "s32le" }, "-"]); + let got = decode(&std::fs::read(&file).expect("read")).expect("decode"); + let bytes = if bits == 16 { 2 } else { 4 }; + let shift = if bytes == 4 { 8 } else { 0 }; + let want: Vec = pcm_values(&theirs, bytes).into_iter().map(|v| v >> shift).collect(); + assert!(interleave(&got.channels) == want, "{channels}ch {bits}-bit: samples differ from ffmpeg's decode"); + } +} + +#[test] +fn rejects_unknown_length_corruption_and_streaminfo_format_mismatch() { + let src = signal(1, 100, 16, 7); + let original = encode(&src, 44_100, 16, 4096, 5); + let at = STREAMINFO_OFFSET as usize; + // STREAMINFO is not CRC-covered. Clear only the 36-bit total sample count. + let mut unknown = original.clone(); + unknown[at + 13] &= 0xf0; + unknown[at + 14..at + 18].fill(0); + assert_eq!(decode(&unknown).unwrap().channels, src); + for tail in [vec![0xff], vec![0x00, 0x00]] { + let mut broken = unknown.clone(); + broken.extend(tail); + assert!(decode(&broken).is_err()); + } + // The encoder spells these in frame headers, independent of STREAMINFO. + let mut wrong_rate = original.clone(); + let mut packed = u64::from_be_bytes(wrong_rate[at + 10..at + 18].try_into().unwrap()); + packed = (packed & ((1u64 << 44) - 1)) | (48_000u64 << 44); + wrong_rate[at + 10..at + 18].copy_from_slice(&packed.to_be_bytes()); + assert!(decode(&wrong_rate).is_err()); + let mut wrong_depth = original; + wrong_depth[at + 12] = (wrong_depth[at + 12] & !1) | 1; + wrong_depth[at + 13] = (wrong_depth[at + 13] & 0x0f) | 0x70; + assert!(decode(&wrong_depth).is_err()); +} + +#[test] +fn caller_constructed_streaminfo_is_validated_before_decoding() { + let file = encode(&signal(1, 100, 16, 9), 44_100, 16, 4096, 5); + let info = StreamInfo::parse(&file[STREAMINFO_OFFSET as usize..]).unwrap(); + for bits in [0, 1, 3, 33, 64, 255] { + let invalid = StreamInfo { bits_per_sample: bits, ..info }; + assert!(matches!(Decoder::new(invalid).decode_frame(&[]), Err(DecodeError::Invalid(_)))); + } +} diff --git a/crates/audio/tests/qoa_oracle.rs b/crates/audio/tests/qoa_oracle.rs new file mode 100644 index 000000000..7b5ac2674 --- /dev/null +++ b/crates/audio/tests/qoa_oracle.rs @@ -0,0 +1,62 @@ +//! QOA against ffmpeg's decoder as the reference (oracle; skipped when ffmpeg is absent, a failure +//! with `CRAFT_MEDIA_REQUIRE_ORACLES=1`): ffmpeg decodes the files our encoder writes to exactly +//! the samples our decoder returns for them, over several sizes (partial slices, partial frames, +//! several frames), channel counts and rates. + +use craft_audio::qoa::{decode, encode}; +use std::process::Command; + +fn signal(channels: usize, frames: usize) -> Vec> { + let mut seed = 0x2545_F491u32; + (0..channels) + .map(|c| { + (0..frames) + .map(|i| { + seed ^= seed << 13; + seed ^= seed >> 17; + seed ^= seed << 5; + let tone = ((i as f64) * (0.02 + 0.013 * c as f64)).sin() * 9000.0; + // Tones, noise bursts, silence and full-scale peaks. + match (i / 700) % 4 { + 0 => tone as i16, + 1 => (tone + f64::from(seed % 6000) - 3000.0) as i16, + 2 => 0, + _ => { + if i % 2 == 0 { + i16::MAX + } else { + i16::MIN + } + } + } + }) + .collect() + }) + .collect() +} + +#[test] +fn ffmpeg_decodes_our_files_to_the_same_samples() { + let Some(ffmpeg) = craft_media_testkit::ffmpeg() else { return }; + let dir = craft_media_testkit::fixture_subdir("qoa"); + for (rate, channels, frames) in + [(44_100u32, 1usize, 1usize), (8_000, 1, 19), (48_000, 2, 5_121), (22_050, 2, 12_345), (96_000, 1, 20_000), (44_100, 4, 6_000)] + { + let src = signal(channels, frames); + let bytes = encode(rate, &src).expect("encode"); + let ours = decode(&bytes).expect("decode"); + let file = dir.join(format!("ours_{rate}_{channels}_{frames}.qoa")); + std::fs::write(&file, &bytes).expect("write fixture"); + let out = Command::new(&ffmpeg) + .args(["-v", "error", "-y", "-i"]) + .arg(&file) + .args(["-f", "s16le", "-acodec", "pcm_s16le", "-"]) + .output() + .expect("run ffmpeg"); + assert!(out.status.success(), "ffmpeg failed on {}: {}", file.display(), String::from_utf8_lossy(&out.stderr)); + let theirs: Vec = out.stdout.as_chunks::<2>().0.iter().map(|b| i16::from_le_bytes(*b)).collect(); + let interleaved: Vec = (0..frames).flat_map(|i| ours.channels.iter().map(move |ch| ch[i])).collect(); + assert_eq!(theirs.len(), interleaved.len(), "{rate} Hz, {channels} ch, {frames} frames: sample count"); + assert!(theirs == interleaved, "{rate} Hz, {channels} ch, {frames} frames: samples differ"); + } +} diff --git a/crates/ber/Cargo.toml b/crates/ber/Cargo.toml new file mode 100644 index 000000000..fa3902e88 --- /dev/null +++ b/crates/ber/Cargo.toml @@ -0,0 +1,11 @@ +[package] +name = "craft-ber" +description = "ASN.1 BER and DER (X.690): reading TLVs, BER to DER, and what an opaque signed CMS holds" +version.workspace = true +edition.workspace = true +license.workspace = true +rust-version.workspace = true +repository.workspace = true + +[lints] +workspace = true diff --git a/crates/ber/src/lib.rs b/crates/ber/src/lib.rs new file mode 100644 index 000000000..e97c1b0f5 --- /dev/null +++ b/crates/ber/src/lib.rs @@ -0,0 +1,432 @@ +// SPDX-FileCopyrightText: MailCraft contributors +// SPDX-License-Identifier: MIT OR Apache-2.0 +//! Reading the TLVs of ASN.1 BER and DER (X.690), and turning BER into DER where they differ in +//! what CMS writers use: indefinite lengths (which streaming signers write) and OCTET STRINGs in +//! pieces (constructed). What is DER already is kept byte for byte, an element with a definite +//! length inside an indefinite one too, so that the bytes a signature covers (signed attributes, +//! a certificate's body) are those the signer wrote. For S/MIME: `craft-smime` checks +//! signatures with it, and a message's reader takes out what an opaque signed one holds +//! ([`signed_content`]). +//! +//! The input is hostile: lengths are checked against what is there, nesting is bounded +//! ([`MAX_DEPTH`]), and the work is linear in the input (times the bounded depth). +#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] + +/// CMS's id-signedData and id-data (RFC 5652 §5.1, §4), as OBJECT IDENTIFIER values. +const SIGNED_DATA: &[u8] = &[0x2a, 0x86, 0x48, 0x86, 0xf7, 0x0d, 0x01, 0x07, 0x02]; +const DATA: &[u8] = &[0x2a, 0x86, 0x48, 0x86, 0xf7, 0x0d, 0x01, 0x07, 0x01]; + +/// What the CMS `cms` holds when it is SignedData with its content inside, of type id-data: the +/// MIME entity an opaque signed message (`application/pkcs7-mime`, RFC 8551 §3.5) signed, for +/// reading as the message. Nothing is checked here (`craft-smime` checks it); `None` for any +/// other CMS (an Encrypted message's enveloped data) or bytes that aren't one. +pub fn signed_content(cms: &[u8]) -> Option> { + let der = to_der(cms)?; + let (info, rest) = read(&der)?; + if info.tag != 0x30 || !rest.is_empty() { + return None; + } + let info = children(info.value)?; + let [oid, explicit] = info.as_slice() else { return None }; + if oid.tag != 0x06 || oid.value != SIGNED_DATA || explicit.tag != 0xa0 { + return None; + } + let (signed, rest) = read(explicit.value)?; + if signed.tag != 0x30 || !rest.is_empty() { + return None; + } + let signed = children(signed.value)?; + let [version, digests, encap, tail @ ..] = signed.as_slice() else { return None }; + if version.tag != 0x02 || !matches!(version.value, [1 | 3 | 4 | 5]) || digests.tag != 0x31 || encap.tag != 0x30 { + return None; + } + if children(digests.value)?.iter().any(|algorithm| algorithm.tag != 0x30) { + return None; + } + // Certificates and revocation lists are optional, but ordered and present at most once. + let tail = if tail.first().is_some_and(|field| field.tag == 0xa0) { tail.get(1..)? } else { tail }; + let tail = if tail.first().is_some_and(|field| field.tag == 0xa1) { tail.get(1..)? } else { tail }; + let [signers] = tail else { return None }; + if signers.tag != 0x31 || children(signers.value)?.iter().any(|signer| signer.tag != 0x30) { + return None; + } + let encap = children(encap.value)?; + let [kind, content] = encap.as_slice() else { return None }; + if kind.tag != 0x06 || kind.value != DATA || content.tag != 0xa0 { + return None; + } + let (octets, rest) = read(content.value)?; + (octets.tag == 0x04 && rest.is_empty()).then(|| octets.value.to_vec()) +} + +/// How deep elements may nest. +pub const MAX_DEPTH: usize = 48; + +/// Maximum elements materialized or indexed in one parsing operation. This bounds the +/// allocation amplification from hostile, shallow input containing many empty TLVs. +pub const MAX_ELEMENTS: usize = 65_536; + +/// The universal tag of an OCTET STRING, and of one in pieces. +const OCTET_STRING: u8 = 0x04; +const OCTET_STRING_PIECES: u8 = 0x24; + +/// One element: its tag bytes, and its value (the content octets). +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct Tlv<'a> { + /// The tag's first byte (class, constructed bit, number below 31). + pub tag: u8, + /// The whole element, tag and length included. + pub whole: &'a [u8], + /// Its content octets. + pub value: &'a [u8], +} + +/// The tag bytes at the start of `bytes`: one, or a high tag number in up to four more. +fn tag_len(bytes: &[u8]) -> Option { + let first = *bytes.first()?; + if first & 0x1f != 0x1f { + return Some(1); + } + // High tag number: base-128 bytes, the last without its top bit. + for (i, b) in bytes.iter().enumerate().skip(1).take(4) { + if b & 0x80 == 0 { + return i.checked_add(1); + } + } + None +} + +/// The length at the start of `bytes`: `(length, bytes it took)`, `None` for an indefinite one +/// (`Some((None, 1))`) is told apart by the caller. +fn length(bytes: &[u8]) -> Option<(Option, usize)> { + let first = *bytes.first()?; + if first < 0x80 { + return Some((Some(usize::from(first)), 1)); + } + if first == 0x80 { + return Some((None, 1)); + } + let n = usize::from(first & 0x7f); + if n > 4 { + return None; + } + let digits = bytes.get(1..n.checked_add(1)?)?; + let len = digits.iter().fold(0usize, |acc, d| (acc << 8) | usize::from(*d)); + Some((Some(len), n.checked_add(1)?)) +} + +/// The element at the start of `bytes` with a definite length, and what follows it. `None` for +/// an indefinite length, or one longer than what is there. +pub fn read(bytes: &[u8]) -> Option<(Tlv<'_>, &[u8])> { + let t = tag_len(bytes)?; + let tag = *bytes.first()?; + let (len, l) = length(bytes.get(t..)?)?; + let len = len?; + let start = t.checked_add(l)?; + let end = start.checked_add(len)?; + let value = bytes.get(start..end)?; + let whole = bytes.get(..end)?; + Some((Tlv { tag, whole, value }, bytes.get(end..)?)) +} + +/// The elements of `value`, each with a definite length; `None` when one can't be read. +pub fn children(mut value: &[u8]) -> Option>> { + let mut out = Vec::new(); + while !value.is_empty() { + if out.len() >= MAX_ELEMENTS { + return None; + } + let (tlv, rest) = read(value)?; + out.push(tlv); + value = rest; + } + Some(out) +} + +/// `bytes` as DER as far as CMS needs it: indefinite lengths made definite and OCTET STRINGs in +/// pieces joined into one, everything else as it was. One element is read, and must be all of +/// `bytes` (bar trailing zero bytes, which some writers pad with). `None` when it isn't BER. +pub fn to_der(bytes: &[u8]) -> Option> { + if let Some((_, rest)) = read(bytes) + && rest.iter().all(|b| *b == 0) + && der_shaped(bytes.get(..bytes.len().saturating_sub(rest.len()))?, 0) + { + return bytes.get(..bytes.len().saturating_sub(rest.len())).map(<[u8]>::to_vec); + } + let mut out = Vec::with_capacity(bytes.len()); + let used = element(bytes, 0, &mut out)?; + let rest = bytes.get(used..)?; + rest.iter().all(|b| *b == 0).then_some(out) +} + +/// Whether the elements of `bytes` all have definite lengths and no OCTET STRING is in pieces, +/// as deep as they go: nothing to change. +fn der_shaped(mut bytes: &[u8], depth: usize) -> bool { + if depth > MAX_DEPTH { + return false; + } + while !bytes.is_empty() { + let Some((tlv, rest)) = read(bytes) else { return false }; + if tlv.tag == OCTET_STRING_PIECES || (tlv.tag & 0x20 != 0 && !der_shaped(tlv.value, depth.saturating_add(1))) { + return false; + } + bytes = rest; + } + true +} + +/// What [`measure`] found of one element: how many bytes of the input it was, and how long its +/// content is as DER. +#[derive(Clone, Copy)] +struct Size { + used: usize, + content: usize, + /// It is DER-shaped already (a definite length, and nothing in it to change): copied as it + /// is, its lengths as the writer wrote them, since a signature may cover them. + verbatim: bool, +} + +/// The sizes of the constructed elements of the input, by where each starts in it: worked out +/// once, so that writing them is one pass with no copy of a level for each level above it. +type Sizes = std::collections::HashMap; + +/// The element at `at` in `input`: how many bytes it was, and its length as DER, header and all. +fn measure(input: &[u8], at: usize, depth: usize, sizes: &mut Sizes) -> Option<(usize, usize)> { + if depth > MAX_DEPTH || sizes.len() >= MAX_ELEMENTS { + return None; + } + let bytes = input.get(at..)?; + if let Some((tlv, _)) = read(bytes) + && der_shaped(tlv.whole, depth) + { + sizes.insert(at, Size { used: tlv.whole.len(), content: tlv.value.len(), verbatim: true }); + return Some((tlv.whole.len(), tlv.whole.len())); + } + let t = tag_len(bytes)?; + let tag = *bytes.first()?; + let (len, l) = length(bytes.get(t..)?)?; + let start = t.checked_add(l)?; + if tag & 0x20 == 0 { + let used = start.checked_add(len?)?; + bytes.get(..used)?; + return Some((used, used_der(t, used.checked_sub(start)?)?)); + } + // The children, one after another, to the end of a definite length or to the end mark. + let mut content = 0usize; + let mut pos = start; + let end = match len { + Some(len) => Some(start.checked_add(len)?), + None => None, + }; + loop { + match end { + Some(end) if pos >= end => break, + None if bytes.get(pos..)?.starts_with(&[0, 0]) => { + pos = pos.checked_add(2)?; + break; + } + _ => {} + } + // (The pieces of an OCTET STRING are OCTET STRINGs, in pieces themselves or not.) + if tag == OCTET_STRING_PIECES && !matches!(bytes.get(pos), Some(&OCTET_STRING | &OCTET_STRING_PIECES)) { + return None; + } + let (n, der) = measure(input, at.checked_add(pos)?, depth.checked_add(1)?, sizes)?; + if n == 0 { + return None; + } + // An OCTET STRING in pieces is one of their contents together. + content = content.checked_add(if tag == OCTET_STRING_PIECES { der_content(input, at.checked_add(pos)?, sizes)? } else { der })?; + pos = pos.checked_add(n)?; + } + if let Some(end) = end + && pos != end + { + return None; + } + if sizes.len() >= MAX_ELEMENTS { + return None; + } + sizes.insert(at, Size { used: pos, content, verbatim: false }); + let header_tag = if tag == OCTET_STRING_PIECES { 1 } else { t }; + Some((pos, used_der(header_tag, content)?)) +} + +/// The length of a DER element with a tag of `tag` bytes and `content` bytes of content. +fn used_der(tag: usize, content: usize) -> Option { + let mut l = Vec::new(); + put_length(&mut l, content); + tag.checked_add(l.len())?.checked_add(content) +} + +/// The DER content length of the (measured) element at `at`: a primitive's own. +fn der_content(input: &[u8], at: usize, sizes: &Sizes) -> Option { + if let Some(s) = sizes.get(&at) { + return Some(s.content); + } + let bytes = input.get(at..)?; + let t = tag_len(bytes)?; + let (len, _) = length(bytes.get(t..)?)?; + len +} + +/// Write the (measured) element at `at` in `input` to `out` as DER; how many bytes it was. +fn write(input: &[u8], at: usize, sizes: &Sizes, pieces: bool, out: &mut Vec) -> Option { + if let Some(s) = sizes.get(&at).filter(|s| s.verbatim && !pieces) { + out.extend_from_slice(input.get(at..at.checked_add(s.used)?)?); + return Some(s.used); + } + let bytes = input.get(at..)?; + let t = tag_len(bytes)?; + let tag_bytes = bytes.get(..t)?; + let tag = *tag_bytes.first()?; + let (len, l) = length(bytes.get(t..)?)?; + let start = t.checked_add(l)?; + if tag & 0x20 == 0 { + let used = start.checked_add(len?)?; + let value = bytes.get(start..used)?; + // (A piece of an OCTET STRING in pieces gives its bytes alone.) + if !pieces { + out.extend_from_slice(tag_bytes); + put_length(out, value.len()); + } + out.extend_from_slice(value); + return Some(used); + } + let size = *sizes.get(&at)?; + if tag == OCTET_STRING_PIECES { + if !pieces { + out.push(OCTET_STRING); + put_length(out, size.content); + } + } else { + out.extend_from_slice(tag_bytes); + put_length(out, size.content); + } + let inner_pieces = tag == OCTET_STRING_PIECES; + let mut pos = start; + while pos < size.used { + if bytes.get(pos..)?.starts_with(&[0, 0]) && len.is_none() { + break; + } + let n = write(input, at.checked_add(pos)?, sizes, inner_pieces, out)?; + pos = pos.checked_add(n)?; + } + Some(size.used) +} + +/// Write the element at the start of `bytes` to `out` as DER; how many bytes of `bytes` it was. +fn element(bytes: &[u8], depth: usize, out: &mut Vec) -> Option { + let mut sizes = Sizes::new(); + let (used, der) = measure(bytes, 0, depth, &mut sizes)?; + out.reserve(der); + write(bytes, 0, &sizes, false, out)?; + Some(used) +} + +/// A definite length in DER's shortest form. +fn put_length(out: &mut Vec, len: usize) { + if len < 0x80 { + out.push(len as u8); + return; + } + let bytes = len.to_be_bytes(); + let skip = bytes.iter().take_while(|b| **b == 0).count(); + let digits = bytes.get(skip..).unwrap_or_default(); + out.push(0x80 | digits.len() as u8); + out.extend_from_slice(digits); +} + +#[cfg(test)] +mod tests { + use super::*; + + fn tlv(tag: u8, value: &[u8]) -> Vec { + let mut out = vec![tag]; + put_length(&mut out, value.len()); + out.extend_from_slice(value); + out + } + + fn cms(outer_tag: u8, signed_tag: u8, oid_tag: u8, long_oid: bool, trailing: bool) -> Vec { + let oid = |value: &[u8]| { + if long_oid { + let mut out = vec![oid_tag, 0x81, 9]; + out.extend_from_slice(value); + out + } else { + tlv(oid_tag, value) + } + }; + let mut encap = oid(DATA); + encap.extend(tlv(0xa0, &tlv(0x04, b"hello"))); + let mut signed = tlv(0x02, &[1]); + signed.extend(tlv(0x31, &[])); + signed.extend(tlv(0x30, &encap)); + signed.extend(tlv(0x31, &[])); + let mut signed = tlv(signed_tag, &signed); + if trailing { + signed.extend(tlv(0x05, &[])); + } + let mut info = oid(SIGNED_DATA); + info.extend(tlv(0xa0, &signed)); + tlv(outer_tag, &info) + } + + #[test] + fn extracts_content_with_der_and_ber_oid_lengths() { + for long_oid in [false, true] { + assert_eq!(signed_content(&cms(0x30, 0x30, 0x06, long_oid, false)), Some(b"hello".to_vec())); + } + } + + #[test] + fn rejects_wrong_cms_container_and_oid_tags() { + for (outer, signed, oid) in [(0x31, 0x30, 0x06), (0x30, 0x31, 0x06), (0x30, 0x30, 0x04)] { + assert_eq!(signed_content(&cms(outer, signed, oid, false, false)), None); + } + } + + #[test] + fn rejects_missing_mandatory_signed_data_fields() { + let mut encap = tlv(0x06, DATA); + encap.extend(tlv(0xa0, &tlv(0x04, b"hello"))); + for prefix in [vec![0x05, 0, 0x05, 0], vec![0x02, 1, 1, 0x31, 0]] { + let mut malformed = prefix; + malformed.extend(tlv(0x30, &encap)); + let mut info = tlv(0x06, SIGNED_DATA); + info.extend(tlv(0xa0, &tlv(0x30, &malformed))); + assert_eq!(signed_content(&tlv(0x30, &info)), None); + } + } + + #[test] + fn rejects_extra_signed_data_inside_explicit_container() { + assert_eq!(signed_content(&cms(0x30, 0x30, 0x06, false, true)), None); + } + + #[test] + fn rejects_wide_inputs_before_unbounded_metadata_allocation() { + let children_bytes = [0x05, 0].repeat(MAX_ELEMENTS + 1); + assert!(children(&children_bytes).is_none()); + let mut indefinite = vec![0x30, 0x80]; + indefinite.extend_from_slice(&children_bytes); + indefinite.extend_from_slice(&[0, 0]); + assert!(to_der(&indefinite).is_none()); + } + + #[test] + fn joins_indefinite_octet_string_pieces() { + assert_eq!(to_der(&[0x24, 0x80, 0x04, 1, b'a', 0x04, 1, b'b', 0, 0]), Some(vec![0x04, 2, b'a', b'b'])); + } + + #[test] + fn rejects_truncated_and_over_nested_input() { + assert_eq!(to_der(&[0x04, 2, b'a']), None); + let mut nested = tlv(0x04, &[]); + for _ in 0..=MAX_DEPTH { + nested = tlv(0x30, &nested); + } + assert_eq!(to_der(&nested), None); + } +} diff --git a/crates/captions/Cargo.toml b/crates/captions/Cargo.toml index df665336d..cd09f7e16 100644 --- a/crates/captions/Cargo.toml +++ b/crates/captions/Cargo.toml @@ -19,3 +19,5 @@ thiserror = { workspace = true } proptest = { workspace = true } # A second XML reader for the tests: checks of written TTML, and the comparison in tests/it/ttml_xml_views.rs. roxmltree = "0.21" +# ffmpeg as the reference for the ASS tests (tests/it/ass.rs). +craft-media-testkit = { workspace = true } diff --git a/crates/captions/README.md b/crates/captions/README.md index 0676d1d72..7380a3ce4 100644 --- a/crates/captions/README.md +++ b/crates/captions/README.md @@ -55,6 +55,7 @@ lists at most 20 characters per stream. | EBU STL `.stl` | yes | yes | EBU Tech 3264-E: GSI block + 128-byte TTI blocks, `STL25.01` (25 fps) or `STL30.01` (read/written as 29.97 fps, non-drop labels), ISO/IEC 6937 Latin text (accents as diacritic + letter), italics / underline codes as `` / ``, extension blocks for long subtitles, start-of-programme (TCP) subtracted. | | TTML `.ttml` (IMSC1) | yes | yes | Writes the IMSC1 Text profile (TTML2 namespaces) with the document's language as `xml:lang` (read back on import): `

` per cue, `
`, styled ``s for `` / `` / ``, `ttm:agent` speakers, exact frame times (`123f`) at the sequence rate. Reads any TTML (TTML1, TTML2, the 2006 `ttaf1` drafts): clock / frame / offset / tick times, `dur`, nested time containers (`body` / `div` begins offset and ends clip their contents; `seq` containers are read as `par` with a warning), timed spans, referenced styles. | | DFXP `.dfxp` | yes | yes | TTML1 with the DFXP presentation profile; clock times in milliseconds (frame-exact after `snap_to_frames`). | +| ASS / SSA `.ass`, `.ssa` | yes | yes (ASS v4+) | The `[Events]` section: `Dialogue:` lines through the file's `Format:` line, `Name` as the speaker, `\N`, `\n`, `\h` and the `\i`, `\b`, `\u` override tags kept (as ``, ``, ``); other override tags and drawing lines are dropped. Times are centiseconds. Written files have one `Default` style. | All cue times are exact `Tick`s. Milliseconds convert exactly (1 ms = 254 016 000 ticks); SCC frames map through `FrameRate::FPS_29_97`. `Document::snap_to_frames` snaps cues to a frame grid and diff --git a/crates/captions/src/ass.rs b/crates/captions/src/ass.rs new file mode 100644 index 000000000..2d0a1df55 --- /dev/null +++ b/crates/captions/src/ass.rs @@ -0,0 +1,244 @@ +//! Advanced SubStation Alpha and SubStation Alpha (`.ass`, `.ssa`). +//! +//! Reading takes the `[Events]` section: its `Format:` line gives the field order (ASS v4+ and SSA +//! v4 differ in the first field only), each `Dialogue:` line becomes a cue; `Comment:` lines and +//! the other sections are skipped. A cue gets the `Name` field as its speaker. The text keeps what +//! the neutral model can show: `\N` and `\n` become a line break and a space, `\h` a no-break +//! space, and the override tags `\i`, `\b` and `\u` become ``, `` and ``. Every other +//! override tag (colours, fonts, positions, karaoke, animation) is dropped, and a line that draws +//! vector shapes (`\p1`) is skipped with a warning. +//! +//! Writing gives a minimal ASS v4+ file: one `Default` style, one `Dialogue:` line per cue with +//! times to the centisecond. Braces in the text, which ASS reads as override blocks, are written +//! as the full-width `ï½›` and `ï½�`. + +use crate::time::{TICKS_PER_SECOND, Tick}; +use crate::{Cue, Document, Error, Result}; + +const TICKS_PER_CS: i64 = TICKS_PER_SECOND / 100; + +/// `H:MM:SS.cc`, the form SSA and ASS use (any number of fraction digits is read). +fn parse_time(s: &str) -> Option { + let s = s.trim(); + let (hms, frac) = s.split_once('.').unwrap_or((s, "")); + let mut parts = hms.split(':'); + let (h, m, sec) = (parts.next()?, parts.next()?, parts.next()?); + if parts.next().is_some() { + return None; + } + let num = |p: &str| -> Option { (!p.is_empty() && p.len() <= 9 && p.bytes().all(|b| b.is_ascii_digit())).then(|| p.parse().ok()).flatten() }; + let (h, m, sec) = (num(h)?, num(m)?, num(sec)?); + if m >= 60 || sec >= 60 || h > 9999 { + return None; + } + let mut ticks = ((h * 60 + m) * 60 + sec) * TICKS_PER_SECOND; + if !frac.is_empty() { + if frac.len() > 9 || !frac.bytes().all(|b| b.is_ascii_digit()) { + return None; + } + let d: i64 = frac.parse().ok()?; + ticks += (i128::from(d) * i128::from(TICKS_PER_SECOND) / 10i128.pow(frac.len() as u32)) as i64; + } + Some(Tick(ticks)) +} + +fn format_time(t: Tick) -> String { + let cs = (i128::from(t.0.max(0)) + i128::from(TICKS_PER_CS) / 2) / i128::from(TICKS_PER_CS); + let (h, m, s, c) = (cs / 360_000, (cs / 6000) % 60, (cs / 100) % 60, cs % 100); + format!("{h}:{m:02}:{s:02}.{c:02}") +} + +/// The Text field of a Dialogue line as cue text; `None` when the line draws shapes. +fn convert_text(raw: &str) -> Option { + let mut out = String::new(); + // Open ``, ``, `` in the order they were opened. + let mut open: Vec = Vec::new(); + let mut chars = raw.chars().peekable(); + let mut drawing = false; + let mut dropped_drawing = false; + while let Some(c) = chars.next() { + match c { + '{' => { + let mut block = String::new(); + let mut closed = false; + for b in chars.by_ref() { + if b == '}' { + closed = true; + break; + } + block.push(b); + } + if !closed { + // An unterminated block is literal text. + out.push('{'); + out.push_str(&block); + continue; + } + for tag in block.split('\\').skip(1) { + let tag = tag.trim_end(); + let (name, val) = tag.split_at(tag.find(|c: char| !c.is_ascii_alphabetic()).unwrap_or(tag.len())); + match name { + "i" | "b" | "u" => { + let letter = name.chars().next().unwrap_or('i'); + let on = val.trim() != "0"; + if on && !open.contains(&letter) { + open.push(letter); + out.push_str(&format!("<{letter}>")); + } else if !on && open.contains(&letter) { + // Close back to it, reopening what was opened after. + let at = open.iter().position(|&o| o == letter).unwrap_or(0); + let after: Vec = open.split_off(at + 1); + for o in after.iter().rev() { + out.push_str(&format!("")); + } + open.pop(); + out.push_str(&format!("")); + for o in after { + open.push(o); + out.push_str(&format!("<{o}>")); + } + } + } + "p" => drawing = val.trim().parse::().unwrap_or(0) != 0, + _ => {} + } + } + } + '\\' => match chars.peek() { + Some('N') => { + chars.next(); + out.push('\n'); + } + Some('n') => { + chars.next(); + out.push(' '); + } + Some('h') => { + chars.next(); + out.push('\u{a0}'); + } + _ => out.push('\\'), + }, + c => { + if drawing { + dropped_drawing = true; + } else { + out.push(c); + } + } + } + } + if dropped_drawing && out.trim().is_empty() { + return None; + } + while let Some(o) = open.pop() { + out.push_str(&format!("")); + } + Some(out) +} + +pub fn parse(text: &str) -> Result { + let text = text.strip_prefix('\u{feff}').unwrap_or(text); + let mut doc = Document::default(); + let mut in_events = false; + let mut saw_events = false; + // Column of each field in a Dialogue line. + let mut cols: Vec = + ["layer", "start", "end", "style", "name", "marginl", "marginr", "marginv", "effect", "text"].map(String::from).to_vec(); + let mut drawings = 0usize; + for (n, line) in text.lines().enumerate() { + let line = line.trim(); + if line.starts_with('[') && line.ends_with(']') { + in_events = line.eq_ignore_ascii_case("[events]"); + saw_events |= in_events; + continue; + } + if !in_events { + continue; + } + let Some((key, rest)) = line.split_once(':') else { continue }; + let rest = rest.trim_start(); + match key.trim().to_ascii_lowercase().as_str() { + "format" => cols = rest.split(',').map(|f| f.trim().to_ascii_lowercase()).collect(), + "dialogue" => { + // The last column (Text) keeps its commas. + let fields: Vec<&str> = rest.splitn(cols.len(), ',').collect(); + let get = |name: &str| cols.iter().position(|c| c == name).and_then(|i| fields.get(i)).copied(); + let (Some(start), Some(end), Some(raw)) = (get("start").and_then(parse_time), get("end").and_then(parse_time), get("text")) else { + doc.warnings.push(format!("line {}: not a Dialogue line this reader understands", n + 1)); + continue; + }; + let Some(text) = convert_text(raw) else { + drawings += 1; + continue; + }; + let name = get("name").map(str::trim).filter(|s| !s.is_empty()); + doc.cues.push(Cue { start, end, text, speaker: name.map(str::to_string), ..Default::default() }); + } + _ => {} + } + } + if !saw_events { + return Err(Error::NotFormat("ASS")); + } + if drawings > 0 { + doc.warnings.push(format!("{drawings} vector drawing line(s) were skipped")); + } + Ok(doc) +} + +fn escape(text: &str) -> String { + let mut out = String::with_capacity(text.len() + 8); + let mut rest = text; + while !rest.is_empty() { + if let Some(r) = rest.strip_prefix("\r\n") { + out.push_str("\\N"); + rest = r; + } else if let Some(r) = rest.strip_prefix("") { + out.push_str("{\\i1}"); + rest = r; + } else if let Some(r) = rest.strip_prefix("") { + out.push_str("{\\i0}"); + rest = r; + } else if let Some(r) = rest.strip_prefix("") { + out.push_str("{\\b1}"); + rest = r; + } else if let Some(r) = rest.strip_prefix("") { + out.push_str("{\\b0}"); + rest = r; + } else if let Some(r) = rest.strip_prefix("") { + out.push_str("{\\u1}"); + rest = r; + } else if let Some(r) = rest.strip_prefix("") { + out.push_str("{\\u0}"); + rest = r; + } else if let Some(c) = rest.chars().next() { + match c { + '\n' | '\r' => out.push_str("\\N"), + // Empty override block separates a literal slash from N/n/h. + '\\' => out.push_str("\\{}"), + '{' => out.push('\u{ff5b}'), + '}' => out.push('\u{ff5d}'), + '\u{a0}' => out.push_str("\\h"), + c => out.push(c), + } + rest = &rest[c.len_utf8()..]; + } + } + out +} + +pub fn write(doc: &Document) -> String { + let mut out = String::from( + "[Script Info]\n; Written by craft-captions\nScriptType: v4.00+\nWrapStyle: 0\nScaledBorderAndShadow: yes\nPlayResX: 384\nPlayResY: 288\n\n\ + [V4+ Styles]\nFormat: Name, Fontname, Fontsize, PrimaryColour, SecondaryColour, OutlineColour, BackColour, Bold, Italic, Underline, StrikeOut, ScaleX, ScaleY, Spacing, Angle, BorderStyle, Outline, Shadow, Alignment, MarginL, MarginR, MarginV, Encoding\n\ + Style: Default,Arial,16,&H00FFFFFF,&H000000FF,&H00000000,&H00000000,0,0,0,0,100,100,0,0,1,1,0,2,10,10,10,1\n\n\ + [Events]\nFormat: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text\n", + ); + for c in &doc.cues { + // A comma in the Name field would shift the columns; the other fields are fixed text. + let name = c.speaker.as_deref().unwrap_or("").replace([',', '\n', '\r'], " "); + out.push_str(&format!("Dialogue: 0,{},{},Default,{},0,0,0,,{}\n", format_time(c.start), format_time(c.end), name, escape(&c.text))); + } + out +} diff --git a/crates/captions/src/lib.rs b/crates/captions/src/lib.rs index 0ba9f2ba5..e8a492722 100644 --- a/crates/captions/src/lib.rs +++ b/crates/captions/src/lib.rs @@ -4,7 +4,7 @@ //! - **Files:** [SubRip](srt) (`.srt`), [WebVTT](vtt) (`.vtt`), [Scenarist SCC](scc) (`.scc`, //! CEA-608 pop-on, roll-up and paint-on on read; pop-on on write), [MacCaption MCC](mcc) //! (`.mcc`, SMPTE ST 334-2 packets with CEA-608 and CEA-708), [EBU STL](stl) (`.stl`, EBU -//! Tech 3264) and [TTML](ttml) (`.ttml` IMSC1 Text profile, `.dfxp` DFXP) parse into a +//! Tech 3264), [ASS/SSA](ass) (`.ass`, `.ssa`) and [TTML](ttml) (`.ttml` IMSC1 Text profile, `.dfxp` DFXP) parse into a //! [`Document`] of [`Cue`]s and write back. Readers are forgiving (BOM, UTF-16, Windows-1252 //! fallback, CRLF/CR, missing SRT indexes, `,` or `.` milliseconds, missing hours, no blank //! line between cues); writers are strict. WebVTT cue identifiers, cue settings, `` @@ -32,6 +32,7 @@ #[doc = include_str!("../README.md")] struct ReadmeExamples; +pub mod ass; pub mod cea608; pub mod mcc; pub mod scc; @@ -69,10 +70,12 @@ pub enum Format { Ttml, /// DFXP (TTML1). Dfxp, + /// Advanced SubStation Alpha (also reads SubStation Alpha). + Ass, } impl Format { - pub const ALL: [Format; 7] = [Format::Srt, Format::WebVtt, Format::Scc, Format::Mcc, Format::Stl, Format::Ttml, Format::Dfxp]; + pub const ALL: [Format; 8] = [Format::Srt, Format::WebVtt, Format::Scc, Format::Mcc, Format::Stl, Format::Ttml, Format::Dfxp, Format::Ass]; pub fn name(self) -> &'static str { match self { Format::Srt => "SubRip (SRT)", @@ -82,6 +85,7 @@ impl Format { Format::Stl => "EBU STL", Format::Ttml => "TTML (IMSC1)", Format::Dfxp => "DFXP (TTML1)", + Format::Ass => "Advanced SubStation Alpha (ASS)", } } pub fn extension(self) -> &'static str { @@ -93,6 +97,7 @@ impl Format { Format::Stl => "stl", Format::Ttml => "ttml", Format::Dfxp => "dfxp", + Format::Ass => "ass", } } pub fn from_name(s: &str) -> Option { @@ -104,6 +109,7 @@ impl Format { "stl" | "ebu-stl" | "ebustl" | "ebu" => Some(Format::Stl), "ttml" | "imsc" | "imsc1" | "xml" => Some(Format::Ttml), "dfxp" => Some(Format::Dfxp), + "ass" | "ssa" => Some(Format::Ass), _ => None, } } @@ -251,6 +257,9 @@ pub fn detect(bytes: &[u8], ext: Option<&str>) -> Option { if head.starts_with("File Format=MacCaption_MCC") { return Some(Format::Mcc); } + if head.starts_with("[Script Info]") { + return Some(Format::Ass); + } let ext = ext.map(|e| e.trim_start_matches('.').to_ascii_lowercase()); if (head.starts_with('<') && (head.contains("") || head.contains(":tt "))) && (head.contains("ns/ttml") || head.contains("ttaf1")) @@ -262,7 +271,7 @@ pub fn detect(bytes: &[u8], ext: Option<&str>) -> Option { match ext.as_deref() { Some("srt") if has_arrow_timing || head.is_empty() => Some(Format::Srt), Some("vtt") if has_arrow_timing => Some(Format::WebVtt), - Some("srt" | "vtt" | "scc" | "mcc" | "stl" | "ttml" | "dfxp" | "xml") => None, + Some("srt" | "vtt" | "scc" | "mcc" | "stl" | "ttml" | "dfxp" | "xml" | "ass" | "ssa") => None, _ if has_arrow_timing => Some(Format::Srt), _ => None, } @@ -296,6 +305,7 @@ fn parse_unbounded(bytes: &[u8], format: Format) -> Result { Format::Scc => scc::parse(&text), Format::Mcc => mcc::parse(&text), Format::Ttml | Format::Dfxp => ttml::parse(&text), + Format::Ass => ass::parse(&text), Format::Stl => stl::parse(bytes), } } @@ -333,6 +343,7 @@ pub fn write(doc: &Document, format: Format, opts: WriteOptions) -> Vec { Format::Stl => stl::write(doc, opts.rate, opts.program), Format::Ttml => ttml::write(doc, ttml::Flavor::Imsc1, opts.rate, doc.language.as_deref().unwrap_or("")).into_bytes(), Format::Dfxp => ttml::write(doc, ttml::Flavor::Dfxp, opts.rate, doc.language.as_deref().unwrap_or("")).into_bytes(), + Format::Ass => ass::write(doc).into_bytes(), } } diff --git a/crates/captions/tests/it/ass.rs b/crates/captions/tests/it/ass.rs new file mode 100644 index 000000000..fb6d4e01e --- /dev/null +++ b/crates/captions/tests/it/ass.rs @@ -0,0 +1,142 @@ +//! SubStation Alpha / Advanced SubStation Alpha: reading, writing, and ffmpeg as the reference +//! (oracle; skipped when ffmpeg is absent, a failure with `CRAFT_MEDIA_REQUIRE_ORACLES=1`). + +use craft_captions::{Cue, Document, Format, Tick, WriteOptions, detect, parse, write}; +use std::process::Command; + +const T: i64 = craft_captions::TICKS_PER_SECOND; + +fn cs(n: i64) -> Tick { + Tick(n * (T / 100)) +} + +const SAMPLE: &str = "\u{feff}[Script Info]\n; a comment\nTitle: Test\nScriptType: v4.00+\n\n[V4+ Styles]\nFormat: Name, Fontname\nStyle: Default,Arial\n\n[Events]\nFormat: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text\nComment: 0,0:00:00.00,0:00:01.00,Default,,0,0,0,,not a cue\nDialogue: 0,0:00:01.50,0:00:03.25,Default,Bob,0,0,0,,Hello, {\\i1}world{\\i0}\\Nsecond {\\b1}line{\\b0}\nDialogue: 0,1:02:03.04,1:02:04.5,Default,,0,0,0,,{\\pos(10,20)\\c&H0000FF&}Red\\hspace\\nsoft\nDialogue: 0,0:00:09.00,0:00:10.00,Default,,0,0,0,,{\\p1}m 0 0 l 100 0 100 100{\\p0}\n"; + +#[test] +fn reads_dialogue_lines_with_commas_overrides_and_line_breaks() { + let doc = parse(SAMPLE.as_bytes(), Format::Ass).unwrap(); + assert_eq!(doc.cues.len(), 2); + assert_eq!(doc.cues[0].start, cs(150)); + assert_eq!(doc.cues[0].end, cs(325)); + assert_eq!(doc.cues[0].text, "Hello, world\nsecond line"); + assert_eq!(doc.cues[0].speaker.as_deref(), Some("Bob")); + // 1:02:03.04, a one-digit fraction, an override block dropped, \h and \n. + assert_eq!(doc.cues[1].start, Tick((3600 + 120 + 3) * T + 4 * (T / 100))); + assert_eq!(doc.cues[1].end, Tick((3600 + 120 + 4) * T + T / 2)); + assert_eq!(doc.cues[1].text, "Red\u{a0}space soft"); + assert!(doc.warnings.iter().any(|w| w.contains("vector drawing")), "{:?}", doc.warnings); +} + +#[test] +fn ssa_v4_files_read_through_their_own_format_line() { + let ssa = "[Script Info]\nScriptType: v4.00\n\n[V4 Styles]\nFormat: Name\n\n[Events]\nFormat: Marked, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text\nDialogue: Marked=0,0:00:01.00,0:00:02.00,*Default,,0000,0000,0000,,Old style\n"; + let doc = parse(ssa.as_bytes(), Format::Ass).unwrap(); + assert_eq!((doc.cues[0].start, doc.cues[0].end, doc.cues[0].text.as_str()), (cs(100), cs(200), "Old style")); +} + +#[test] +fn write_then_read_keeps_times_text_and_speaker() { + let cues = vec![ + Cue { start: cs(100), end: cs(250), text: "Line one\nline two".into(), speaker: Some("Ann".into()), ..Default::default() }, + Cue { + start: Tick(7200 * T + 99 * (T / 100)), + end: Tick(7203 * T), + text: "bold and under {braces}".into(), + ..Default::default() + }, + ]; + let doc = Document { cues, ..Default::default() }; + let bytes = write(&doc, Format::Ass, WriteOptions::default()); + assert_eq!(detect(&bytes, Some("ass")), Some(Format::Ass)); + let back = parse(&bytes, Format::Ass).unwrap(); + assert_eq!(back.cues[0], doc.cues[0]); + assert_eq!(back.cues[1].start, doc.cues[1].start); + // Braces are written full-width, as ASS reads `{` as the start of an override block. + assert_eq!(back.cues[1].text, "bold and under \u{ff5b}braces\u{ff5d}"); +} + +#[test] +fn files_without_an_events_section_are_not_ass() { + assert!(parse(b"[Script Info]\nTitle: x\n", Format::Ass).is_err()); + assert!(parse(b"hello", Format::Ass).is_err()); + assert_eq!(detect(b"hello", Some("ass")), None); +} + +#[test] +fn bad_lines_are_skipped_with_a_warning() { + let text = "[Events]\nFormat: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text\nDialogue: 0,garbage,0:00:02.00,Default,,0,0,0,,x\nDialogue: 0,0:00:01.00,0:00:02.00,Default,,0,0,0,,ok\n"; + let doc = parse(text.as_bytes(), Format::Ass).unwrap(); + assert_eq!(doc.cues.len(), 1); + assert_eq!(doc.warnings.len(), 1); +} + +fn ffmpeg_run(ff: &std::path::Path, args: &[&str]) { + let out = Command::new(ff).args(["-v", "error", "-nostdin", "-y"]).args(args).output().expect("run ffmpeg"); + assert!(out.status.success(), "ffmpeg {args:?}: {}", String::from_utf8_lossy(&out.stderr)); +} + +const SRT: &str = "1\n00:00:01,500 --> 00:00:03,250\nHello, world\nsecond line\n\n2\n00:00:04,000 --> 00:00:05,990\nPlain, with a comma\n\n3\n01:00:00,010 --> 01:00:02,999\nUnder and italic both\n"; + +#[test] +fn we_read_what_ffmpeg_writes() { + let Some(ff) = craft_media_testkit::ffmpeg() else { return }; + let dir = craft_media_testkit::fixture_subdir("captions_ass"); + let (srt, ass) = (dir.join("in.srt"), dir.join("ffmpeg.ass")); + std::fs::write(&srt, SRT).expect("write"); + ffmpeg_run(&ff, &["-i", srt.to_str().expect("path"), ass.to_str().expect("path")]); + let from_ass = parse(&std::fs::read(&ass).expect("read"), Format::Ass).expect("parse ass"); + let from_srt = parse(SRT.as_bytes(), Format::Srt).expect("parse srt"); + assert_eq!(from_ass.cues.len(), from_srt.cues.len()); + for (a, s) in from_ass.cues.iter().zip(&from_srt.cues) { + // ASS times are centiseconds; the SRT ones round to them. + assert_eq!(a.start, cs((s.start.0 + T / 200) / (T / 100)), "start of {:?}", s.text); + assert_eq!(a.end, cs((s.end.0 + T / 200) / (T / 100)), "end of {:?}", s.text); + assert_eq!(a.text, s.text); + } +} + +#[test] +fn ffmpeg_reads_what_we_write() { + let Some(ff) = craft_media_testkit::ffmpeg() else { return }; + let dir = craft_media_testkit::fixture_subdir("captions_ass"); + let mut doc = parse(SRT.as_bytes(), Format::Srt).expect("parse srt"); + doc.cues.push(Cue { start: cs(400_000), end: cs(400_100), text: r"C:\New\notes\hello and \N \n \h".into(), ..Default::default() }); + let (ass, srt) = (dir.join("ours.ass"), dir.join("back.srt")); + std::fs::write(&ass, write(&doc, Format::Ass, WriteOptions::default())).expect("write"); + ffmpeg_run(&ff, &["-i", ass.to_str().expect("path"), srt.to_str().expect("path")]); + let theirs = parse(&std::fs::read(&srt).expect("read"), Format::Srt).expect("parse back"); + assert_eq!(theirs.cues.len(), doc.cues.len()); + for (t, d) in theirs.cues.iter().zip(&doc.cues) { + assert_eq!(t.start, cs((d.start.0 + T / 200) / (T / 100))); + assert_eq!(t.end, cs((d.end.0 + T / 200) / (T / 100))); + assert_eq!(t.text, d.text, "text of the cue at {:?}", d.start); + } +} + +#[test] +fn literal_backslashes_survive_ass_round_trip() { + let text = r"C:\New\notes\hello and \N \n \h"; + let doc = Document { cues: vec![Cue { start: cs(100), end: cs(200), text: text.into(), ..Default::default() }], ..Default::default() }; + let bytes = write(&doc, Format::Ass, WriteOptions::default()); + assert_eq!(parse(&bytes, Format::Ass).unwrap().cues[0].text, text); +} + +#[test] +fn carriage_returns_cannot_create_dialogue_records() { + let doc = Document { + cues: vec![Cue { + start: cs(100), + end: cs(200), + text: "first\rsecond\r\nthird".into(), + speaker: Some("Ann\rOther".into()), + ..Default::default() + }], + ..Default::default() + }; + let bytes = write(&doc, Format::Ass, WriteOptions::default()); + assert!(!bytes.contains(&b'\r')); + let back = parse(&bytes, Format::Ass).unwrap(); + assert_eq!(back.cues.len(), 1); + assert_eq!(back.cues[0].text, "first\nsecond\nthird"); + assert_eq!(back.cues[0].speaker.as_deref(), Some("Ann Other")); +} diff --git a/crates/captions/tests/it/main.rs b/crates/captions/tests/it/main.rs index 16f8fcbb6..be03d2855 100644 --- a/crates/captions/tests/it/main.rs +++ b/crates/captions/tests/it/main.rs @@ -1,3 +1,4 @@ +mod ass; mod limits; mod literal_roundtrips; mod mcc_708_timing; diff --git a/crates/cms/src/builtin.rs b/crates/cms/src/builtin.rs index ca7034286..5a8b03228 100644 --- a/crates/cms/src/builtin.rs +++ b/crates/cms/src/builtin.rs @@ -424,6 +424,18 @@ fn shipped_cmyk(bytes: &[u8], b: Builtin) -> Profile { }) } +/// Generates the compact ICC v2.1 sRGB matrix/TRC profile used by PDFCraft. +/// +/// Unlike [`Builtin::Srgb`]'s v4 profile, this uses a 1024-sample tone curve and +/// v2 `desc`/`text` tags for consumers that require a v2 output intent. The legacy +/// description `sRGB (PdfCraft)` is retained to preserve existing document bytes. +/// The generated profile contains no external assets; callers choose where it is +/// appropriate (including their own PDF/A conformance policy). +/// All sizes are fixed; no input-controlled allocation or recursion is involved. +pub fn srgb_v2() -> Vec { + crate::builtin_v2::srgb() +} + #[cfg(test)] mod tests { use super::*; diff --git a/crates/cms/src/builtin_v2.rs b/crates/cms/src/builtin_v2.rs new file mode 100644 index 000000000..4a9e2e25b --- /dev/null +++ b/crates/cms/src/builtin_v2.rs @@ -0,0 +1,143 @@ +//! A small sRGB ICC profile (version 2, display class, matrix/TRC), written from the published +//! sRGB primaries (IEC 61966-2-1) adapted to D50, with the sRGB tone curve sampled at 1024 +//! points. Used as the PDF/A output intent's destination profile. + +fn s15f16(v: f64) -> [u8; 4] { + ((v * 65536.0).round() as i32).to_be_bytes() +} + +fn xyz(x: f64, y: f64, z: f64) -> Vec { + let mut t = b"XYZ \0\0\0\0".to_vec(); + t.extend(s15f16(x)); + t.extend(s15f16(y)); + t.extend(s15f16(z)); + t +} + +fn desc(text: &str) -> Vec { + let mut t = b"desc\0\0\0\0".to_vec(); + let ascii: Vec = text.bytes().chain(std::iter::once(0)).collect(); + t.extend((ascii.len() as u32).to_be_bytes()); + t.extend(&ascii); + // No Unicode or ScriptCode descriptions. + t.extend([0u8; 4 + 4]); + t.extend([0u8; 2 + 1 + 67]); + t +} + +fn text(text: &str) -> Vec { + let mut t = b"text\0\0\0\0".to_vec(); + t.extend(text.bytes()); + t.push(0); + t +} + +fn curve() -> Vec { + const N: usize = 1024; + let mut t = b"curv\0\0\0\0".to_vec(); + t.extend((N as u32).to_be_bytes()); + for i in 0..N { + let v = i as f64 / (N - 1) as f64; + let lin = if v <= 0.04045 { v / 12.92 } else { ((v + 0.055) / 1.055).powf(2.4) }; + t.extend(((lin * 65535.0).round() as u16).to_be_bytes()); + } + t +} + +/// Fixed-size profile bytes; all tag sizes and loop counts are constants. +pub fn srgb() -> Vec { + // sRGB primaries, Bradford-adapted to the D50 profile connection space. + let tags: Vec<(&[u8; 4], Vec)> = vec![ + (b"desc", desc("sRGB (PdfCraft)")), + (b"cprt", text("No copyright, use freely")), + (b"wtpt", xyz(0.9642, 1.0, 0.8249)), + (b"rXYZ", xyz(0.4361, 0.2225, 0.0139)), + (b"gXYZ", xyz(0.3851, 0.7169, 0.0971)), + (b"bXYZ", xyz(0.1431, 0.0606, 0.7141)), + (b"rTRC", curve()), + ]; + // gTRC and bTRC share rTRC's data. + let count = tags.len() + 2; + let table_len = 4 + count * 12; + let mut offset = 128 + table_len; + let mut table = (count as u32).to_be_bytes().to_vec(); + let mut data = Vec::new(); + let mut trc = (0, 0); + for (sig, body) in &tags { + let len = body.len(); + table.extend(*sig); + table.extend((offset as u32).to_be_bytes()); + table.extend((len as u32).to_be_bytes()); + if *sig == b"rTRC" { + trc = (offset, len); + } + data.extend(body); + // Tags start on 4-byte boundaries. + let pad = (4 - len % 4) % 4; + data.extend(std::iter::repeat_n(0u8, pad)); + offset += len + pad; + } + for sig in [b"gTRC", b"bTRC"] { + table.extend(sig); + table.extend((trc.0 as u32).to_be_bytes()); + table.extend((trc.1 as u32).to_be_bytes()); + } + let size = 128 + table.len() + data.len(); + let mut h = Vec::with_capacity(size); + h.extend((size as u32).to_be_bytes()); + h.extend(b"\0\0\0\0"); // preferred CMM + h.extend([2, 0x10, 0, 0]); // version 2.1 + h.extend(b"mntr"); + h.extend(b"RGB "); + h.extend(b"XYZ "); + h.extend([0u8; 12]); // date + h.extend(b"acsp"); + h.extend(b"\0\0\0\0"); // platform + h.extend([0u8; 4]); // flags + h.extend([0u8; 4]); // manufacturer + h.extend([0u8; 4]); // model + h.extend([0u8; 8]); // attributes + h.extend([0, 0, 0, 0]); // perceptual intent + h.extend(s15f16(0.9642)); + h.extend(s15f16(1.0)); + h.extend(s15f16(0.8249)); + h.extend([0u8; 4]); // creator + h.resize(128, 0); + h.extend(table); + h.extend(data); + h +} + +#[cfg(test)] +mod tests { + #[test] + fn parses_as_v2_with_shared_rgb_curves() { + let bytes = super::srgb(); + let p = crate::Profile::parse(&bytes).unwrap(); + assert_eq!(p.version, (2, 0x10)); + assert_eq!(p.color_space, crate::ColorSpace::Rgb); + assert_eq!(p.pcs, crate::Pcs::Xyz); + assert_eq!(p.description, "sRGB (PdfCraft)"); + let curves = p.trc.unwrap(); + assert_eq!(curves[0], curves[1]); + assert_eq!(curves[1], curves[2]); + assert!(matches!(&curves[0], crate::Curve::Table(v) if v.len() == 1024 && v[0] == 0.0 && v[1023] == 1.0)); + assert_eq!(bytes, super::srgb()); + } + + #[test] + fn profile_is_well_formed() { + let p = super::srgb(); + assert_eq!(u32::from_be_bytes(p[0..4].try_into().unwrap()) as usize, p.len()); + assert_eq!(&p[36..40], b"acsp"); + assert_eq!(&p[12..20], b"mntrRGB "); + let n = u32::from_be_bytes(p[128..132].try_into().unwrap()) as usize; + assert_eq!(n, 9); + for i in 0..n { + let e = 132 + i * 12; + let off = u32::from_be_bytes(p[e + 4..e + 8].try_into().unwrap()) as usize; + let len = u32::from_be_bytes(p[e + 8..e + 12].try_into().unwrap()) as usize; + assert!(off.is_multiple_of(4) && off + len <= p.len(), "tag {i}"); + } + } +} diff --git a/crates/cms/src/lib.rs b/crates/cms/src/lib.rs index 8d4f99f49..2f4abdc50 100644 --- a/crates/cms/src/lib.rs +++ b/crates/cms/src/lib.rs @@ -20,6 +20,7 @@ #![forbid(unsafe_code)] pub mod builtin; +mod builtin_v2; pub mod clut; pub mod curve; pub mod gamut; @@ -28,11 +29,15 @@ pub mod lut3d; pub mod lutfile; pub mod math; pub mod named; +mod normalize; pub mod pipeline; pub mod profile; +pub mod sony_lut; +pub mod spi1d; pub mod synth; pub mod transform; mod write; +pub mod write_v2; pub use builtin::Builtin; pub use clut::Clut; @@ -41,6 +46,7 @@ pub use gamut::GamutCheck; pub use gpu_matrix_shaper::GpuMatrixShaper; pub use lut3d::Lut3d; pub use named::{NamedColor, NamedColors}; +pub use normalize::normalize_for_embedding; pub use profile::{ColorSpace, Pcs, Profile, ProfileClass}; pub use transform::{SampleKind, Transform, TransformOptions, black_point, cached}; pub use write::profile_id; diff --git a/crates/cms/src/lutfile.rs b/crates/cms/src/lutfile.rs index 404b5a315..e2e0a5980 100644 --- a/crates/cms/src/lutfile.rs +++ b/crates/cms/src/lutfile.rs @@ -150,62 +150,142 @@ fn decode_text(bytes: &[u8]) -> std::borrow::Cow<'_, str> { } } -/// Adobe / Resolve `.cube`. 1D LUTs (`LUT_1D_SIZE`) are expanded to a 33³ table. -pub fn parse_cube(text: &str) -> Result { - let mut title = String::new(); - let (mut size3, mut size1) = (0usize, 0usize); - let (mut dmin, mut dmax) = ([0.0f32; 3], [1.0f32; 3]); - let mut rows: Vec = Vec::new(); - for line in text.lines() { - // Only a quote at the beginning of a TITLE value protects '#'. Quotes inside an - // unquoted title's comment must not turn that comment into part of the title. - if let Some(quoted) = line - .trim_start() - .strip_prefix("TITLE") - .filter(|rest| rest.starts_with(char::is_whitespace)) - .and_then(|rest| rest.trim_start().strip_prefix('"')) +/// A bounded, three-dimensional Cube profile with RGB rows in file order. +/// Unlike [`parse_cube`], this preserves the optional title and does not resample 1D curves. +#[derive(Clone, Debug, PartialEq)] +pub struct Cube3d { + pub title: Option, + pub size: usize, + pub data: Vec<[f32; 3]>, + pub domain_min: [f32; 3], + pub domain_max: [f32; 3], +} + +/// Read the simple 3D profile dialect used by camera-profile consumers. +/// +/// `max_size` must be in 2..=256; the row budget is checked before each allocation. +/// Whole-line comments and literal titles are preserved. Unknown alphabetic directives +/// are ignored; 1D tables are refused. Finite domains are returned unchanged so the +/// consumer retains its existing policy for equal or reversed endpoints. +/// Numeric fields must consist of exactly three finite numbers (never silently skip tokens). +pub fn parse_cube_3d(text: &str, max_size: usize) -> Result { + let p = read_cube(text, Some(max_size))?; + if p.size3 == 0 { + return err("not a 3D .cube LUT (no LUT_3D_SIZE)"); + } + let expected = p.size3.checked_pow(3).ok_or_else(|| LutError("LUT size overflow".into()))?; + if p.rows.len() != expected { + return err(format!("expected {expected} entries, found {}", p.rows.len())); + } + Ok(Cube3d { title: p.title, size: p.size3, data: p.rows, domain_min: p.min, domain_max: p.max }) +} + +struct CubeRows { + title: Option, + size3: usize, + size1: usize, + min: [f32; 3], + max: [f32; 3], + rows: Vec<[f32; 3]>, +} + +fn cube_triplet(text: &str) -> Result<[f32; 3], LutError> { + let mut it = text.split_whitespace(); + let mut row = [0.0; 3]; + for value in &mut row { + *value = it.next().and_then(|s| s.parse::().ok()).filter(|v| v.is_finite()).ok_or_else(|| LutError(format!("bad line `{text}`")))?; + } + if it.next().is_some() { + return err(format!("bad line `{text}`")); + } + Ok(row) +} + +fn read_cube(text: &str, profile_max: Option) -> Result { + if profile_max.is_some_and(|n| !(2..=256).contains(&n)) { + return err("3D profile size limit must be in 2..=256"); + } + let mut p = CubeRows { title: None, size3: 0, size1: 0, min: [0.0; 3], max: [1.0; 3], rows: Vec::new() }; + let row_limit = profile_max.unwrap_or(256).checked_pow(3).ok_or_else(|| LutError("LUT size overflow".into()))?; + for raw in text.lines() { + let profile = profile_max.is_some(); + let line = if profile { raw.trim() } else { raw.split('#').next().unwrap_or("").trim() }; + // The standard dialect protects '#' inside quoted titles; profile titles are literal. + if !profile + && let Some(quoted) = + raw.trim_start().strip_prefix("TITLE").filter(|r| r.starts_with(char::is_whitespace)).and_then(|r| r.trim_start().strip_prefix('"')) { - let Some((value, _)) = quoted.split_once('"') else { return err("unterminated quoted TITLE") }; - title = value.to_string(); + let Some((title, _)) = quoted.split_once('"') else { return err("unterminated quoted TITLE") }; + p.title = Some(title.to_string()); continue; } - let line = line.split('#').next().unwrap_or("").trim(); - if line.is_empty() { + if line.is_empty() || line.starts_with('#') { continue; } - let mut it = line.split_whitespace(); - let key = it.next().unwrap_or(""); - let nums = |it: std::str::SplitWhitespace<'_>| -> Result<[f32; 3], LutError> { - let v: Vec = it.filter_map(|s| s.parse().ok()).collect(); - if v.len() == 3 { Ok([v[0], v[1], v[2]]) } else { err(format!("bad line `{line}`")) } + let (key, rest) = if profile { + ["TITLE", "LUT_3D_SIZE", "LUT_1D_SIZE", "DOMAIN_MIN", "DOMAIN_MAX"] + .into_iter() + .find_map(|key| line.strip_prefix(key).map(|rest| (key, rest.trim()))) + .unwrap_or_else(|| line.split_once(char::is_whitespace).unwrap_or((line, ""))) + } else { + line.split_once(char::is_whitespace).map_or((line, ""), |(key, rest)| (key, rest.trim())) }; match key { - "TITLE" => title = line[5..].trim().trim_matches('"').to_string(), - "LUT_3D_SIZE" => size3 = it.next().and_then(|s| s.parse().ok()).ok_or_else(|| LutError("bad LUT_3D_SIZE".into()))?, - "LUT_1D_SIZE" => size1 = it.next().and_then(|s| s.parse().ok()).ok_or_else(|| LutError("bad LUT_1D_SIZE".into()))?, - "DOMAIN_MIN" => dmin = nums(it)?, - "DOMAIN_MAX" => dmax = nums(it)?, - "LUT_3D_INPUT_RANGE" | "LUT_1D_INPUT_RANGE" => { - let v: Vec = it.filter_map(|s| s.parse().ok()).collect(); - if v.len() == 2 { - dmin = [v[0]; 3]; - dmax = [v[1]; 3]; + "TITLE" => p.title = Some(rest.trim_matches('"').to_string()), + "LUT_3D_SIZE" => { + let value = if profile { rest } else { rest.split_whitespace().next().unwrap_or("") }; + p.size3 = value.parse().map_err(|_| LutError(format!("bad LUT_3D_SIZE `{value}`")))?; + let limit = profile_max.unwrap_or(MAX_SIZE); + if !(2..=limit).contains(&p.size3) { + return err(format!("LUT_3D_SIZE {} out of range", p.size3)); } } - k if k.starts_with(|c: char| c.is_ascii_digit() || c == '-' || c == '+' || c == '.') => { - let mut v = vec![k.parse::().map_err(|_| LutError(format!("bad number in `{line}`")))?]; - for value in it { - v.push(value.parse::().map_err(|_| LutError(format!("bad number in `{line}`")))?); + "LUT_1D_SIZE" => { + if profile { + return err("1D LUTs are not supported (only 3D .cube files)"); } - if v.len() != 3 { - return err(format!("bad data line `{line}`")); + p.size1 = rest.split_whitespace().next().and_then(|s| s.parse().ok()).ok_or_else(|| LutError("bad LUT_1D_SIZE".into()))?; + if p.size1 > row_limit { + return err("1D LUT exceeds row budget"); + } + } + "DOMAIN_MIN" => p.min = cube_triplet(rest).map_err(|_| LutError("bad DOMAIN_MIN".into()))?, + "DOMAIN_MAX" => p.max = cube_triplet(rest).map_err(|_| LutError("bad DOMAIN_MAX".into()))?, + "LUT_3D_INPUT_RANGE" | "LUT_1D_INPUT_RANGE" if !profile => { + let mut it = rest.split_whitespace(); + let min = it.next().and_then(|s| s.parse::().ok()); + let max = it.next().and_then(|s| s.parse::().ok()); + if let (Some(min), Some(max), None) = (min, max, it.next()) { + p.min = [min; 3]; + p.max = [max; 3]; + } + } + _ => { + let numeric = key.starts_with(|c: char| c.is_ascii_digit() || c == '-' || c == '+' || c == '.'); + let nonfinite = key.parse::().is_ok(); + if numeric || nonfinite || (profile && !key.starts_with(|c: char| c.is_ascii_alphabetic())) { + let row = cube_triplet(line)?; + if p.rows.len() >= row_limit { + return err("LUT exceeds row budget"); + } + // Explicit fallible growth prevents capacity growth past the configured budget. + if p.rows.len() == p.rows.capacity() { + let grow = (p.rows.capacity().max(1)).min(row_limit - p.rows.len()); + p.rows.try_reserve_exact(grow).map_err(|_| LutError("LUT allocation failed".into()))?; + } + p.rows.push(row); } - rows.extend(v); } - // Unknown keywords (e.g. Resolve's LUT_IN_VIDEO_RANGE) are ignored, as the spec allows. - _ => {} } } + Ok(p) +} + +/// Adobe / Resolve `.cube`. 1D LUTs (`LUT_1D_SIZE`) are expanded to a 33³ table. +pub fn parse_cube(text: &str) -> Result { + let p = read_cube(text, None)?; + let (title, size3, size1, dmin, dmax) = (p.title.unwrap_or_default(), p.size3, p.size1, p.min, p.max); + let rows: Vec = p.rows.into_iter().flatten().collect(); let domain_ok = (0..3).all(|k| dmin[k] < dmax[k] && dmin[k].is_finite() && dmax[k].is_finite()); if (size3 > 0 || size1 >= 2) && !domain_ok { return err(format!("DOMAIN_MIN {:?} must lie below DOMAIN_MAX {:?}", dmin, dmax)); @@ -531,12 +611,52 @@ pub fn write_cube(l: &LutFile) -> String { s.push_str(&format!("DOMAIN_MAX {} {} {}\n", l.domain_max[0], l.domain_max[1], l.domain_max[2])); } s.push_str(&format!("LUT_3D_SIZE {}\n", l.size)); - for c in l.data.as_chunks::<3>().0 { + cube_rows(s, l.data.as_chunks::<3>().0.iter().copied()) +} + +/// Writes a `.cube` file with a creator comment, section comments and an explicit input domain. +/// +/// This layout is useful for exported LUTs; [`write_cube`] keeps its compact layout for embedded +/// tables. Values retain six decimal places, and their order and quantisation belong to the +/// caller. A default domain is written as `0.0` and `1.0` for compatibility with export consumers. +pub fn write_cube_commented(l: &LutFile, creator: &str) -> String { + let mut s = format!("#Created by: {creator}\nTITLE \"{}\"\n\n#LUT size\nLUT_3D_SIZE {}\n\n#data domain\n", l.title.replace('"', "'"), l.size); + if l.domain_is_default() { + s.push_str("DOMAIN_MIN 0.0 0.0 0.0\nDOMAIN_MAX 1.0 1.0 1.0\n"); + } else { + s.push_str(&format!("DOMAIN_MIN {} {} {}\n", l.domain_min[0], l.domain_min[1], l.domain_min[2])); + s.push_str(&format!("DOMAIN_MAX {} {} {}\n", l.domain_max[0], l.domain_max[1], l.domain_max[2])); + } + s.push_str("\n#LUT data points\n"); + cube_rows(s, l.data.as_chunks::<3>().0.iter().copied()) +} + +fn cube_rows(mut s: String, rows: impl Iterator) -> String { + for c in rows { s.push_str(&format!("{:.6} {:.6} {:.6}\n", c[0], c[1], c[2])); } s } +/// Annotated identity-domain CUBE export, streamed in red-fastest order without a table copy. +/// Callers supply exactly `size³` rows for a valid size (2..=[`MAX_SIZE`]), as with [`write_cube`]. +/// Integer-depth exports clamp/round to 1/32768 steps; floating exports clamp without rounding. +pub fn write_cube_export(creator: &str, title: &str, size: usize, integer_depth: bool, rows: impl Iterator) -> String { + let s = format!( + "#Created by: {creator}\nTITLE \"{}\"\n\n#LUT size\nLUT_3D_SIZE {size}\n\n#data domain\nDOMAIN_MIN 0.0 0.0 0.0\nDOMAIN_MAX 1.0 1.0 1.0\n\n#LUT data points\n", + title.replace('"', "'") + ); + cube_rows( + s, + rows.map(|c| { + c.map(|v| { + let v = v.clamp(0.0, 1.0); + if integer_depth { (v * 32768.0).round() / 32768.0 } else { v } + }) + }), + ) +} + fn luma(c: [f32; 3]) -> f32 { 0.299 * c[0] + 0.587 * c[1] + 0.114 * c[2] } @@ -602,6 +722,30 @@ pub fn builtin(id: &str) -> Option { mod tests { use super::*; + #[test] + fn commented_cube_preserves_export_header_and_rows() { + let mut lut = super::LutFile::identity(2).unwrap(); + lut.title = "A \"look\"".into(); + let text = super::write_cube_commented(&lut, "ExampleCraft"); + assert!(text.starts_with("#Created by: ExampleCraft\nTITLE \"A 'look'\"\n\n#LUT size\nLUT_3D_SIZE 2\n\n#data domain\nDOMAIN_MIN 0.0 0.0 0.0\nDOMAIN_MAX 1.0 1.0 1.0\n\n#LUT data points\n")); + let parsed = super::parse_cube(&text).unwrap(); + assert_eq!(parsed.data, lut.data); + assert_eq!(parsed.title, "A 'look'"); + assert_eq!(text.lines().count(), 11 + 8); + let rows = |s: &str| s.lines().filter(|line| line.starts_with(|c: char| c.is_ascii_digit())).map(str::to_owned).collect::>(); + assert_eq!(rows(&text), rows(&super::write_cube(&lut))); + } + + #[test] + fn commented_cube_keeps_nondefault_domain_and_sample_precision() { + let mut lut = super::LutFile::identity(2).unwrap(); + lut.domain_min = [0.125, 0.25, 0.375]; + lut.domain_max = [0.625, 0.75, 0.875]; + lut.data[0] = 0.833344; + let parsed = super::parse_cube(&super::write_cube_commented(&lut, "ExampleCraft")).unwrap(); + assert_eq!(parsed, lut); + } + #[test] fn cube_title_hashes_and_comment_quotes_are_distinct() { let rows = "LUT_1D_SIZE 2\n0 0 0 # low\n1 1 1 # high\n"; @@ -1299,4 +1443,75 @@ LUT_3D_SIZE 2 assert_eq!(back.domain_min, l.domain_min, "the written table must mean the same"); assert_eq!(back.domain_max, l.domain_max); } + #[test] + fn annotated_cube_export_preserves_header_and_integer_precision() { + let mut lut = LutFile::identity(2).unwrap(); + lut.title = "My \"Look\"".into(); + lut.data[..3].copy_from_slice(&[1.0 / 6.0, 5.0 / 6.0, 160.0 / 255.0]); + let output = write_cube_export("PhotoCraft", &lut.title, lut.size, true, lut.data.as_chunks::<3>().0.iter().copied()); + assert!(output.starts_with("#Created by: PhotoCraft\nTITLE \"My 'Look'\"\n\n#LUT size\nLUT_3D_SIZE 2\n\n#data domain\nDOMAIN_MIN 0.0 0.0 0.0\nDOMAIN_MAX 1.0 1.0 1.0\n\n#LUT data points\n0.166656 0.833344 0.627441\n")); + assert_eq!(output.lines().count(), 11 + 8); + let parsed = parse_cube(&output).unwrap(); + assert_eq!(parsed.size, 2); + assert_eq!(&parsed.data[3..], &lut.data[3..]); + let float = write_cube_export("PhotoCraft", &lut.title, lut.size, false, lut.data.as_chunks::<3>().0.iter().copied()); + assert!(float.contains("0.166667 0.833333 0.627451\n")); + } + + #[test] + fn annotated_empty_title_and_float_clamping() { + let text = write_cube_export("Exporter", "", 2, false, std::iter::repeat_n([-0.5, 0.25, 1.5], 8)); + assert!(text.contains("TITLE \"\"\n")); + let back = parse_cube(&text).unwrap(); + assert_eq!(back.data, [0.0, 0.25, 1.0].repeat(8)); + } +} + +#[cfg(test)] +mod profile_cube_tests { + use super::*; + + fn cube(prefix: &str) -> String { + format!("{prefix}\nLUT_3D_SIZE 2\n{}", "0.25 0.5 0.75\n".repeat(8)) + } + + #[test] + fn profile_preserves_optional_literal_title_and_domains() { + let no_title = parse_cube_3d(&cube(""), 256).unwrap(); + assert_eq!(no_title.title, None); + let value = parse_cube_3d(&cube("TITLE \"look # literal\"\nDOMAIN_MIN 1 0 2\nDOMAIN_MAX 0 0 1"), 256).unwrap(); + assert_eq!(value.title.as_deref(), Some("look # literal")); + assert_eq!(value.domain_min, [1.0, 0.0, 2.0]); + assert_eq!(value.domain_max, [0.0, 0.0, 1.0]); + assert_eq!(value.data, vec![[0.25, 0.5, 0.75]; 8]); + assert_eq!(parse_cube_3d(&cube("TITLE \"\""), 256).unwrap().title.as_deref(), Some("")); + assert!(parse_cube(&cube("DOMAIN_MIN 1 0 2\nDOMAIN_MAX 0 0 1")).is_err()); + } + + #[test] + fn profile_limits_are_checked_without_large_allocations() { + assert!(parse_cube_3d("LUT_3D_SIZE 256", 256).unwrap_err().0.contains("16777216")); + assert!(parse_cube_3d("LUT_3D_SIZE 257", 256).unwrap_err().0.contains("out of range")); + assert!(parse_cube_3d("LUT_3D_SIZE 256", 129).unwrap_err().0.contains("out of range")); + assert!(parse_cube_3d(&cube(""), usize::MAX).is_err()); + assert!(parse_cube_3d(&cube(""), 1).is_err()); + assert!(parse_cube_3d(&format!("{}0 0 0\n", cube("")), 2).unwrap_err().0.contains("budget")); + assert!(parse_cube_3d("LUT_1D_SIZE 2\n0 0 0\n1 1 1", 256).is_err()); + } + + #[test] + fn every_numeric_token_must_be_valid_and_finite() { + for row in ["0 ignored 0 0", "0 0 0 ignored", "NaN 0 0", "inf 0 0", "0 0 -inf", "1e39 0 0", "0 0", "0 0 0 0"] { + assert!(parse_cube_3d(&format!("LUT_3D_SIZE 2\n{row}\n{}", "0 0 0\n".repeat(7)), 256).is_err(), "{row}"); + assert!(parse_cube_3d(&cube(&format!("DOMAIN_MIN {row}")), 256).is_err(), "{row}"); + } + } + + #[test] + fn standard_reader_keeps_quoted_comments_and_1d_expansion() { + assert_eq!(parse_cube(&cube("TITLE \"look # preserved\" # comment")).unwrap().title, "look # preserved"); + let value = parse_cube("LUT_1D_SIZE 2\n0 0 0\n1 1 1").unwrap(); + assert_eq!(value.size, 33); + assert!(parse_cube_3d(&cube("DOMAIN_MIN 0 0 0 # not a whole-line comment"), 256).is_err()); + } } diff --git a/crates/cms/src/normalize.rs b/crates/cms/src/normalize.rs new file mode 100644 index 000000000..85d690783 --- /dev/null +++ b/crates/cms/src/normalize.rs @@ -0,0 +1,142 @@ +//! Bounded encoded ICC layout normalization, originating in LightCraft. + +use std::collections::BTreeMap; + +/// Zero the creation date, align tag payloads to four bytes, preserve shared payloads, +/// update the profile length and clear its stale profile ID. No colour tags are interpreted. +/// +/// This is a layout operation, not profile validation. Malformed tables or layouts exceeding +/// 4096 tags or 64 MiB output are returned unchanged except for the creation date. The owned +/// input is reused on failure, so rejected input does not require an additional full copy. +/// Consumers needing rejection of invalid profiles must separately use `Profile::parse`. +pub fn normalize_for_embedding(mut bytes: Vec) -> Vec { + if let Some(date) = bytes.get_mut(24..36) { + date.fill(0); + } + align(&bytes).unwrap_or(bytes) +} + +const MAX_BYTES: usize = 64 * 1024 * 1024; +const MAX_TAGS: usize = 4096; + +fn be(bytes: &[u8], at: usize) -> Option { + let s = bytes.get(at..at.checked_add(4)?)?; + usize::try_from(u32::from_be_bytes([s[0], s[1], s[2], s[3]])).ok() +} + +fn align(bytes: &[u8]) -> Option> { + if bytes.len() > MAX_BYTES { + return None; + } + let count = be(bytes, 128)?; + if count > MAX_TAGS { + return None; + } + let end = 132usize.checked_add(count.checked_mul(12)?)?; + bytes.get(..end)?; + let mut placements = BTreeMap::new(); + let mut offsets = Vec::new(); + offsets.try_reserve_exact(count).ok()?; + let mut size = end; + for index in 0..count { + let entry = 132 + index * 12; + let off = be(bytes, entry + 4)?; + let len = be(bytes, entry + 8)?; + if off < end { + return None; + } + bytes.get(off..off.checked_add(len)?)?; + let new_off = if let Some(&placed) = placements.get(&(off, len)) { + placed + } else { + let aligned = size.checked_add(3)? & !3; + size = aligned.checked_add(len)?; + if size > MAX_BYTES { + return None; + } + placements.insert((off, len), aligned); + aligned + }; + offsets.push(new_off); + } + size = size.checked_add(3)? & !3; + if size > MAX_BYTES { + return None; + } + let mut out = Vec::new(); + out.try_reserve_exact(size).ok()?; + out.resize(size, 0); + out[..end].copy_from_slice(&bytes[..end]); + for ((off, len), target) in placements { + out[target..target + len].copy_from_slice(&bytes[off..off + len]); + } + for (index, off) in offsets.into_iter().enumerate() { + let field = 132 + index * 12 + 4; + out[field..field + 4].copy_from_slice(&u32::try_from(off).ok()?.to_be_bytes()); + } + out[..4].copy_from_slice(&u32::try_from(size).ok()?.to_be_bytes()); + out[84..100].fill(0); + Some(out) +} + +#[cfg(test)] +#[allow(clippy::unwrap_used, clippy::expect_used, clippy::panic)] +#[path = "normalize_legacy_tests.rs"] +mod legacy; + +#[cfg(test)] +#[allow(clippy::unwrap_used, clippy::expect_used, clippy::panic)] +mod tests { + use super::*; + + #[test] + fn matches_frozen_writer_and_preserves_aliases() { + for count in 0..20usize { + let end = 132 + count * 12; + let mut bytes = vec![17; end + 31]; + bytes[128..132].copy_from_slice(&(count as u32).to_be_bytes()); + for index in 0..count { + let field = 132 + index * 12 + 4; + let off = end + index % 3; + bytes[field..field + 4].copy_from_slice(&(off as u32).to_be_bytes()); + bytes[field + 4..field + 8].copy_from_slice(&7u32.to_be_bytes()); + } + let expected = legacy::align_tags(&legacy::fixed_date(bytes.clone())); + let result = normalize_for_embedding(bytes); + assert_eq!(result, expected); + assert_eq!(normalize_for_embedding(result.clone()), result); + } + } + + #[test] + fn real_profile_preserves_color_tags_and_legacy_bytes() { + let encoded = crate::Builtin::Srgb.profile().clone().with_encoded_bytes(); + let input = encoded.to_bytes().to_vec(); + let normalized = normalize_for_embedding(input.clone()); + assert_eq!(normalized, legacy::align_tags(&legacy::fixed_date(input))); + assert_eq!(&normalized[24..36], &[0; 12]); + assert_eq!(&normalized[84..100], &[0; 16]); + assert!(crate::Profile::parse(&normalized).is_ok()); + for index in 0..be(&normalized, 128).unwrap() { + assert_eq!(be(&normalized, 136 + index * 12).unwrap() % 4, 0); + } + } + + #[test] + fn hostile_tables_return_owned_input_without_expansion() { + for len in 0..132 { + let bytes = vec![1; len]; + assert_eq!(normalize_for_embedding(bytes.clone()), legacy::fixed_date(bytes)); + } + for count in [4097u32, u32::MAX] { + let mut bytes = vec![1; 144]; + bytes[128..132].copy_from_slice(&count.to_be_bytes()); + assert_eq!(normalize_for_embedding(bytes.clone()), legacy::fixed_date(bytes)); + } + let mut bytes = vec![1; 144]; + bytes[128..132].copy_from_slice(&1u32.to_be_bytes()); + bytes[136..140].copy_from_slice(&u32::MAX.to_be_bytes()); + bytes[140..144].copy_from_slice(&u32::MAX.to_be_bytes()); + assert_eq!(normalize_for_embedding(bytes.clone()), legacy::fixed_date(bytes)); + } +} diff --git a/crates/cms/src/normalize_legacy_tests.rs b/crates/cms/src/normalize_legacy_tests.rs new file mode 100644 index 000000000..dae5c2770 --- /dev/null +++ b/crates/cms/src/normalize_legacy_tests.rs @@ -0,0 +1,43 @@ +//! Encoded ICC normalization extracted from LightCraft. +pub(super) fn align_tags(b: &[u8]) -> Vec { + let be = |i: usize| b.get(i..i + 4).map(|s| u32::from_be_bytes([s[0], s[1], s[2], s[3]]) as usize); + let Some(n) = be(128) else { return b.to_vec() }; + let table_end = 132 + 12 * n; + let tags: Option> = (0..n).map(|t| Some((be(132 + 12 * t + 4)?, be(132 + 12 * t + 8)?))).collect(); + let Some(tags) = tags.filter(|t| table_end <= b.len() && t.iter().all(|&(o, l)| o >= table_end && o + l <= b.len())) else { + return b.to_vec(); + }; + let mut out = b[..table_end].to_vec(); + let mut placed: Vec<((usize, usize), usize)> = Vec::new(); + for (t, &(off, len)) in tags.iter().enumerate() { + let new_off = match placed.iter().find(|(k, _)| *k == (off, len)) { + // tags sharing data keep sharing it + Some(&(_, o)) => o, + None => { + while !out.len().is_multiple_of(4) { + out.push(0); + } + let o = out.len(); + out.extend_from_slice(&b[off..off + len]); + placed.push(((off, len), o)); + o + } + }; + out[132 + 12 * t + 4..132 + 12 * t + 8].copy_from_slice(&(new_off as u32).to_be_bytes()); + } + while !out.len().is_multiple_of(4) { + out.push(0); + } + let size = out.len() as u32; + out[0..4].copy_from_slice(&size.to_be_bytes()); + // the profile ID (MD5) would no longer match: clear it (= "not computed") + out[84..100].fill(0); + out +} + +pub(super) fn fixed_date(mut b: Vec) -> Vec { + if let Some(date) = b.get_mut(24..36) { + date.fill(0); + } + b +} diff --git a/crates/cms/src/sony_lut.rs b/crates/cms/src/sony_lut.rs new file mode 100644 index 000000000..cc3f8fb00 --- /dev/null +++ b/crates/cms/src/sony_lut.rs @@ -0,0 +1,82 @@ +//! Sony Imageworks SPI3D and SPIMTX text readers, extracted from EffectCraft. +//! +//! These preserve the source reader's format tolerance. SPI3D sample values are not +//! clamped or sanitized; rendering policy belongs to the caller. Duplicate coordinates +//! retain their last value, and the total row count must match the lattice volume. + +/// Maximum lattice side accepted by the source reader. +pub const MAX_SIDE: usize = 129; + +/// Row-major homogeneous transform with a separate normalized code-value offset. +#[derive(Clone, Debug, PartialEq)] +pub struct Matrix { + pub matrix: [f64; 16], + pub offset: [f64; 4], +} + +/// A cubic RGB table, red coordinate varying fastest. +#[derive(Clone, Debug, PartialEq)] +pub struct Lattice { + pub size: usize, + pub values: Vec<[f32; 3]>, +} + +/// Read exactly twelve finite SPIMTX numbers, dividing the offset column by 65535. +/// Parsing uses a fixed array, not storage proportional to an oversized input. +pub fn parse_spimtx(text: &str) -> Option { + let mut words = text.split_whitespace(); + let mut v = [0.0f64; 12]; + for value in &mut v { + *value = words.next()?.parse().ok()?; + if !value.is_finite() { + return None; + } + } + if words.next().is_some() { + return None; + } + let mut matrix = [0.0; 16]; + let mut offset = [0.0; 4]; + for (r, row) in v.as_chunks::<4>().0.iter().enumerate() { + matrix[r * 4..r * 4 + 3].copy_from_slice(&row[..3]); + offset[r] = row[3] / 65535.0; + } + matrix[15] = 1.0; + Some(Matrix { matrix, offset }) +} + +/// Read a SPI3D table with a bounded 2..=129 cubic lattice and exactly six fields per row. +/// A `SPILUT` prefix is accepted and the second header line is ignored, as in EffectCraft. +pub fn parse_spi3d(text: &str) -> Option { + let mut lines = text.lines().map(str::trim).filter(|l| !l.is_empty()); + if !lines.next()?.starts_with("SPILUT") { + return None; + } + lines.next()?; + let mut sizes = lines.next()?.split_whitespace(); + let n: usize = sizes.next()?.parse().ok()?; + let g: usize = sizes.next()?.parse().ok()?; + let b: usize = sizes.next()?.parse().ok()?; + if sizes.next().is_some() || g != n || b != n || !(2..=MAX_SIDE).contains(&n) { + return None; + } + let count = n * n * n; + let mut values = vec![[0.0f32; 3]; count]; + let mut seen = 0usize; + for line in lines { + if seen == count { + return None; + } + let mut fields = line.split_whitespace(); + let r: usize = fields.next()?.parse().ok()?; + let g: usize = fields.next()?.parse().ok()?; + let b: usize = fields.next()?.parse().ok()?; + let color = [fields.next()?.parse().ok()?, fields.next()?.parse().ok()?, fields.next()?.parse().ok()?]; + if fields.next().is_some() || r >= n || g >= n || b >= n { + return None; + } + *values.get_mut(r + g * n + b * n * n)? = color; + seen += 1; + } + (seen == count).then_some(Lattice { size: n, values }) +} diff --git a/crates/cms/src/spi1d.rs b/crates/cms/src/spi1d.rs new file mode 100644 index 000000000..ea77bdc88 --- /dev/null +++ b/crates/cms/src/spi1d.rs @@ -0,0 +1,59 @@ +//! Sony Imageworks SPI1D tables, preserving EffectCraft's permissive header syntax. +//! +//! Unknown header lines and trailing text after the closing brace are ignored. Table +//! samples are not clamped or sanitized: callers own their numerical rendering policy. +//! The declared length is validated before storing samples; surplus samples are refused. + +/// Largest accepted table (the existing EffectCraft interpolation limit). +pub const MAX_ENTRIES: usize = 1 << 24; + +/// A one-dimensional RGB lookup table and its common input domain. +#[derive(Clone, Debug, PartialEq)] +pub struct Table { + pub values: Vec<[f32; 3]>, + pub domain: [f32; 2], +} + +/// Parse a SPI1D table with one or three components, expanding monochrome samples to RGB. +/// No allocation is based only on the claimed length. At most `MAX_ENTRIES` RGB entries +/// are retained. Missing final braces remain accepted, as in the source reader. +pub fn parse(text: &str) -> Option { + let (mut from, mut len, mut comps) = ([0.0f32, 1.0], 0usize, 1usize); + let mut values = Vec::new(); + let mut body = false; + let mut sample = [0.0; 3]; + let mut channel = 0; + for line in text.lines().map(str::trim).filter(|l| !l.is_empty()) { + if body { + if line.starts_with('}') { + break; + } + for t in line.split_whitespace() { + if values.len() == len { + return None; + } + sample[channel] = t.parse().ok()?; + channel += 1; + if channel == comps { + values.push(if comps == 1 { [sample[0]; 3] } else { sample }); + channel = 0; + } + } + continue; + } + let mut t = line.split_whitespace(); + match t.next()? { + "{" => { + if !(2..=MAX_ENTRIES).contains(&len) || !matches!(comps, 1 | 3) || from[1] <= from[0] || from.iter().any(|v| !v.is_finite()) { + return None; + } + body = true; + } + "From" => from = [t.next()?.parse().ok()?, t.next()?.parse().ok()?], + "Length" => len = t.next()?.parse().ok()?, + "Components" => comps = t.next()?.parse().ok()?, + _ => {} + } + } + (body && values.len() == len && channel == 0).then_some(Table { values, domain: from }) +} diff --git a/crates/cms/src/synth.rs b/crates/cms/src/synth.rs index a59058f08..b1535f78c 100644 --- a/crates/cms/src/synth.rs +++ b/crates/cms/src/synth.rs @@ -111,9 +111,9 @@ impl CmykModel { // Overprints with black: multiplicative in reflectance above a first-surface floor. let floor = 0.006; let mut prim = [[0.0; 3]; 16]; - for m in 0..16 { + for (m, slot) in prim.iter_mut().enumerate() { let p = base[m & 7]; - prim[m] = if m & 8 == 0 { + *slot = if m & 8 == 0 { p } else { std::array::from_fn(|i| { @@ -133,7 +133,7 @@ impl CmykModel { (v + 4.0 * self.tvi[i] * v * (1.0 - v)).clamp(0.0, 1.0) }); let mut s = [0.0; 3]; - for m in 0..16 { + for (m, prim) in self.prim_n.iter().enumerate() { let mut w = 1.0; for (i, ai) in a.iter().enumerate() { w *= if m & (1 << i) != 0 { *ai } else { 1.0 - ai }; @@ -141,7 +141,7 @@ impl CmykModel { if w == 0.0 { continue; } - for (acc, p) in s.iter_mut().zip(&self.prim_n[m]) { + for (acc, p) in s.iter_mut().zip(prim) { *acc += w * p; } } diff --git a/crates/cms/src/write_v2.rs b/crates/cms/src/write_v2.rs new file mode 100644 index 000000000..63933c801 --- /dev/null +++ b/crates/cms/src/write_v2.rs @@ -0,0 +1,195 @@ +//! Bounded ICC v2.1 matrix/TRC display-profile writer, extracted from TextureCraft. +//! Colour-space selection and transfer-curve sampling remain caller policy. + +const D50_XYZ: [f64; 3] = [0.9642, 1.0, 0.8249]; + +/// One common red/green/blue transfer curve. +#[derive(Clone, Copy)] +pub enum ToneCurve<'a> { + /// ICC gamma 1.0 (`curv` with one u8Fixed8 entry). + Linear, + /// Already quantized samples, inclusive endpoints, between 2 and 4096 entries. + Samples(&'a [u16]), +} + +/// Parameters for an ICC v2.1 RGB display profile. +pub struct MatrixProfile<'a> { + /// RGB-to-XYZ matrix already adapted to PCS D50, in row-major order. + pub matrix: [[f64; 3]; 3], + /// Common RGB transfer curve; its storage is shared by all three tags. + pub curve: ToneCurve<'a>, + /// ASCII description, at most 1024 bytes with no embedded NUL. + pub description: &'a str, + /// ASCII copyright text, at most 1024 bytes with no embedded NUL. + pub copyright: &'a str, + /// Raw ICC date fields: year, month, day, hour, minute, second. + pub date: [u16; 6], +} + +fn s15f16(v: f64) -> [u8; 4] { + ((v * 65536.0).round().clamp(f64::from(i32::MIN), f64::from(i32::MAX)) as i32).to_be_bytes() +} + +fn xyz_tag(v: [f64; 3]) -> Vec { + let mut b = b"XYZ \0\0\0\0".to_vec(); + for c in v { + b.extend_from_slice(&s15f16(c)); + } + b +} + +fn text_tag(text: &str) -> Vec { + let mut b = b"text\0\0\0\0".to_vec(); + b.extend_from_slice(text.as_bytes()); + b.push(0); + b +} + +/// `textDescriptionType` with the ASCII description only. +fn desc_tag(text: &str) -> Vec { + let mut b = b"desc\0\0\0\0".to_vec(); + b.extend_from_slice(&(text.len() as u32 + 1).to_be_bytes()); + b.extend_from_slice(text.as_bytes()); + b.push(0); + // No Unicode text and no Macintosh ScriptCode text: empty fields and the 67-byte filler. + b.extend_from_slice(&[0; 4 + 4 + 2 + 1 + 67]); + b +} + +fn curve_tag(curve: ToneCurve<'_>) -> Vec { + let mut b = b"curv\0\0\0\0".to_vec(); + match curve { + ToneCurve::Linear => { + b.extend_from_slice(&1u32.to_be_bytes()); + b.extend_from_slice(&0x0100u16.to_be_bytes()); + } + ToneCurve::Samples(samples) => { + b.extend_from_slice(&(samples.len() as u32).to_be_bytes()); + for sample in samples { + b.extend_from_slice(&sample.to_be_bytes()); + } + } + } + b +} + +/// Write a deterministic ICC v2.1 RGB display profile with one shared RGB TRC. +/// +/// Returns `None` for nonfinite matrix values, descriptions longer than 1024 ASCII +/// bytes or containing NUL, or sampled curves outside 2..=4096 entries. These caps +/// bound every allocation independently of caller input. Matrix values outside the +/// signed 15.16 range saturate, matching the legacy exporter. The caller owns colour +/// adaptation, transfer sampling, names and timestamp policy. +pub fn matrix_profile(params: &MatrixProfile<'_>) -> Option> { + let valid_text = |s: &str| s.len() <= 1024 && s.is_ascii() && !s.contains('\0'); + if !valid_text(params.description) + || !valid_text(params.copyright) + || !params.matrix.iter().flatten().all(|v| v.is_finite()) + || matches!(params.curve, ToneCurve::Samples(s) if !(2..=4096).contains(&s.len())) + { + return None; + } + let m = params.matrix; + let column = |c: usize| [m[0][c], m[1][c], m[2][c]]; + let curve = curve_tag(params.curve); + // The green and blue curves are the red one's entry. + let tags: [(&[u8; 4], Vec); 9] = [ + (b"desc", desc_tag(params.description)), + (b"cprt", text_tag(params.copyright)), + (b"wtpt", xyz_tag(D50_XYZ)), + (b"rXYZ", xyz_tag(column(0))), + (b"gXYZ", xyz_tag(column(1))), + (b"bXYZ", xyz_tag(column(2))), + (b"rTRC", curve.clone()), + (b"gTRC", curve.clone()), + (b"bTRC", curve), + ]; + // Header (128) + tag count (4) + 12 per tag, then the tag data, each tag padded to 4 bytes. + let table_end = 128 + 4 + 12 * tags.len(); + let mut offsets = Vec::new(); + let mut data: Vec = Vec::new(); + for (i, (_, bytes)) in tags.iter().enumerate() { + let shared = i >= 7; + let at = if shared { offsets.get(6).copied().unwrap_or(0) } else { table_end + data.len() }; + offsets.push(at); + if !shared { + data.extend_from_slice(bytes); + while !data.len().is_multiple_of(4) { + data.push(0); + } + } + } + let total = table_end + data.len(); + let mut out = vec![0u8; 128]; + out.get_mut(0..4)?.copy_from_slice(&(total as u32).to_be_bytes()); + out.get_mut(8..12)?.copy_from_slice(&[2, 0x10, 0, 0]); // version 2.1.0 + out.get_mut(12..16)?.copy_from_slice(b"mntr"); + out.get_mut(16..20)?.copy_from_slice(b"RGB "); + out.get_mut(20..24)?.copy_from_slice(b"XYZ "); + // Serialize caller-supplied ICC date fields; no calendar arithmetic. + for (i, value) in params.date.iter().enumerate() { + out.get_mut(24 + i * 2..26 + i * 2)?.copy_from_slice(&value.to_be_bytes()); + } + out.get_mut(36..40)?.copy_from_slice(b"acsp"); + for (k, v) in D50_XYZ.iter().enumerate() { + out.get_mut(68 + 4 * k..72 + 4 * k)?.copy_from_slice(&s15f16(*v)); + } + out.extend_from_slice(&(tags.len() as u32).to_be_bytes()); + for ((sig, bytes), at) in tags.iter().zip(&offsets) { + out.extend_from_slice(*sig); + out.extend_from_slice(&(*at as u32).to_be_bytes()); + out.extend_from_slice(&(bytes.len() as u32).to_be_bytes()); + } + out.extend_from_slice(&data); + Some(out) +} + +#[cfg(test)] +mod tests { + #![allow(clippy::unwrap_used)] + use super::*; + + fn params() -> MatrixProfile<'static> { + MatrixProfile { + matrix: [[0.4361, 0.3851, 0.1431], [0.2225, 0.7169, 0.0606], [0.0139, 0.0971, 0.7141]], + curve: ToneCurve::Linear, + description: "v2", + copyright: "free", + date: [2026, 1, 1, 0, 0, 0], + } + } + + #[test] + fn valid_profile_parses_and_aliases_curves() { + for curve in [ToneCurve::Linear, ToneCurve::Samples(&[0, 32768, 65535])] { + let bytes = matrix_profile(&MatrixProfile { curve, ..params() }).unwrap(); + let p = crate::Profile::parse(&bytes).unwrap(); + assert_eq!(p.version, (2, 0x10)); + assert_eq!(p.description, "v2"); + assert_eq!(&bytes[24..28], &[7, 234, 0, 1]); + assert_eq!(&bytes[208..216], &bytes[220..228]); + assert_eq!(&bytes[208..216], &bytes[232..240]); + assert_eq!(bytes, matrix_profile(&MatrixProfile { curve, ..params() }).unwrap()); + } + } + + #[test] + fn bounded_inputs_reject_before_serializing() { + let long = "x".repeat(1025); + for description in [long.as_str(), "bad\0name", "non-ascii Ă©"] { + assert!(matrix_profile(&MatrixProfile { description, ..params() }).is_none()); + } + for samples in [&[][..], &[0][..], &vec![0; 4097][..]] { + assert!(matrix_profile(&MatrixProfile { curve: ToneCurve::Samples(samples), ..params() }).is_none()); + } + for value in [f64::NAN, f64::INFINITY, f64::NEG_INFINITY] { + let mut p = params(); + p.matrix[0][0] = value; + assert!(matrix_profile(&p).is_none()); + } + let samples = vec![0; 4096]; + let name = "x".repeat(1024); + let bytes = matrix_profile(&MatrixProfile { description: &name, copyright: &name, curve: ToneCurve::Samples(&samples), ..params() }).unwrap(); + assert!(bytes.len() < 12 * 1024); + } +} diff --git a/crates/cms/tests/sony_lut.rs b/crates/cms/tests/sony_lut.rs new file mode 100644 index 000000000..065792a27 --- /dev/null +++ b/crates/cms/tests/sony_lut.rs @@ -0,0 +1,46 @@ +use craft_cms::sony_lut::{MAX_SIDE, parse_spi3d, parse_spimtx}; + +#[test] +fn spimtx_is_finite_exact_and_scales_offsets() { + let m = parse_spimtx("2 0 0 6553.5\n0 3 0 -65535\n0 0 4 0").unwrap(); + assert_eq!(m.matrix, [2.0, 0.0, 0.0, 0.0, 0.0, 3.0, 0.0, 0.0, 0.0, 0.0, 4.0, 0.0, 0.0, 0.0, 0.0, 1.0]); + assert_eq!(m.offset, [0.1, -1.0, 0.0, 0.0]); + for bad in ["", "1 0 0 0", "NaN 0 0 0 0 1 0 0 0 0 1 0", "inf 0 0 0 0 1 0 0 0 0 1 0", "1 0 0 0 0 1 0 0 0 0 1 0 9"] { + assert!(parse_spimtx(bad).is_none()); + } +} + +fn table() -> String { + let mut text = String::from("SPILUT 1.0\n3 3\n2 2 2\n"); + // Reverse row order to prove coordinates, rather than row order, determine storage. + for i in (0..8).rev() { + let (r, g, b) = (i & 1, (i >> 1) & 1, i >> 2); + text.push_str(&format!("{r} {g} {b} {r} {g} {b}\n")); + } + text +} + +#[test] +fn spi3d_red_fast_coordinates_and_limits() { + let text = table(); + let lut = parse_spi3d(&text).unwrap(); + assert_eq!(lut.size, 2); + for (i, row) in lut.values.iter().enumerate() { + assert_eq!(*row, [(i & 1) as f32, ((i >> 1) & 1) as f32, (i >> 2) as f32]); + } + assert!(parse_spi3d(&format!("{text}0 0 0 0 0 0\n")).is_none()); + assert!(parse_spi3d("SPILUT 1.0\n3 3\n2 2 2\n0 0 0 0 0 0\n").is_none()); + assert!(parse_spi3d(&format!("SPILUT 1.0\n3 3\n{0} {0} {0}\n", MAX_SIDE + 1)).is_none()); + assert!(parse_spi3d(&text.replace("2 2 2", "2 2 3")).is_none()); + assert!(parse_spi3d(&text.replace("0 0 0 0 0 0", "2 0 0 0 0 0")).is_none()); + assert!(parse_spi3d(&text.replace("0 0 0 0 0 0", "0 0 0 0 0 0 extra")).is_none()); +} + +#[test] +fn inherited_tolerance_is_explicit() { + let text = "SPILUT alternative\nignored\n2 2 2\n".to_owned() + &"0 0 0 -1 2 NaN\n".repeat(8); + let lut = parse_spi3d(&text).unwrap(); + assert_eq!(&lut.values[0][..2], &[-1.0, 2.0]); + assert!(lut.values[0][2].is_nan()); + assert_eq!(lut.values[1], [0.0; 3]); +} diff --git a/crates/cms/tests/spi1d.rs b/crates/cms/tests/spi1d.rs new file mode 100644 index 000000000..f4e27d051 --- /dev/null +++ b/crates/cms/tests/spi1d.rs @@ -0,0 +1,32 @@ +use craft_cms::spi1d::{MAX_ENTRIES, Table, parse}; + +#[test] +fn monochrome_and_three_component_samples_preserve_domain() { + assert_eq!( + parse("Version 1\nFrom -1 2\nLength 2\nComponents 1\n{\n-2\n3\n}"), + Some(Table { values: vec![[-2.0; 3], [3.0; 3]], domain: [-1.0, 2.0] }) + ); + assert_eq!( + parse("Length 2\nComponents 3\n{\n0 1\n2 3 4 5\n}"), + Some(Table { values: vec![[0.0, 1.0, 2.0], [3.0, 4.0, 5.0]], domain: [0.0, 1.0] }) + ); +} + +#[test] +fn legacy_header_and_eof_tolerance_is_preserved() { + let table = parse("Ignored arbitrary header\nLength 2 extra\n{\n0 1"); + assert!(table.is_some()); + assert_eq!(table, parse("Length 2\n{\n0 1\n} ignored\ninvalid trailing text")); + assert!(parse("Length 2\n{\nNaN inf\n}").is_some(), "sample sanitization belongs to the caller"); +} + +#[test] +fn invalid_headers_partial_and_surplus_tables_are_refused() { + for header in ["Length 1", "Length 2\nComponents 2", "Length 2\nFrom 2 1", "Length 2\nFrom NaN 1"] { + assert!(parse(&format!("{header}\n{{\n0 1\n}}")).is_none()); + } + assert!(parse(&format!("Length {}\n{{\n0 1", MAX_ENTRIES + 1)).is_none()); + assert!(parse("Length 2\nComponents 3\n{\n0 1 2 3 4").is_none()); + assert!(parse("Length 2\n{\n0 1 2").is_none()); + assert!(parse("Length 2\n{\n").is_none()); +} diff --git a/crates/control-protocol/Cargo.toml b/crates/control-protocol/Cargo.toml new file mode 100644 index 000000000..f00bbc1c6 --- /dev/null +++ b/crates/control-protocol/Cargo.toml @@ -0,0 +1,15 @@ +[package] +name = "craft-control-protocol" +description = "Authenticated bounded JSON-lines control protocol" +version.workspace = true +edition.workspace = true +license.workspace = true +rust-version.workspace = true +repository.workspace = true + +[lints] +workspace = true + +[dependencies] +serde_json = { workspace = true } +subtle = "2" diff --git a/crates/control-protocol/src/lib.rs b/crates/control-protocol/src/lib.rs new file mode 100644 index 000000000..8bdb08d66 --- /dev/null +++ b/crates/control-protocol/src/lib.rs @@ -0,0 +1,234 @@ +//! Authenticated JSON-lines request framing and dispatch for local control services. +#![cfg_attr(not(test), deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable))] +use serde_json::{Value, json}; +use std::io::{BufRead, BufReader, Read, Write}; +use subtle::ConstantTimeEq; + +/// Whether a session token has the control protocol's 64 hexadecimal character form. +pub fn valid_token(token: &str) -> bool { + token.len() == 64 && token.bytes().all(|b| b.is_ascii_hexdigit()) +} + +/// Encode one client request, including its terminating newline. +pub fn request_line(id: u64, token: Option<&str>, method: &str, params: &Value) -> String { + format!("{}\n", json!({"id": id, "token": token, "method": method, "params": params})) +} + +/// Extract the result or application error from a parsed control reply. +pub fn reply_result(reply: &Value) -> Result { + if reply.get("ok").and_then(Value::as_bool) == Some(true) { + Ok(reply.get("result").cloned().unwrap_or(Value::Null)) + } else { + Err(reply.get("error").and_then(Value::as_str).unwrap_or("error").to_owned()) + } +} + +/// One parsed line. +enum Line { + Request { + id: Value, + method: String, + params: Value, + }, + Blank, + /// Not a request: reply with this error and close the connection. + Reject(String), +} + +fn parse_line(bytes: &[u8], token: &str) -> Line { + let Ok(text) = std::str::from_utf8(bytes) else { + return Line::Reject("not a control request (invalid UTF-8); closing the connection".into()); + }; + let text = text.trim(); + if text.is_empty() { + return Line::Blank; + } + let msg = match serde_json::from_str::(text) { + Ok(v) => v, + Err(e) => return Line::Reject(format!("bad JSON ({e}); expected one JSON request object per line; closing the connection")), + }; + let Some(obj) = msg.as_object() else { + return Line::Reject("not a control request (expected a JSON object); closing the connection".into()); + }; + let Some(method) = obj.get("method").and_then(Value::as_str) else { + return Line::Reject("not a control request (missing string `method`); closing the connection".into()); + }; + // Authenticate every request before constructing or enqueueing a ControlRequest. A valid + // earlier request must not authorize a later one with a missing or incorrect credential. + let authenticated = obj.get("token").and_then(Value::as_str).is_some_and(|supplied| bool::from(supplied.as_bytes().ct_eq(token.as_bytes()))); + if !authenticated { + return Line::Reject("control authentication failed; closing the connection".into()); + } + Line::Request { + id: obj.get("id").cloned().unwrap_or(Value::Null), + method: method.to_string(), + params: obj.get("params").cloned().unwrap_or(json!({})), + } +} + +#[derive(Debug, PartialEq, Eq)] +pub enum ReadFailure { + Timeout, + Rejected(String), +} + +/// Read one `\n`-terminated line of at most `max` bytes into `buf` (newline stripped). +/// `Ok(false)` at end of stream; `Err` when the line is too long or the read fails. +pub fn read_bounded_line(reader: &mut impl BufRead, buf: &mut Vec, max: usize) -> Result { + buf.clear(); + // `take` bounds the bytes buffered for one line: a peer can't make us allocate more than `max`. + let limit = u64::try_from(max).unwrap_or(u64::MAX).saturating_add(1); + let n = reader.take(limit).read_until(b'\n', buf).map_err(|e| match e.kind() { + std::io::ErrorKind::TimedOut | std::io::ErrorKind::WouldBlock => ReadFailure::Timeout, + _ => ReadFailure::Rejected(e.to_string()), + })?; + if n == 0 { + return Ok(false); + } + if buf.last() == Some(&b'\n') { + buf.pop(); + if buf.last() == Some(&b'\r') { + buf.pop(); + } + return Ok(true); + } + if buf.len() > max { + return Err(ReadFailure::Rejected(format!("request line longer than {max} bytes; closing the connection"))); + } + // Last line without a newline, then end of stream. + Ok(true) +} + +/// Serve requests from `read`, answering on `out`, until the peer closes, sends something that is +/// not a request, or `handle` returns `None` (app gone). `handle` gets `(method, params)` and +/// returns the reply object (the request's `id` is added here). +pub fn serve_lines(read: R, mut out: W, token: &str, max_line: usize, mut handle: impl FnMut(String, Value) -> Option) { + let mut reader = BufReader::new(read); + let mut buf = Vec::new(); + loop { + let reject = match read_bounded_line(&mut reader, &mut buf, max_line) { + Ok(false) => return, + // Expiration is not a request: a pooled client must see EOF and reconnect, rather + // than read a stale error reply. A partial line must not be parsed on timeout. + Err(ReadFailure::Timeout) => return, + Err(ReadFailure::Rejected(e)) => e, + Ok(true) => match parse_line(&buf, token) { + Line::Blank => continue, + Line::Reject(e) => e, + Line::Request { id, method, params } => { + let Some(mut r) = handle(method, params) else { return }; + if let Some(o) = r.as_object_mut() { + o.insert("id".into(), id); + } + if writeln!(out, "{r}").and_then(|()| out.flush()).is_err() { + return; + } + continue; + } + }, + }; + let _ = writeln!(out, "{}", json!({"ok": false, "error": reject})).and_then(|()| out.flush()); + return; + } +} + +#[cfg(test)] +mod tests { + use super::*; + #[test] + fn rejects_bad_request_shapes_without_dispatching() { + for input in [&b"\xff\n"[..], &b"[]\n"[..], &b"{}\n"[..]] { + let mut out = Vec::new(); + serve_lines(input, &mut out, "secret", 1024, |_, _| panic!("must not dispatch")); + let reply: Value = serde_json::from_slice(&out).unwrap(); + assert_eq!(reply["ok"], false); + assert!(reply["error"].as_str().unwrap().contains("closing the connection")); + } + } + #[test] + fn timeout_after_partial_line_never_dispatches() { + struct Partial(bool); + impl Read for Partial { + fn read(&mut self, buf: &mut [u8]) -> std::io::Result { + if !self.0 { + self.0 = true; + buf[..2].copy_from_slice(b"{\""); + return Ok(2); + } + Err(std::io::ErrorKind::TimedOut.into()) + } + } + let mut out = Vec::new(); + serve_lines(Partial(false), &mut out, "secret", 1024, |_, _| panic!("must not dispatch")); + assert!(out.is_empty()); + } + #[test] + fn line_boundaries_and_closed_dispatch_are_preserved() { + let mut buf = Vec::new(); + assert_eq!(read_bounded_line(&mut &b"abc\r\n"[..], &mut buf, 4), Ok(true)); + assert_eq!(buf, b"abc"); + assert_eq!(read_bounded_line(&mut &b"abcd\n"[..], &mut buf, 4), Ok(true)); + assert_eq!(buf, b"abcd"); + let mut out = Vec::new(); + serve_lines(&b"{\"method\":\"x\",\"token\":\"s\"}\n"[..], &mut out, "s", 1024, |_, _| None); + assert!(out.is_empty()); + } + #[test] + fn output_failure_stops_following_requests() { + struct Closed; + impl Write for Closed { + fn write(&mut self, _: &[u8]) -> std::io::Result { + Err(std::io::ErrorKind::BrokenPipe.into()) + } + fn flush(&mut self) -> std::io::Result<()> { + Ok(()) + } + } + let input = b"{\"method\":\"x\",\"token\":\"s\"}\n{\"method\":\"y\",\"token\":\"s\"}\n"; + let mut calls = 0; + serve_lines(&input[..], Closed, "s", 1024, |_, _| { + calls += 1; + Some(json!({"ok": true})) + }); + assert_eq!(calls, 1); + } + #[test] + fn client_request_and_reply_keep_optional_fields() { + let line = request_line(7, None, "status", &json!({"a": 1})); + assert_eq!(serde_json::from_str::(&line).unwrap(), json!({"id": 7, "token": null, "method": "status", "params": {"a": 1}})); + assert_eq!(reply_result(&json!({"ok": true})).unwrap(), Value::Null); + assert_eq!(reply_result(&json!({"ok": false, "error": "busy"})).unwrap_err(), "busy"); + } + #[test] + fn validates_session_token_shape() { + assert!(valid_token(&"aB09".repeat(16))); + assert!(!valid_token(&"a".repeat(63))); + assert!(!valid_token(&"g".repeat(64))); + } + #[test] + fn authenticates_each_request_and_closes_after_failure() { + let input = + b"{\"id\":1,\"method\":\"first\",\"token\":\"secret\"}\n{\"id\":2,\"method\":\"second\"}\n{\"method\":\"third\",\"token\":\"secret\"}\n"; + let mut output = Vec::new(); + let mut calls = Vec::new(); + serve_lines(&input[..], &mut output, "secret", 4096, |method, _| { + calls.push(method); + Some(json!({"ok": true})) + }); + assert_eq!(calls, ["first"]); + let lines: Vec = std::str::from_utf8(&output).unwrap().lines().map(|s| serde_json::from_str(s).unwrap()).collect(); + assert_eq!(lines[0], json!({"ok": true, "id": 1})); + assert_eq!(lines[1]["error"], "control authentication failed; closing the connection"); + } + #[test] + fn bounds_requests_and_preserves_last_unterminated_line() { + let mut buf = Vec::new(); + assert_eq!(read_bounded_line(&mut &b"abc"[..], &mut buf, 3), Ok(true)); + assert_eq!(buf, b"abc"); + assert!(matches!(read_bounded_line(&mut &b"abcd"[..], &mut buf, 3), Err(ReadFailure::Rejected(_)))); + let mut out = Vec::new(); + serve_lines(&b"{broken}\n"[..], &mut out, "secret", 1024, |_, _| panic!("must not dispatch")); + let reply: Value = serde_json::from_slice(&out).unwrap(); + assert!(reply["error"].as_str().unwrap().starts_with("bad JSON (")); + } +} diff --git a/crates/csv/README.md b/crates/csv/README.md index c677ec9eb..6369ce8b4 100644 --- a/crates/csv/README.md +++ b/crates/csv/README.md @@ -11,7 +11,7 @@ mapping a SYLK sheet onto a document are the caller's work. | Function | What it does | |---|---| | `decode_text(bytes)` | UTF-8 and UTF-16 byte order marks; without one, valid UTF-8 is kept; other bytes are Windows-1256 when they read as Arabic, else Windows-1252 | -| `sniff_delimiter(text, quote)` | comma, semicolon or tab: the one with the most consistent count per line over the first 20 non-blank lines, outside quotes | +| `sniff_delimiter(text, quote)` | comma, semicolon or tab: the one with the most consistent field count over the first 20 non-blank records in a 64 KiB sample; quote boundaries must be coherent, and ties prefer tab, comma, then semicolon | | `parse_records(text, delim, quote, max_fields, on_record)` | records of fields, one callback per record (return `false` to stop); CR, LF and CRLF line ends; quoted fields with doubled quotes and line breaks; a quote in the middle of a field is text | | `records(text, delim, quote, max_records, max_fields)` | the same, collected | | `push_field(out, field, delim)`, `field_len(field, delim)` | one field as written, quoted only when it holds the delimiter, a quote or a line break; its exact length, to enforce a size cap before writing | diff --git a/crates/csv/src/decode.rs b/crates/csv/src/decode.rs index 238f7f18e..63b4bb590 100644 --- a/crates/csv/src/decode.rs +++ b/crates/csv/src/decode.rs @@ -19,6 +19,16 @@ const CP1256: [u32; 128] = [ 0x200F, 0x06D2, ]; +/// Decode strictly as UTF-8, stripping a leading UTF-8 BOM when present. +/// +/// Unlike [`decode_text`], this does not guess a legacy encoding or accept UTF-16. It is useful +/// when a format or application contract requires UTF-8 and malformed bytes must be rejected. +pub fn decode_utf8(bytes: &[u8]) -> Result { + let bytes = bytes.strip_prefix(&[0xEF, 0xBB, 0xBF]).unwrap_or(bytes); + let text = std::str::from_utf8(bytes)?; + Ok(text.strip_prefix('\u{FEFF}').unwrap_or(text).to_string()) +} + /// Evidence that bytes which aren't UTF-8 are Windows-1256 Arabic rather than Windows-1252. #[derive(Default)] struct Cp1256Evidence { @@ -165,6 +175,16 @@ mod tests { assert_eq!(decode_text(&[0xFF, 0xFE, 0x00, 0xD8]), "\u{FFFD}", "a lone surrogate"); } + #[test] + fn strict_utf8_strips_bom_and_rejects_legacy_or_malformed_bytes() { + assert_eq!(decode_utf8(b"\xEF\xBB\xBFName,City\nAda,Town\n").unwrap(), "Name,City\nAda,Town\n"); + assert_eq!(decode_utf8("cafĂ©".as_bytes()).unwrap(), "cafĂ©"); + assert!(decode_utf8(&[0xFF, 0xFE, 0x00]).is_err()); + assert!(decode_utf8(b"\xEF\xBB\xBF\xFF").is_err()); + // The permissive text decoder still accepts legacy encodings for callers that want it. + assert_eq!(decode_text(&[0x80, b'a', 0xE9]), "€aĂ©"); + } + #[test] fn windows_1256_table_matches_the_unicode_mapping() { let at = |b: u8| CP1256.get(usize::from(b - 0x80)).and_then(|&c| char::from_u32(c)); diff --git a/crates/csv/src/lib.rs b/crates/csv/src/lib.rs index a773f066f..0aec971ac 100644 --- a/crates/csv/src/lib.rs +++ b/crates/csv/src/lib.rs @@ -3,6 +3,7 @@ //! //! - [`decode_text`] turns a file's bytes into a string: byte order marks, UTF-8, else //! Windows-1256 when the bytes read as Arabic and Windows-1252 otherwise. +//! - [`decode_utf8`] requires valid UTF-8 and strips a leading UTF-8 byte order mark. //! - [`sniff_delimiter`] picks comma, semicolon or tab. //! - [`parse_records`] splits text into records of fields (RFC 4180 with the usual leniency), //! handing each record to a callback, with a cap on the fields kept per record. [`records`] @@ -27,46 +28,98 @@ mod decode; pub mod sylk; -pub use decode::decode_text; +pub use decode::{decode_text, decode_utf8}; -/// Picks comma, semicolon or tab: the candidate with the most consistent count per line -/// (outside quotes) over the first non-blank lines. A candidate must appear on every such line, otherwise comma wins. +/// Picks comma, semicolon or tab by consistent field counts in the first non-blank records. +/// Quotes start only at a candidate field boundary; candidates with incoherent quoted fields +/// are discarded when another coherent candidate separates the records. Ties prefer tab, +/// then comma, then semicolon. Comma is the fallback when no candidate separates the text. pub fn sniff_delimiter(text: &str, quote: char) -> u8 { - let mut best = (b',', 0usize, 0usize); - for d in [',', ';', '\t'] { - let mut counts = vec![]; - let mut n = 0usize; - let mut in_q = false; - let mut blank = true; - for ch in text.chars().take(64 * 1024) { - if !in_q && ch == '\n' { - // Blank lines carry no delimiter evidence. - if !blank { - counts.push(n); - } - n = 0; - blank = true; - if counts.len() >= 20 { - break; + let mut candidates = [(b'\t', 0usize, 0usize, true), (b',', 0, 0, true), (b';', 0, 0, true)]; + const SAMPLE_BYTES: usize = 64 * 1024; + let mut sample_end = text.len().min(SAMPLE_BYTES); + while !text.is_char_boundary(sample_end) { + sample_end -= 1; + } + let sample = &text[..sample_end]; + let truncated = sample_end < text.len(); + for (delimiter, min_count, total, coherent) in &mut candidates { + let mut counts = Vec::new(); + let mut count = 0usize; + let mut quoted = false; + let mut after_quote = false; + let mut field_start = true; + let mut record_has_content = false; + let mut record_coherent = true; + let mut candidate_coherent = true; + let mut reached_record_limit = false; + let mut chars = sample.chars().peekable(); + while let Some(ch) = chars.next() { + if quoted { + if ch == quote { + if chars.peek() == Some("e) { + chars.next(); + record_has_content = true; + } else { + quoted = false; + after_quote = true; + } + } else if !ch.is_whitespace() { + record_has_content = true; } continue; } - if ch != '\r' { - blank = false; + if ch == '\r' || ch == '\n' { + if ch == '\r' && chars.peek() == Some(&'\n') { + chars.next(); + } + if record_has_content { + counts.push(count); + candidate_coherent &= record_coherent; + if counts.len() >= 20 { + reached_record_limit = true; + break; + } + } + count = 0; + quoted = false; + after_quote = false; + field_start = true; + record_has_content = false; + record_coherent = true; + continue; } - if ch == quote { - in_q = !in_q; - } else if !in_q && ch == d { - n += 1; + if ch == quote && field_start { + quoted = true; + field_start = false; + record_has_content = true; + } else if ch == char::from(*delimiter) { + count += 1; + field_start = true; + after_quote = false; + } else { + record_coherent &= !after_quote || ch.is_whitespace(); + record_has_content |= !ch.is_whitespace(); + field_start = false; } } - if !blank { - counts.push(n); + if !reached_record_limit && !truncated { + if quoted { + candidate_coherent = false; + } else if record_has_content { + candidate_coherent &= record_coherent; + counts.push(count); + } } - let min = counts.iter().copied().min().unwrap_or(0); - let total: usize = counts.iter().sum(); - if min > 0 && (min, total) > (best.1, best.2) { - best = (d as u8, min, total); + *coherent &= candidate_coherent; + *min_count = counts.iter().copied().min().unwrap_or(0); + *total = counts.iter().sum(); + } + let prefer_coherent = candidates.iter().any(|(_, min, _, coherent)| *min > 0 && *coherent); + let mut best = (b',', 0usize, 0usize); + for (delimiter, min, total, coherent) in candidates { + if min > 0 && (!prefer_coherent || coherent) && (min, total) > (best.1, best.2) { + best = (delimiter, min, total); } } best.0 @@ -204,6 +257,9 @@ mod tests { assert_eq!(sniff_delimiter("a\tb\n1\t2\n", '"'), b'\t'); assert_eq!(sniff_delimiter("a,b\n\"x;y;z\",2\n", '"'), b','); assert_eq!(sniff_delimiter("single", '"'), b','); + assert_eq!(sniff_delimiter("a,b\tc\n1,2\td\n", '"'), b'\t'); + assert_eq!(sniff_delimiter("a,b;c\n1,2;c\n", '"'), b','); + assert_eq!(sniff_delimiter("a;b\tc\n1;2\td\n", '"'), b'\t'); // A delimiter missing from some lines is text, not a separator (single-column export round trip). assert_eq!(sniff_delimiter("alpha;beta\r\ngamma\r\n", '"'), b','); assert_eq!(sniff_delimiter("alpha\tbeta\r\ngamma\r\n", '"'), b','); @@ -211,6 +267,48 @@ mod tests { assert_eq!(sniff_delimiter("x;y\n1,5;2", '"'), b';'); } + #[test] + fn delimiter_sniffing_preserves_candidate_quote_boundaries() { + for ending in ["\n", "\r\n", "\r"] { + for (text, expected) in [ + (format!("Name,\"preferred\tCity\"{ending}Ada\tLondon{ending}"), b'\t'), + (format!("\"Last, First\";City{ending}Ada;London{ending}"), b';'), + (format!("\"Name\";\"Last, First\"{ending}\"Ada\";\"Lovelace, Countess\"{ending}\"Grace\";\"Hopper, Admiral\"{ending}"), b';'), + (format!("\"Name\"suffix,Title\tCity{ending}\"Ada\"suffix,Analyst\tLondon\tignored{ending}"), b'\t'), + (format!("\r\n \r\n\"Name\";City{ending}Ada;London{ending}"), b';'), + ] { + assert_eq!(sniff_delimiter(&text, '"'), expected, "{text:?}"); + } + } + } + + #[test] + fn delimiter_sniffing_counts_empty_quoted_fields_and_ignores_whitespace_records() { + assert_eq!(sniff_delimiter("\"\";name\n\"\";Ada\n", '"'), b';'); + + let blank_rows = "\t \t\n".repeat(25); + assert_eq!(sniff_delimiter(&format!("{blank_rows}a;b\nc;d\n"), '"'), b';'); + } + + #[test] + fn delimiter_sniffing_does_not_count_the_twentieth_record_twice() { + let mut text = "a;b\tc\td\n".repeat(19); + text.push('a'); + text.push_str(&";".repeat(19)); + text.push_str("\tc\n"); + assert_eq!(sniff_delimiter(&text, '"'), b'\t'); + } + + #[test] + fn delimiter_sniffing_discards_an_incomplete_record_at_the_sample_boundary() { + let complete = "a\tb\tc\n1\t2\t3\n"; + let remaining = 64 * 1024 - complete.len(); + let partial = format!("{}x\t\"unfinished", "z".repeat(remaining - 4)); + assert!(partial.len() >= remaining); + let text = format!("{complete}{partial}\"\n"); + assert_eq!(sniff_delimiter(&text, '"'), b'\t'); + } + #[test] fn parser_stops_collecting_fields_at_the_cap() { let (max, extra) = (16_384usize, 50); diff --git a/crates/dav/src/lib.rs b/crates/dav/src/lib.rs index 87040feae..db137a050 100644 --- a/crates/dav/src/lib.rs +++ b/crates/dav/src/lib.rs @@ -74,7 +74,14 @@ pub fn parse_multistatus(bytes: &[u8]) -> Result { if !text.chars().all(craft_xml::is_xml_char) { return Err(Error::Characters); } - let limits = Limits { max_input: MAX_BYTES, max_depth: MAX_DEPTH, max_nodes: MAX_NODES, max_attrs: MAX_ATTRS, max_attrs_total: MAX_ATTRS_TOTAL }; + let limits = Limits { + max_input: MAX_BYTES, + max_depth: MAX_DEPTH, + max_nodes: MAX_NODES, + max_elements: MAX_NODES, + max_attrs: MAX_ATTRS, + max_attrs_total: MAX_ATTRS_TOTAL, + }; let options = Options { doctype: Doctype::Reject, ..Options::canonical(&CANONICAL).strict().with_limits(limits) }; let root = options.parse(bytes).map_err(Error::Xml)?; if root.name != "d:multistatus" { diff --git a/crates/deckcraft-format/Cargo.toml b/crates/deckcraft-format/Cargo.toml new file mode 100644 index 000000000..d7c6b035f --- /dev/null +++ b/crates/deckcraft-format/Cargo.toml @@ -0,0 +1,18 @@ +[package] +name = "craft-deckcraft-format" +description = "The DeckCraft .deckcraft package: identity, manifest envelope, NaN-safe JSON, member names and bounded reads" +version.workspace = true +edition.workspace = true +license.workspace = true +rust-version.workspace = true +repository.workspace = true + +[lints] +workspace = true + +[dependencies] +craft-zip = { workspace = true } +serde = { workspace = true } +serde_json = { workspace = true } +thiserror = { workspace = true } +zip = { version = "9", default-features = false, features = ["deflate-flate2"] } diff --git a/crates/deckcraft-format/README.md b/crates/deckcraft-format/README.md new file mode 100644 index 000000000..94f5d90e8 --- /dev/null +++ b/crates/deckcraft-format/README.md @@ -0,0 +1,25 @@ +# craft-deckcraft-format + +The DeckCraft `.deckcraft` package. A file is a ZIP that holds, in this order: + +- `mimetype`, stored: `application/vnd.storyteller.deckcraft+zip`. Files from before the + SlideCraft rename carry the legacy type, and `sniff` accepts both. +- `presentation.json`, deflated: `{"format":"deckcraft","version":1,"generator":…,"presentation":…}`. + It is compact JSON, and non-finite numbers are written as `0` (`Finite`). +- `media/` and `chart-parts/` binary members. A member whose content type is already + compressed (JPEG, PNG, video, MP3/MP4 audio, anything `zip`) is stored, and any other member is deflated. + +The presentation is a type parameter: DeckCraft owns the model, decides which media and chart +parts become members, and repairs a loaded deck. This crate owns the bytes: + +- `write(presentation, generator, members)` writes the package with `craft-zip`'s deterministic writer. +- `Package::open_manifest(bytes, max)` reads and checks the manifest. A file with no + `presentation.json` or a different `format` is `NotOurs`. `parse_manifest` reads files written + before `Finite`, where serde_json saved NaN as `null`, as 0 within fixed patch and re-parse budgets. +- `Package::read_max` refuses a member that declares more than its limit and stops one that + inflates past it. `read_charged` charges a shared budget for every byte it inflates, including + the bytes of a read that then fails its CRC. +- `Budget` and the `MAX_*` constants are the format's limits: 2 GiB per member, 4 GiB of media and + chart parts, 100,000 items and a 256 MiB manifest. + +There is no `unsafe` and no panics, and it builds for `wasm32-unknown-unknown`. diff --git a/crates/deckcraft-format/src/finite.rs b/crates/deckcraft-format/src/finite.rs new file mode 100644 index 000000000..834108a33 --- /dev/null +++ b/crates/deckcraft-format/src/finite.rs @@ -0,0 +1,185 @@ +//! A serde serializer adapter that writes non-finite floats as 0. +//! +//! serde_json writes NaN and ±infinity as `null`, which doesn't read back as a number: a single +//! NaN anywhere in the model (a rotation, a crop, a stroke width) made a saved file fail to open +//! as "damaged". [`Finite`] wraps any `Serialize` value and replaces those floats on the way out; +//! everything else passes through unchanged. + +use serde::ser::{self, Serialize, Serializer}; + +/// `T`, serialized with non-finite floats written as 0. +pub struct Finite<'a, T: ?Sized>(pub &'a T); + +impl Serialize for Finite<'_, T> { + fn serialize(&self, s: S) -> Result { + self.0.serialize(Wrap(s)) + } +} + +struct Wrap(S); + +macro_rules! pass { + ($($m:ident($t:ty)),* $(,)?) => { + $(fn $m(self, v: $t) -> Result { self.0.$m(v) })* + }; +} + +impl Serializer for Wrap { + type Ok = S::Ok; + type Error = S::Error; + type SerializeSeq = Wrap; + type SerializeTuple = Wrap; + type SerializeTupleStruct = Wrap; + type SerializeTupleVariant = Wrap; + type SerializeMap = Wrap; + type SerializeStruct = Wrap; + type SerializeStructVariant = Wrap; + + pass!( + serialize_bool(bool), + serialize_i8(i8), + serialize_i16(i16), + serialize_i32(i32), + serialize_i64(i64), + serialize_i128(i128), + serialize_u8(u8), + serialize_u16(u16), + serialize_u32(u32), + serialize_u64(u64), + serialize_u128(u128), + serialize_char(char), + serialize_str(&str), + serialize_bytes(&[u8]), + ); + + fn serialize_f32(self, v: f32) -> Result { + self.0.serialize_f32(if v.is_finite() { v } else { 0.0 }) + } + fn serialize_f64(self, v: f64) -> Result { + self.0.serialize_f64(if v.is_finite() { v } else { 0.0 }) + } + fn serialize_none(self) -> Result { + self.0.serialize_none() + } + fn serialize_some(self, v: &T) -> Result { + self.0.serialize_some(&Finite(v)) + } + fn serialize_unit(self) -> Result { + self.0.serialize_unit() + } + fn serialize_unit_struct(self, name: &'static str) -> Result { + self.0.serialize_unit_struct(name) + } + fn serialize_unit_variant(self, name: &'static str, i: u32, variant: &'static str) -> Result { + self.0.serialize_unit_variant(name, i, variant) + } + fn serialize_newtype_struct(self, name: &'static str, v: &T) -> Result { + self.0.serialize_newtype_struct(name, &Finite(v)) + } + fn serialize_newtype_variant(self, name: &'static str, i: u32, variant: &'static str, v: &T) -> Result { + self.0.serialize_newtype_variant(name, i, variant, &Finite(v)) + } + fn serialize_seq(self, len: Option) -> Result { + self.0.serialize_seq(len).map(Wrap) + } + fn serialize_tuple(self, len: usize) -> Result { + self.0.serialize_tuple(len).map(Wrap) + } + fn serialize_tuple_struct(self, name: &'static str, len: usize) -> Result { + self.0.serialize_tuple_struct(name, len).map(Wrap) + } + fn serialize_tuple_variant(self, name: &'static str, i: u32, variant: &'static str, len: usize) -> Result { + self.0.serialize_tuple_variant(name, i, variant, len).map(Wrap) + } + fn serialize_map(self, len: Option) -> Result { + self.0.serialize_map(len).map(Wrap) + } + fn serialize_struct(self, name: &'static str, len: usize) -> Result { + self.0.serialize_struct(name, len).map(Wrap) + } + fn serialize_struct_variant( + self, + name: &'static str, + i: u32, + variant: &'static str, + len: usize, + ) -> Result { + self.0.serialize_struct_variant(name, i, variant, len).map(Wrap) + } + fn is_human_readable(&self) -> bool { + self.0.is_human_readable() + } +} + +macro_rules! compound { + ($tr:ident, $m:ident) => { + impl ser::$tr for Wrap { + type Ok = S::Ok; + type Error = S::Error; + fn $m(&mut self, v: &T) -> Result<(), S::Error> { + self.0.$m(&Finite(v)) + } + fn end(self) -> Result { + self.0.end() + } + } + }; +} +compound!(SerializeSeq, serialize_element); +compound!(SerializeTuple, serialize_element); +compound!(SerializeTupleStruct, serialize_field); +compound!(SerializeTupleVariant, serialize_field); + +impl ser::SerializeMap for Wrap { + type Ok = S::Ok; + type Error = S::Error; + fn serialize_key(&mut self, k: &T) -> Result<(), S::Error> { + self.0.serialize_key(&Finite(k)) + } + fn serialize_value(&mut self, v: &T) -> Result<(), S::Error> { + self.0.serialize_value(&Finite(v)) + } + fn end(self) -> Result { + self.0.end() + } +} + +macro_rules! fields { + ($tr:ident) => { + impl ser::$tr for Wrap { + type Ok = S::Ok; + type Error = S::Error; + fn serialize_field(&mut self, key: &'static str, v: &T) -> Result<(), S::Error> { + self.0.serialize_field(key, &Finite(v)) + } + fn skip_field(&mut self, key: &'static str) -> Result<(), S::Error> { + self.0.skip_field(key) + } + fn end(self) -> Result { + self.0.end() + } + } + }; +} +fields!(SerializeStruct); +fields!(SerializeStructVariant); + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn non_finite_floats_are_written_as_zero_and_nothing_else_changes() { + #[derive(serde::Serialize)] + struct T { + a: f64, + b: Option, + c: Vec<(f32, f64)>, + d: Option, + e: std::collections::BTreeMap, + } + let t = T { a: f64::NAN, b: Some(f64::INFINITY), c: vec![(f32::NEG_INFINITY, 1.5)], d: None, e: [("k".to_string(), f64::NAN)].into() }; + let s = serde_json::to_string(&Finite(&t)).unwrap_or_default(); + assert_eq!(s, r#"{"a":0.0,"b":0.0,"c":[[0.0,1.5]],"d":null,"e":{"k":0.0}}"#); + } +} diff --git a/crates/deckcraft-format/src/lib.rs b/crates/deckcraft-format/src/lib.rs new file mode 100644 index 000000000..b59363c99 --- /dev/null +++ b/crates/deckcraft-format/src/lib.rs @@ -0,0 +1,226 @@ +//! The native `.deckcraft` package: a ZIP holding a stored `mimetype`, `presentation.json` (the +//! manifest: format name, version, generator and the presentation, minus media bytes), and +//! `media/` and `chart-parts/` binary members. Readers ignore unknown fields, so newer +//! files open in older builds with what those builds understand. +//! +//! DeckCraft owns the presentation model and its mapping: which media and chart parts become +//! members, and how a loaded deck is repaired. This crate owns the bytes: the package identity, +//! the manifest envelope, the NaN-safe JSON writer, the reader that opens files with NaN saved as +//! `null`, member names and compression, and the bounded member reads and their budgets. +#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] +#![forbid(unsafe_code)] + +mod finite; +mod manifest; + +#[cfg(test)] +mod tests; + +use std::io::{Cursor, Read}; + +pub use finite::Finite; +pub use manifest::parse_manifest; + +pub const MIMETYPE: &str = "application/vnd.storyteller.deckcraft+zip"; +/// Files written before the SlideCraft → DeckCraft rename (`.slidecraft`) are still read. +pub const LEGACY_MIMETYPE: &str = concat!("application/vnd.storyteller.", "slide", "craft+zip"); +pub const EXTENSION: &str = "deckcraft"; +/// The manifest's `format`. +pub const FORMAT: &str = "deckcraft"; +/// Format version written into the manifest. +pub const VERSION: u32 = 1; +/// The stored member naming the package type. +pub const MIMETYPE_ENTRY: &str = "mimetype"; +/// The manifest member. +pub const MANIFEST_ENTRY: &str = "presentation.json"; +/// Largest entry we will inflate (guards against zip bombs). +pub const MAX_ENTRY: u64 = 2 << 30; +/// All embedded media and chart sidecars together, inflated (storytold/deckcraft#393). +pub const MAX_TOTAL_MEDIA: u64 = 4 << 30; +/// Media items and chart sidecar parts one file may list. +pub const MAX_MEDIA_ITEMS: usize = 100_000; +/// Largest `presentation.json` we will inflate. The manifest holds no media bytes, so it is far +/// smaller than a media entry may be; inflating and parsing one is what dominates memory. +pub const MAX_MANIFEST: u64 = 256 << 20; + +const MEDIA_PREFIX: &str = "media/"; +const CHART_PARTS_PREFIX: &str = "chart-parts/"; + +/// What one load may inflate. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct Budget { + /// Bytes all media and chart members may inflate to together. + pub total: u64, + /// Media items and chart parts the manifest may list. + pub items: usize, + /// Bytes `presentation.json` may inflate to. + pub manifest: u64, +} + +impl Default for Budget { + fn default() -> Self { + Budget { total: MAX_TOTAL_MEDIA, items: MAX_MEDIA_ITEMS, manifest: MAX_MANIFEST } + } +} + +impl Budget { + /// A media budget with the default manifest limit. + pub const fn media(total: u64, items: usize) -> Self { + Budget { total, items, manifest: MAX_MANIFEST } + } +} + +#[derive(Debug, thiserror::Error)] +pub enum FormatError { + #[error("not a DeckCraft file: {0}")] + NotOurs(String), + #[error("damaged file: {0}")] + Damaged(String), + #[error("write failed: {0}")] + Write(String), +} + +pub type Result = std::result::Result; + +/// `presentation.json`: the envelope around the app's presentation. +#[derive(serde::Serialize, serde::Deserialize)] +pub struct Manifest { + pub format: String, + pub version: u32, + pub generator: String, + pub presentation: T, +} + +/// The member holding media item `id`'s bytes. +pub fn media_entry(id: u32) -> String { + format!("{MEDIA_PREFIX}{id}") +} + +/// The member holding the `index`th chart part a save separated out. +pub fn chart_part_entry(index: usize) -> String { + format!("{CHART_PARTS_PREFIX}{index}") +} + +/// Whether `entry` is a name [`chart_part_entry`] gives, exactly: no other member, no path +/// tricks, no leading zeros. +pub fn is_chart_part_entry(entry: &str) -> bool { + entry.strip_prefix(CHART_PARTS_PREFIX).and_then(|s| s.parse::().ok()).is_some_and(|n| entry == chart_part_entry(n)) +} + +/// Whether a member of this content type is stored rather than deflated: it is compressed already. +pub fn is_compressed(content_type: &str) -> bool { + content_type.starts_with("image/jpeg") + || content_type.starts_with("image/png") + || content_type.starts_with("video/") + || content_type.starts_with("audio/mpeg") + || content_type.starts_with("audio/mp4") + || content_type.contains("zip") +} + +/// Does this look like a `.deckcraft` file? +pub fn sniff(bytes: &[u8]) -> bool { + bytes.starts_with(b"PK") && [MIMETYPE, LEGACY_MIMETYPE].iter().any(|m| bytes.windows(m.len()).take(200).any(|w| w == m.as_bytes())) +} + +/// A binary member of a package being written. +#[derive(Clone, Copy, Debug)] +pub struct Member<'a> { + /// Its name, from [`media_entry`] or [`chart_part_entry`]. + pub name: &'a str, + /// Its content type, which decides whether it is stored or deflated. + pub content_type: &'a str, + pub bytes: &'a [u8], +} + +/// The package: `mimetype` (stored), the manifest naming `generator` around `presentation` +/// (deflated, with non-finite numbers written as 0), then `members` in order. +pub fn write(presentation: &T, generator: &str, members: &[Member<'_>]) -> Result> { + use craft_zip::Compression::{Deflated, Stored}; + let w = |e: craft_zip::Error| FormatError::Write(e.to_string()); + let mut z = craft_zip::Writer::new(); + z.add(MIMETYPE_ENTRY, MIMETYPE.as_bytes(), Stored).map_err(w)?; + let m = Manifest { format: FORMAT.to_string(), version: VERSION, generator: generator.to_string(), presentation }; + let json = serde_json::to_vec(&Finite(&m)).map_err(|e| FormatError::Write(e.to_string()))?; + z.add(MANIFEST_ENTRY, &json, Deflated).map_err(w)?; + for member in members { + let compression = if is_compressed(member.content_type) { Stored } else { Deflated }; + z.add(member.name, member.bytes, compression).map_err(w)?; + } + z.finish().map_err(w) +} + +/// A package open for reading. Nothing is inflated until a member is read. +pub struct Package<'a> { + zip: zip::ZipArchive>, +} + +impl<'a> Package<'a> { + /// Index `bytes` as a ZIP. + pub fn open(bytes: &'a [u8]) -> Result { + let zip = zip::ZipArchive::new(Cursor::new(bytes)).map_err(|e| FormatError::NotOurs(e.to_string()))?; + Ok(Package { zip }) + } + + /// Open `bytes` and read its manifest, inflating at most `max_manifest` bytes of it. Missing + /// `presentation.json` or a different `format` is [`FormatError::NotOurs`]. + pub fn open_manifest(bytes: &'a [u8], max_manifest: u64) -> Result<(Manifest, Self)> { + let mut package = Package::open(bytes)?; + if !package.contains(MANIFEST_ENTRY) { + return Err(FormatError::NotOurs(format!("no {MANIFEST_ENTRY}"))); + } + let json = package.read_max(MANIFEST_ENTRY, max_manifest)?; + let m: Manifest = parse_manifest(&json)?; + if m.format != FORMAT { + return Err(FormatError::NotOurs(m.format)); + } + Ok((m, package)) + } + + /// Whether the package has a member named `name`. + pub fn contains(&self, name: &str) -> bool { + self.zip.index_for_name(name).is_some() + } + + /// The size `name` declares, inflated; `None` when there is no such readable member. + pub fn declared_size(&mut self, name: &str) -> Option { + self.zip.by_name(name).ok().map(|f| f.size()) + } + + /// Read one member within [`MAX_ENTRY`]. + pub fn read(&mut self, name: &str) -> Result> { + self.read_max(name, MAX_ENTRY) + } + + /// Read one member, refusing it before inflating if it declares more than `max` bytes, and + /// stopping at `max` if it inflates further than it declared. + pub fn read_max(&mut self, name: &str, max: u64) -> Result> { + self.read_counted(name, max, &mut 0) + } + + /// [`Package::read_max`] within `left` (and [`MAX_ENTRY`]), charging `left` for every byte + /// inflated, including those inflated before a decode or CRC error, so a member that fails + /// late still spends the shared budget it used. + pub fn read_charged(&mut self, name: &str, left: &mut u64) -> Result> { + let mut spent = 0; + let out = self.read_counted(name, (*left).min(MAX_ENTRY), &mut spent); + *left = left.saturating_sub(spent); + out + } + + /// [`Package::read_max`], adding the bytes inflated to `spent` whether or not the read succeeds. + fn read_counted(&mut self, name: &str, max: u64, spent: &mut u64) -> Result> { + let f = self.zip.by_name(name).map_err(|e| FormatError::Damaged(format!("{name}: {e}")))?; + if f.size() > max { + return Err(FormatError::Damaged(format!("{name} is too large"))); + } + let mut out = Vec::with_capacity(usize::try_from(f.size().min(64 << 20)).unwrap_or(0)); + let mut f = f.take(max.saturating_add(1)); + let read = f.read_to_end(&mut out); + *spent = spent.saturating_add(out.len() as u64); + read.map_err(|e| FormatError::Damaged(format!("{name}: {e}")))?; + if out.len() as u64 > max { + return Err(FormatError::Damaged(format!("{name} is too large"))); + } + Ok(out) + } +} diff --git a/crates/deckcraft-format/src/manifest.rs b/crates/deckcraft-format/src/manifest.rs new file mode 100644 index 000000000..03f017ace --- /dev/null +++ b/crates/deckcraft-format/src/manifest.rs @@ -0,0 +1,99 @@ +//! Reading `presentation.json`, including files written with NaN or infinity saved as `null`. + +use serde::de::DeserializeOwned; + +use crate::{FormatError, Manifest, Result}; + +/// Most `null`s standing in for numbers that [`parse_manifest`] patches in one file, and the +/// candidates it tries for each. +const MAX_NULL_FIXES: usize = 256; +const MAX_TRIES_PER_FIX: usize = 32; +/// Bytes [`parse_manifest`] may re-parse in all while patching, so a hostile file full of `null`s +/// can't keep it busy: past this the file is reported damaged. +const MAX_REPARSE_BYTES: usize = 256 * 1024 * 1024; + +/// Where a parse failed, comparable: later is progress. +fn failure(src: &[u8]) -> std::result::Result, (String, (usize, usize))> { + serde_json::from_slice::>(src).map_err(|e| (e.to_string(), (e.line(), e.column()))) +} + +/// Parse `presentation.json`. serde_json writes a NaN or infinite number as `null`, which then +/// fails to read back as a number ("invalid type: null, expected f64"), so DeckCraft could save +/// a file it couldn't open. Each such `null` is read as 0 instead (the app then brings the +/// value into range); anything else that doesn't parse is damage. +/// +/// serde reports the failing `null` exactly, except inside tagged enums (buffered), where it +/// reports the end of the enclosing object: there the `null`s before that point are tried +/// nearest first, and a patch is kept only if the parse then gets further. (Files saved since +/// [`crate::Finite`] never contain such `null`s; this reads older ones.) +pub fn parse_manifest(json: &[u8]) -> Result> { + // Parse the inflated text as it is: only a file that fails on a `null` number is copied for + // patching (the copy is bounded by `MAX_MANIFEST`). + let mut last = match failure(json) { + Ok(m) => return Ok(m), + Err(f) => f, + }; + if !is_null_number(&last.0) { + return Err(FormatError::Damaged(last.0)); + } + let mut src = json.to_vec(); + let mut budget = MAX_REPARSE_BYTES; + 'fixes: for _ in 0..MAX_NULL_FIXES { + let (msg, pos) = &last; + if !is_null_number(msg) { + break; + } + let end = offset(&src, pos.0, pos.1).unwrap_or(src.len()); + let mut progressed = false; + let candidates: Vec = value_nulls_before(&src, end + 4).take(MAX_TRIES_PER_FIX).collect(); + for at in candidates { + let (Some(head), Some(tail)) = (src.get(..at), src.get(at.saturating_add(4)..)) else { continue }; + let Some(left) = budget.checked_sub(src.len()) else { break 'fixes }; + budget = left; + let mut next = Vec::with_capacity(src.len()); + next.extend_from_slice(head); + next.push(b'0'); + next.extend_from_slice(tail); + match failure(&next) { + Ok(m) => return Ok(m), + // The patch shortened the text by 3 bytes before the error: compare positions + // allowing for that. + Err(f) if f.1 > (pos.0, pos.1.saturating_sub(3)) => { + src = next; + last = f; + progressed = true; + break; + } + Err(_) => {} + } + } + if !progressed { + break; + } + } + Err(FormatError::Damaged(last.0)) +} + +/// Is this parse error a `null` where a number belongs (a NaN or infinity serde_json wrote)? +fn is_null_number(msg: &str) -> bool { + msg.starts_with("invalid type: null, expected f") || msg.starts_with("invalid type: null, expected a number") +} + +/// Byte offset of a 1-based `line`:`column`. +fn offset(src: &[u8], line: usize, column: usize) -> Option { + let mut start = 0; + for _ in 1..line { + start += src.get(start..)?.iter().position(|b| *b == b'\n')? + 1; + } + Some((start + column).min(src.len())) +} + +/// Starts of `null` tokens in value position (after `:`, `[` or `,`) before `end`, nearest first. +fn value_nulls_before(src: &[u8], end: usize) -> impl Iterator + '_ { + let end = end.min(src.len()); + (0..end.saturating_sub(3)).rev().filter(move |&i| { + src.get(i..i + 4) == Some(b"null") + && src[..i].iter().rev().find(|b| !b.is_ascii_whitespace()).is_some_and(|b| matches!(b, b':' | b'[' | b',')) + && src.get(i + 4).is_none_or(|b| !b.is_ascii_alphanumeric()) + }) +} diff --git a/crates/deckcraft-format/src/tests.rs b/crates/deckcraft-format/src/tests.rs new file mode 100644 index 000000000..8090d709a --- /dev/null +++ b/crates/deckcraft-format/src/tests.rs @@ -0,0 +1,235 @@ +#![allow(clippy::unwrap_used, clippy::expect_used, clippy::panic)] + +use serde::{Deserialize, Serialize}; + +use super::*; + +/// A stand-in for an app's presentation: plain numbers, an optional number and a tagged enum +/// (whose `null`s serde reports only at the end of the enclosing object). +#[derive(Debug, Default, PartialEq, Serialize, Deserialize)] +struct Deck { + #[serde(default)] + width: f64, + #[serde(default)] + zoom: Option, + #[serde(default)] + shapes: Vec, +} + +#[derive(Debug, PartialEq, Serialize, Deserialize)] +#[serde(tag = "kind")] +enum Shape { + Box { x: f64, y: f64 }, + Line { len: f64 }, +} + +fn package(members: &[Member<'_>]) -> Vec { + write(&Deck { width: 960.0, ..Default::default() }, "Test 1", members).unwrap() +} + +#[test] +fn a_package_holds_mimetype_then_manifest_then_members_in_order() { + let png = [0x89, b'P', b'N', b'G']; + let svg = b"".repeat(20); + let bytes = package(&[ + Member { name: &media_entry(7), content_type: "image/png", bytes: &png }, + Member { name: &media_entry(8), content_type: "image/svg+xml", bytes: &svg }, + Member { name: &chart_part_entry(0), content_type: "application/zip", bytes: b"PK" }, + ]); + assert!(sniff(&bytes)); + let mut z = zip::ZipArchive::new(std::io::Cursor::new(bytes.as_slice())).unwrap(); + assert_eq!(z.len(), 5); + let methods: Vec<_> = (0..z.len()) + .map(|i| { + let f = z.by_index(i).unwrap(); + (f.name().unwrap().into_owned(), f.compression()) + }) + .collect(); + use zip::CompressionMethod::{Deflated, Stored}; + assert_eq!( + methods, + [("mimetype", Stored), ("presentation.json", Deflated), ("media/7", Stored), ("media/8", Deflated), ("chart-parts/0", Stored)] + .map(|(n, m)| (n.to_string(), m)) + ); + // The mimetype is the first, stored member, readable in place. + assert_eq!(craft_zip::stored_mimetype(&bytes), Some(MIMETYPE.as_bytes())); + let (m, mut p) = Package::open_manifest::(&bytes, MAX_MANIFEST).unwrap(); + assert_eq!((m.format.as_str(), m.version, m.generator.as_str(), m.presentation.width), (FORMAT, VERSION, "Test 1", 960.0)); + assert_eq!(p.read(&media_entry(8)).unwrap(), svg); + assert_eq!(p.declared_size(&media_entry(7)), Some(4)); + assert_eq!(p.declared_size("media/99"), None); +} + +#[test] +fn the_manifest_is_compact_json_with_its_fields_in_order() { + let bytes = package(&[]); + let mut p = Package::open(&bytes).unwrap(); + let json = String::from_utf8(p.read(MANIFEST_ENTRY).unwrap()).unwrap(); + assert_eq!(json, r#"{"format":"deckcraft","version":1,"generator":"Test 1","presentation":{"width":960.0,"zoom":null,"shapes":[]}}"#); +} + +#[test] +fn non_finite_numbers_are_written_as_zero_and_read_back() { + let deck = Deck { width: f64::NAN, zoom: Some(f64::INFINITY), shapes: vec![Shape::Box { x: f64::NEG_INFINITY, y: 2.0 }] }; + let bytes = write(&deck, "t", &[]).unwrap(); + let (m, _) = Package::open_manifest::(&bytes, MAX_MANIFEST).unwrap(); + assert_eq!(m.presentation, Deck { width: 0.0, zoom: Some(0.0), shapes: vec![Shape::Box { x: 0.0, y: 2.0 }] }); +} + +/// Files written before [`Finite`] (plain serde_json: NaN as `null`) open too, including a +/// `null` inside a tagged enum, where serde reports only the enclosing object. +#[test] +fn older_manifests_with_null_numbers_parse() { + let cases: [(&str, Deck); 3] = [ + (r#"{"format":"deckcraft","version":1,"generator":"old","presentation":{"width":null}}"#, Deck::default()), + ( + r#"{"format":"deckcraft","version":1,"generator":"old","presentation":{"width":3,"shapes":[{"kind":"Box","x":null,"y":4}]}}"#, + Deck { width: 3.0, shapes: vec![Shape::Box { x: 0.0, y: 4.0 }], ..Default::default() }, + ), + ( + "{\n \"format\": \"deckcraft\",\n \"version\": 1,\n \"generator\": \"t\",\n \"presentation\": {\"shapes\": [{\"kind\": \"Line\", \"len\": null}]}\n}", + Deck { shapes: vec![Shape::Line { len: 0.0 }], ..Default::default() }, + ), + ]; + for (json, want) in cases { + let m = parse_manifest::(json.as_bytes()).unwrap_or_else(|e| panic!("{json}: {e}")); + assert_eq!(m.presentation, want, "{json}"); + } +} + +/// Only `null`s where numbers belong are patched: other damage still fails. +#[test] +fn other_manifest_damage_is_still_reported() { + let ok = br#"{"format": "deckcraft", "version": 1, "generator": "t", "presentation": {}}"#; + assert!(parse_manifest::(ok).is_ok()); + let bad = br#"{"format": null, "version": 1, "generator": "t", "presentation": {}}"#; + assert!(matches!(parse_manifest::(bad), Err(FormatError::Damaged(_)))); + assert!(parse_manifest::(&ok[..ok.len() - 3]).is_err()); +} + +#[test] +fn a_manifest_full_of_nulls_is_refused_within_bounds() { + let shapes = (0..2000).map(|_| r#"{"kind":"Box","x":1,"y":"not a number"}"#).collect::>().join(","); + let json = format!(r#"{{"format":"deckcraft","version":1,"generator":"t","presentation":{{"width":null,"shapes":[{shapes}]}}}}"#); + assert!(matches!(parse_manifest::(json.as_bytes()), Err(FormatError::Damaged(_)))); +} + +#[test] +fn what_is_not_a_package_is_not_ours() { + for bytes in [&b""[..], b"PK\x03\x04garbage", &[0u8; 1000]] { + assert!(matches!(Package::open_manifest::(bytes, MAX_MANIFEST), Err(FormatError::NotOurs(_)))); + } + // A ZIP without the manifest, and one from another app. + let mut z = craft_zip::Writer::new(); + z.add("mimetype", b"application/zip", craft_zip::Compression::Stored).unwrap(); + let no_manifest = z.finish().unwrap(); + let err = Package::open_manifest::(&no_manifest, MAX_MANIFEST).err().unwrap(); + assert_eq!(err.to_string(), "not a DeckCraft file: no presentation.json"); + let mut z = craft_zip::Writer::new(); + z.add(MANIFEST_ENTRY, br#"{"format":"other","version":1,"generator":"x","presentation":{}}"#, craft_zip::Compression::Deflated).unwrap(); + let other = z.finish().unwrap(); + let err = Package::open_manifest::(&other, MAX_MANIFEST).err().unwrap(); + assert_eq!(err.to_string(), "not a DeckCraft file: other"); + // A truncated package is an error, not a panic. + let mut b = package(&[]); + b.truncate(b.len() / 2); + assert!(Package::open_manifest::(&b, MAX_MANIFEST).is_err()); +} + +#[test] +fn legacy_slidecraft_packages_are_recognised() { + let mut z = craft_zip::Writer::new(); + z.add(MIMETYPE_ENTRY, LEGACY_MIMETYPE.as_bytes(), craft_zip::Compression::Stored).unwrap(); + assert!(sniff(&z.finish().unwrap())); + assert!(!sniff(b"PK\x03\x04 application/zip")); + assert!(!sniff(MIMETYPE.as_bytes())); +} + +#[test] +fn an_oversized_manifest_is_refused_before_inflating() { + let bytes = package(&[]); + let size = Package::open(&bytes).unwrap().declared_size(MANIFEST_ENTRY).unwrap(); + assert!(Package::open_manifest::(&bytes, size).is_ok()); + let err = Package::open_manifest::(&bytes, size - 1).err().unwrap(); + assert_eq!(err.to_string(), "damaged file: presentation.json is too large"); +} + +#[test] +fn a_member_that_inflates_past_its_declared_size_stops_at_the_limit() { + // Declared sizes are the file's word: a stored member whose header lies is still capped. + let data = vec![7u8; 100]; + let bytes = package(&[Member { name: &media_entry(1), content_type: "image/png", bytes: &data }]); + let mut p = Package::open(&bytes).unwrap(); + assert_eq!(p.read_max(&media_entry(1), 100).unwrap().len(), 100); + assert!(p.read_max(&media_entry(1), 99).is_err()); + let err = p.read("media/2").unwrap_err(); + assert!(matches!(err, FormatError::Damaged(ref m) if m.starts_with("media/2: ")), "{err}"); +} + +#[test] +fn corrupt_members_are_charged_for_what_they_inflated() { + let a = b"PICTURE-AAAAAAAAAAAA"; + let b = b"PICTURE-BBBBBBBBBBBB"; + let mut bytes = package(&[ + Member { name: &media_entry(1), content_type: "image/png", bytes: a }, + Member { name: &media_entry(2), content_type: "image/png", bytes: b }, + ]); + // Corrupt both stored members, so each fails its CRC only after inflating all its bytes. + for tag in [b"PICTURE-AAAA", b"PICTURE-BBBB"] { + let at = bytes.windows(tag.len()).position(|w| w == tag).unwrap(); + bytes[at] ^= 0xff; + } + let mut p = Package::open(&bytes).unwrap(); + let mut left = 30; + assert!(p.read_charged(&media_entry(1), &mut left).is_err()); + assert_eq!(left, 10, "a failed read still spends what it inflated"); + // The second fits a fresh budget on its own, but not what is left: refused before inflating. + let err = p.read_charged(&media_entry(2), &mut left).unwrap_err(); + assert_eq!(err.to_string(), "damaged file: media/2 is too large"); + assert_eq!(left, 10); +} + +#[test] +fn chart_part_names_are_exact() { + assert_eq!(chart_part_entry(3), "chart-parts/3"); + assert!(is_chart_part_entry("chart-parts/0")); + assert!(is_chart_part_entry("chart-parts/12")); + for not in ["media/1", "chart-parts/../0", "chart-parts/00", "chart-parts/-1", "chart-parts/", "chart-parts/+1", "chart-parts/1 "] { + assert!(!is_chart_part_entry(not), "{not}"); + } +} + +#[test] +fn compressed_content_types_are_stored() { + for stored in [ + "image/jpeg", + "image/png", + "video/mp4", + "audio/mpeg", + "audio/mp4", + "application/zip", + "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet+zip", + ] { + assert!(is_compressed(stored), "{stored}"); + } + for deflated in ["image/svg+xml", "image/gif", "audio/wav", "application/octet-stream", "text/plain"] { + assert!(!is_compressed(deflated), "{deflated}"); + } +} + +#[test] +fn duplicate_members_are_a_write_error() { + let err = write( + &Deck::default(), + "t", + &[Member { name: "media/1", content_type: "x", bytes: b"a" }, Member { name: "media/1", content_type: "x", bytes: b"b" }], + ) + .unwrap_err(); + assert!(matches!(err, FormatError::Write(_)), "{err}"); +} + +#[test] +fn the_default_budget_is_the_formats_limits() { + assert_eq!(Budget::default(), Budget { total: 4 << 30, items: 100_000, manifest: 256 << 20 }); + assert_eq!(Budget::media(10, 2), Budget { total: 10, items: 2, manifest: MAX_MANIFEST }); +} diff --git a/crates/designcraft-format/Cargo.toml b/crates/designcraft-format/Cargo.toml new file mode 100644 index 000000000..19557b7ae --- /dev/null +++ b/crates/designcraft-format/Cargo.toml @@ -0,0 +1,18 @@ +[package] +name = "craft-designcraft-format" +description = "Versioned DesignCraft document JSON codec and legacy migrations over craft-document-bundle" +version.workspace = true +edition.workspace = true +license.workspace = true +rust-version.workspace = true +repository.workspace = true + +[lints] +workspace = true + +[dependencies] +craft-document-bundle = { workspace = true } +craft-eps = { workspace = true } +serde = { workspace = true } +serde_json = { workspace = true } +thiserror = { workspace = true } diff --git a/crates/designcraft-format/src/lib.rs b/crates/designcraft-format/src/lib.rs new file mode 100644 index 000000000..eab431eab --- /dev/null +++ b/crates/designcraft-format/src/lib.rs @@ -0,0 +1,294 @@ +//! Versioned JSON and bundle codec for DesignCraft's native document format. +//! +//! The application owns its document model. This crate owns the on-disk schema version, +//! serialization, legacy JSON upgrades, version inference and the shared ZIP envelope. +#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] +#![forbid(unsafe_code)] + +use std::collections::{HashMap, HashSet}; +use std::io::{Cursor, Read, Seek, Write}; + +pub use craft_document_bundle::Asset; +use craft_document_bundle::Spec; +use serde::{Serialize, de::DeserializeOwned}; +use serde_json::Value; + +pub const MIME: &str = "application/vnd.designcraft+zip"; +pub const VERSION: u32 = 5; +pub const EXTENSION: &str = "designcraft"; +pub const PREVIEW: &str = "preview.png"; +pub const PROXIES_DIRECTORY: &str = "proxies"; +pub const MAX_PREVIEW_BYTES: u64 = 4 << 20; + +const BUNDLE: Spec<'static> = Spec { + mimetype_entry: "mimetype", + mimetype: MIME, + metadata_entry: "meta.json", + document_entry: "document.json", + preview_entry: PREVIEW, + assets_directory: "assets", + proxies_directory: PROXIES_DIRECTORY, + format_name: "designcraft", + legacy_assets_field: Some("assetData"), +}; + +#[derive(Debug, thiserror::Error)] +pub enum Error { + #[error("not a DesignCraft document: {0}")] + NotOurs(String), + #[error("document is from a newer DesignCraft (format {0}); please update")] + TooNew(u64), + #[error("{0}")] + Io(String), + #[error("document cannot be saved: {0}")] + Unsaveable(String), +} + +impl From for Error { + fn from(error: craft_document_bundle::Error) -> Self { + match error { + craft_document_bundle::Error::TooNew(version) => Error::TooNew(version), + craft_document_bundle::Error::NotBundle(message) => Error::NotOurs(message), + craft_document_bundle::Error::Read(message) | craft_document_bundle::Error::Write(message) => Error::Io(message), + } + } +} + +/// A decoded native document and the separately stored embedded assets. +#[derive(Clone, Debug, PartialEq)] +pub struct Loaded { + pub document: T, + pub assets: Vec<(String, Vec)>, + /// Proxy previews (JPEG or PNG bytes) of linked assets, keyed by asset id. + pub proxies: Vec<(String, Vec)>, +} + +/// The reader capability required by `document`, inferred from its neutral JSON schema and +/// embedded assets. Traversal beyond 64 levels conservatively requires version 4. A linked +/// asset (`/assets//storage == "linked"`) requires version 5. +pub fn required_version(document: &Value, assets: &[Asset<'_>]) -> u32 { + if has_linked_asset(document) { + return 5; + } + if has_version_four_feature(document) { + return 4; + } + if assets.iter().any(|asset| craft_eps::is_eps(asset.data)) { 3 } else { 2 } +} + +/// Serialize a document, its embedded assets and the proxy previews of its linked assets as a +/// `.designcraft` ZIP bundle. Linked assets have a record in the document and no `assets` entry. +pub fn save( + document: &T, + generator: &str, + preview: Option<&[u8]>, + assets: &[Asset<'_>], + proxies: &[Asset<'_>], +) -> Result, Error> { + let mut out = Cursor::new(Vec::new()); + write_to(&mut out, document, generator, preview, assets, proxies)?; + Ok(out.into_inner()) +} + +/// Like [`save`], streaming the ZIP into `out`. A writer that returns an error (to cancel a +/// long save, say) stops the write and the error is returned as [`Error::Io`]. +pub fn write_to( + out: W, + document: &T, + generator: &str, + preview: Option<&[u8]>, + assets: &[Asset<'_>], + proxies: &[Asset<'_>], +) -> Result<(), Error> { + let value = serde_json::to_value(document).map_err(|error| Error::Io(error.to_string()))?; + let embedded: Vec> = assets.iter().filter(|asset| !is_linked(&value, asset.id)).copied().collect(); + let assets = embedded.as_slice(); + let version = required_version(&value, assets); + let json = serde_json::to_vec_pretty(&value).map_err(|error| Error::Io(error.to_string()))?; + craft_document_bundle::write_to(out, &json, version, generator, preview, assets, proxies, BUNDLE).map_err(Into::into) +} + +/// Read a `.designcraft` ZIP or legacy single-file JSON, applying all versioned schema migrations +/// before deserializing the caller's model. +pub fn load(bytes: &[u8]) -> Result, Error> { + let mut package = craft_document_bundle::read(bytes, VERSION, BUNDLE)?; + upgrade_legacy_json(&mut package.document, package.version); + let document = serde_json::from_value(package.document).map_err(|error| Error::NotOurs(error.to_string()))?; + Ok(Loaded { document, assets: package.assets, proxies: package.proxies }) +} + +/// Whether `bytes` are a native DesignCraft ZIP package. +pub fn is_designcraft(bytes: &[u8]) -> bool { + craft_document_bundle::has_document(bytes, BUNDLE) +} + +/// Whether `prefix` begins with this format's stored mimetype entry. +pub fn starts_designcraft(prefix: &[u8]) -> bool { + craft_document_bundle::starts_with_mimetype(prefix, BUNDLE) +} + +/// Read the first-page PNG preview from a ZIP package, with a fixed 4 MiB cap. +pub fn read_preview(reader: R) -> Option> { + craft_document_bundle::read_preview_from(reader, PREVIEW) +} + +fn is_linked(document: &Value, id: &str) -> bool { + document.get("assets").and_then(|assets| assets.get(id)).and_then(|asset| asset.get("storage")).and_then(Value::as_str) == Some("linked") +} + +fn has_linked_asset(document: &Value) -> bool { + document + .get("assets") + .and_then(Value::as_object) + .is_some_and(|assets| assets.values().any(|asset| asset.get("storage").and_then(Value::as_str) == Some("linked"))) +} + +fn has_version_four_feature(document: &Value) -> bool { + let mut stack = vec![(document, 0usize)]; + let mut visited = 0usize; + while let Some((value, depth)) = stack.pop() { + if depth > 64 || visited >= 1_000_000 { + return true; + } + visited += 1; + match value { + Value::Object(object) => { + if object.get("layerOptions").is_some_and(|options| !options.is_null()) + || (object.get("pdfLayerOverride").and_then(Value::as_bool) == Some(true) + && object.get("pdfHiddenLayers").and_then(Value::as_array).is_none_or(Vec::is_empty)) + { + return true; + } + stack.extend(object.values().map(|child| (child, depth + 1))); + } + Value::Array(children) => stack.extend(children.iter().map(|child| (child, depth + 1))), + _ => {} + } + } + false +} + +fn upgrade_legacy_json(document: &mut Value, from: u64) { + let mut vertical = HashSet::new(); + vertical_frames(document, &mut vertical, 0); + if let Some(stories) = document.get_mut("stories").and_then(Value::as_object_mut) { + for story in stories.values_mut() { + let first = story.get("frames").and_then(|frames| frames.get(0)).and_then(Value::as_u64); + if let (Some(frame), Some(story)) = (first, story.as_object_mut()) + && vertical.contains(&frame) + && !story.contains_key("vertical") + { + story.insert("vertical".into(), Value::Bool(true)); + } + } + } + if from < 2 { + pin_paragraph_defaults(document); + } +} + +fn vertical_frames(value: &Value, out: &mut HashSet, depth: usize) { + if depth > 256 { + return; + } + match value { + Value::Object(object) => { + if let (Some(id), Some(content)) = (object.get("id").and_then(Value::as_u64), object.get("content")) + && content.get("type").and_then(Value::as_str) == Some("text") + && content.get("options").and_then(|options| options.get("vertical")).and_then(Value::as_bool) == Some(true) + { + out.insert(id); + } + for child in object.values() { + vertical_frames(child, out, depth + 1); + } + } + Value::Array(children) => { + for child in children { + vertical_frames(child, out, depth + 1); + } + } + _ => {} + } +} + +fn pin_paragraph_defaults(document: &mut Value) { + const MAX_STYLE_DEPTH: usize = 64; + let styles = document.pointer("/styles/paragraph").and_then(Value::as_array).cloned().unwrap_or_default(); + let mut by_name: HashMap, bool, bool)> = HashMap::new(); + for style in &styles { + let Some(name) = style.get("name").and_then(Value::as_str) else { continue }; + let based_on = style.get("basedOn").and_then(Value::as_str).map(str::to_string); + let attrs = style.get("para"); + let missing_keep = attrs.and_then(|attrs| attrs.get("keepAllLines")).is_none(); + let missing_drop_cap = attrs.and_then(|attrs| attrs.get("dropCapAlignLeft")).is_none(); + by_name.entry(name.to_string()).or_insert((based_on, missing_keep, missing_drop_cap)); + } + + let mut roots = HashSet::new(); + for name in by_name.keys() { + let mut chain = Vec::new(); + let mut seen = HashSet::new(); + let mut current = Some(name.as_str()); + while let Some(style_name) = current { + let Some((based_on, missing_keep, missing_drop_cap)) = by_name.get(style_name) else { break }; + if chain.len() >= MAX_STYLE_DEPTH || !seen.insert(style_name.to_string()) { + break; + } + chain.push((style_name, *missing_keep, *missing_drop_cap)); + current = based_on.as_deref(); + } + if !chain.is_empty() + && chain.iter().all(|(_, keep, drop_cap)| *keep && *drop_cap) + && let Some((root, _, _)) = chain.last() + { + roots.insert((*root).to_string()); + } + } + if let Some(styles) = document.pointer_mut("/styles/paragraph").and_then(Value::as_array_mut) { + for style in styles { + if let Some(style) = style.as_object_mut() + && style.get("name").and_then(Value::as_str).is_some_and(|name| roots.contains(name)) + { + let attrs = style.entry("para").or_insert_with(|| Value::Object(Default::default())); + if let Some(attrs) = attrs.as_object_mut() { + attrs.entry("keepAllLines").or_insert(Value::Bool(true)); + attrs.entry("dropCapAlignLeft").or_insert(Value::Bool(false)); + } + } + } + } + let known_styles: HashSet = by_name.into_keys().collect(); + pin_unstyled_paragraphs(document, &known_styles, 0); +} + +fn pin_unstyled_paragraphs(value: &mut Value, known_styles: &HashSet, depth: usize) { + if depth > 256 { + return; + } + match value { + Value::Object(object) => { + if let Some(style) = object.get("style").and_then(Value::as_str) + && !known_styles.contains(style) + { + let attrs = object.entry("para").or_insert_with(|| Value::Object(Default::default())); + if let Some(attrs) = attrs.as_object_mut() { + attrs.entry("keepAllLines").or_insert(Value::Bool(true)); + attrs.entry("dropCapAlignLeft").or_insert(Value::Bool(false)); + } + } + for child in object.values_mut() { + pin_unstyled_paragraphs(child, known_styles, depth + 1); + } + } + Value::Array(children) => { + for child in children { + pin_unstyled_paragraphs(child, known_styles, depth + 1); + } + } + _ => {} + } +} + +#[cfg(test)] +mod tests; diff --git a/crates/designcraft-format/src/tests.rs b/crates/designcraft-format/src/tests.rs new file mode 100644 index 000000000..38cf1c34f --- /dev/null +++ b/crates/designcraft-format/src/tests.rs @@ -0,0 +1,169 @@ +use super::*; +use serde_json::json; + +#[test] +fn native_bundle_roundtrips_document_assets_and_preview() -> Result<(), Box> { + let document = json!({"title":"A page", "assets":{"7":{"mime":"image/png"}}}); + let image = [1u8, 2, 3, 4]; + let assets = [Asset { id: "7", mime: "image/png", data: &image }]; + let preview = b"preview png"; + let bytes = save(&document, "DesignCraft test", Some(preview), &assets, &[])?; + + assert!(is_designcraft(&bytes)); + assert!(starts_designcraft(&bytes[..bytes.len().min(80)])); + let loaded = load::(&bytes)?; + assert_eq!(loaded.document, document); + assert_eq!(loaded.assets, [("7".to_owned(), image.to_vec())]); + assert_eq!(read_preview(std::io::Cursor::new(&bytes)).as_deref(), Some(&preview[..])); + Ok(()) +} + +#[test] +fn preview_reader_rejects_an_entry_over_its_declared_read_budget() -> Result<(), Box> { + let oversized = vec![0u8; usize::try_from(MAX_PREVIEW_BYTES + 1)?]; + let bytes = save(&json!({}), "DesignCraft test", Some(&oversized), &[], &[])?; + assert_eq!(read_preview(std::io::Cursor::new(bytes)), None); + Ok(()) +} + +#[test] +fn required_version_preserves_the_reader_capability_contract() { + assert_eq!(required_version(&json!({"title":"plain"}), &[]), 2); + assert_eq!(required_version(&json!({}), &[Asset { id: "1", mime: "application/postscript", data: b"%!PS-Adobe-3.0 EPSF-3.0" }]), 3); + assert_eq!(required_version(&json!({"layerOptions":{}}), &[]), 4); + assert_eq!(required_version(&json!({"pdfLayerOverride":true,"pdfHiddenLayers":[]}), &[]), 4); + assert_eq!(required_version(&json!({"pdfLayerOverride":true,"pdfHiddenLayers":["guide"]}), &[]), 2); + + let mut nested = json!(null); + for _ in 0..65 { + nested = json!([nested]); + } + assert_eq!(required_version(&nested, &[]), 4); +} + +#[test] +fn legacy_json_extracts_assets_and_applies_v1_migrations() -> Result<(), Box> { + let legacy = json!({ + "assetData":{"7":"AQID"}, + "frames":[{"id":8,"content":{"type":"text","options":{"vertical":true}}}], + "stories":{"vertical":{"frames":[8]},"explicit":{"frames":[8],"vertical":false}}, + "styles":{"paragraph":[{"name":"Base"},{"name":"Child","basedOn":"Base","para":{"keepAllLines":false}}]}, + "paragraphs":[{"style":"Missing"}] + }); + let loaded = load::(&serde_json::to_vec(&legacy)?)?; + + assert_eq!(loaded.assets, [("7".to_owned(), vec![1, 2, 3])]); + assert_eq!(loaded.document.get("assetData"), None); + assert_eq!(loaded.document["stories"]["vertical"]["vertical"], true); + assert_eq!(loaded.document["stories"]["explicit"]["vertical"], false); + assert_eq!(loaded.document["styles"]["paragraph"][0]["para"]["keepAllLines"], true); + assert_eq!(loaded.document["styles"]["paragraph"][0]["para"]["dropCapAlignLeft"], false); + assert_eq!(loaded.document["styles"]["paragraph"][1]["para"]["keepAllLines"], false); + assert_eq!(loaded.document["paragraphs"][0]["para"]["keepAllLines"], true); + Ok(()) +} + +#[test] +fn newer_metadata_precedes_malformed_document_json() -> Result<(), Box> { + let bytes = craft_document_bundle::write(b"{", VERSION + 1, "future", None, &[], &[], BUNDLE)?; + assert!(matches!(load::(&bytes), Err(Error::TooNew(version)) if version == u64::from(VERSION + 1))); + assert!(matches!(load::(b"not a package"), Err(Error::NotOurs(_)))); + Ok(()) +} + +fn linked_doc() -> Value { + json!({"assets":{"7":{"storage":"linked","mime":"image/png"},"8":{"mime":"image/png"}}}) +} + +#[test] +fn embedded_documents_keep_their_version() { + assert_eq!(required_version(&json!({"assets":{"7":{"storage":"embedded"},"8":{}}}), &[]), 2); +} + +#[test] +fn a_linked_asset_requires_version_five_and_old_readers_refuse_it() -> Result<(), Box> { + assert_eq!(required_version(&linked_doc(), &[]), 5); + let bytes = save(&linked_doc(), "t", None, &[], &[])?; + assert_eq!(craft_document_bundle::read(&bytes, 4, BUNDLE), Err(craft_document_bundle::Error::TooNew(5))); + let loaded = load::(&bytes)?; + assert_eq!(loaded.document, linked_doc()); + assert!(loaded.assets.is_empty()); + Ok(()) +} + +#[test] +fn proxies_round_trip_and_unknown_or_oversized_ones_are_skipped() -> Result<(), Box> { + let big = vec![0u8; usize::try_from(craft_document_bundle::PROXY_LIMIT)? + 1]; + let proxies = [ + Asset { id: "7", mime: "image/jpeg", data: b"jpeg" }, + Asset { id: "8", mime: "image/png", data: &big }, + Asset { id: "99", mime: "image/png", data: b"unknown" }, + ]; + let bytes = save(&linked_doc(), "t", None, &[], &proxies)?; + assert_eq!(load::(&bytes)?.proxies, [("7".to_owned(), b"jpeg".to_vec())]); + Ok(()) +} + +#[test] +fn proxy_paths_must_be_exactly_proxies_id_ext() -> Result<(), Box> { + let proxies = [ + Asset { id: "../x", mime: "image/png", data: b"a" }, + Asset { id: "a/b", mime: "image/png", data: b"b" }, + Asset { id: "7/8", mime: "image/png", data: b"c" }, + ]; + let bytes = save(&linked_doc(), "t", None, &[], &proxies)?; + assert!(load::(&bytes)?.proxies.is_empty()); + Ok(()) +} + +#[test] +fn write_to_stops_when_the_writer_fails() { + struct Failing { + inner: Cursor>, + budget: usize, + failed: bool, + calls_after_failure: std::rc::Rc>, + } + impl Write for Failing { + fn write(&mut self, buf: &[u8]) -> std::io::Result { + if self.failed { + self.calls_after_failure.set(self.calls_after_failure.get() + 1); + } + if self.budget < buf.len() { + self.failed = true; + return Err(std::io::Error::other("cancelled")); + } + self.budget -= buf.len(); + self.inner.write(buf) + } + fn flush(&mut self) -> std::io::Result<()> { + if self.failed { + self.calls_after_failure.set(self.calls_after_failure.get() + 1); + } + self.inner.flush() + } + } + impl Seek for Failing { + fn seek(&mut self, pos: std::io::SeekFrom) -> std::io::Result { + if self.failed { + self.calls_after_failure.set(self.calls_after_failure.get() + 1); + } + self.inner.seek(pos) + } + } + let data = vec![1u8; 4096]; + let assets = [Asset { id: "8", mime: "image/png", data: &data }]; + let calls = std::rc::Rc::new(std::cell::Cell::new(0)); + let out = Failing { inner: Cursor::new(Vec::new()), budget: 1024, failed: false, calls_after_failure: calls.clone() }; + let result = write_to(out, &linked_doc(), "t", None, &assets, &[]); + assert_eq!(calls.get(), 0, "no write, flush or seek reaches the writer after it failed"); + assert!(matches!(result, Err(Error::Io(message)) if message.contains("cancelled"))); +} + +#[test] +fn bytes_supplied_for_a_linked_asset_are_not_written() -> Result<(), Box> { + let assets = [Asset { id: "7", mime: "image/png", data: b"linked bytes" }, Asset { id: "8", mime: "image/png", data: b"kept" }]; + let bytes = save(&linked_doc(), "t", None, &assets, &[])?; + assert_eq!(load::(&bytes)?.assets, [("8".to_owned(), b"kept".to_vec())]); + Ok(()) +} diff --git a/crates/doc-binary/src/fib.rs b/crates/doc-binary/src/fib.rs index 7bb3ec149..4224f6962 100644 --- a/crates/doc-binary/src/fib.rs +++ b/crates/doc-binary/src/fib.rs @@ -138,7 +138,9 @@ impl Fib { } let w_ident = u16le(0); if w_ident == 0xA5DC { - return Err(DocbinError::NotWord("Word 6.0/95 format, only Word 97-2003 is supported".into())); + // `read` sends Word 6.0/95 streams to `word6` before this FIB is parsed. This keeps any + // other caller from reading one with the Word 97 layout, which it doesn't have. + return Err(DocbinError::NotWord("a Word 6.0/95 file doesn't have the Word 97 FIB layout".into())); } if w_ident != W_IDENT { return Err(DocbinError::NotWord(format!("magic 0x{w_ident:04X}, expected 0x{W_IDENT:04X}"))); diff --git a/crates/doc-binary/src/fkp.rs b/crates/doc-binary/src/fkp.rs index 3de272da2..3a795d4a4 100644 --- a/crates/doc-binary/src/fkp.rs +++ b/crates/doc-binary/src/fkp.rs @@ -29,6 +29,19 @@ pub(crate) struct Bins { sorted: bool, /// Page number (Ă— 512 = byte offset in the WordDocument stream). pns: Vec, + /// Runs read ahead of time instead of pages (Word 6.0/95, whose FKPs and sprms are another + /// layout, translated by [`crate::word6`]): sorted by `lo`, not overlapping. + translated: Option>, +} + +/// One run of a [`Bins::translated`] table: the FCs `[lo, hi)` it covers, the paragraph style +/// (for PAPX runs) and a Word 97 grpprl. +#[derive(Clone, Debug, PartialEq)] +pub(crate) struct TranslatedRun { + pub(crate) lo: u32, + pub(crate) hi: u32, + pub(crate) istd: u16, + pub(crate) grpprl: Vec, } impl Bins { @@ -37,7 +50,7 @@ impl Bins { let fc = fc as usize; let lcb = lcb as usize; if lcb < 4 || !(lcb - 4).is_multiple_of(8) { - return Ok(Bins { fcs: Vec::new(), sorted: true, pns: Vec::new() }); + return Ok(Bins { fcs: Vec::new(), sorted: true, pns: Vec::new(), translated: None }); } let end = fc.checked_add(lcb).ok_or_else(|| DocbinError::Malformed("bin table overflow".into()))?; let plc = table.get(fc..end).ok_or_else(|| DocbinError::Malformed("bin table outside the Table stream".into()))?; @@ -56,7 +69,31 @@ impl Bins { pns.push(at(n + 1 + i).ok_or_else(|| DocbinError::Malformed("truncated bin aPn".into()))? & 0x003F_FFFF); } let sorted = fcs.windows(2).all(|w| w.first() <= w.get(1)); - Ok(Bins { fcs, sorted, pns }) + Ok(Bins { fcs, sorted, pns, translated: None }) + } + + /// A table of runs read ahead of time ([`TranslatedRun`]). Runs are sorted by start; one that + /// overlaps an earlier one (a corrupt file) or covers nothing is dropped. + pub(crate) fn translated(mut runs: Vec) -> Bins { + runs.sort_by_key(|r| r.lo); + let mut kept: Vec = Vec::with_capacity(runs.len()); + for r in runs { + if r.lo < r.hi && kept.last().is_none_or(|last| r.lo >= last.hi) { + kept.push(r); + } + } + Bins { fcs: Vec::new(), sorted: true, pns: Vec::new(), translated: Some(kept) } + } + + /// For a translated table: the run covering `fc`, or the gap `[lo, hi)` between runs that + /// holds it. + fn translated_at(runs: &[TranslatedRun], fc: u32) -> Result<&TranslatedRun, (u32, u32)> { + let i = runs.partition_point(|r| r.lo <= fc); + let prev = i.checked_sub(1).and_then(|i| runs.get(i)); + match prev { + Some(r) if fc < r.hi => Ok(r), + _ => Err((prev.map_or(0, |r| r.hi), runs.get(i).map_or(u32::MAX, |r| r.lo))), + } } /// The page containing `fc`, as a slice of the WordDocument stream, and the FC range @@ -77,13 +114,19 @@ impl Bins { /// The grpprl of the Chpx covering `fc` ([MS-DOC] ChpxFkp: rgfc u32s, rgb offsets Ă—2, /// crun in the last byte). - pub(crate) fn chpx<'w>(&self, word: &'w [u8], fc: u32) -> &'w [u8] { + pub(crate) fn chpx<'w>(&'w self, word: &'w [u8], fc: u32) -> &'w [u8] { self.chpx_run(word, fc).grpprl } /// [`Bins::chpx`] with the FC range over which the same grpprl is the answer, so a walk /// over consecutive characters looks it up once per run, not once per character. - pub(crate) fn chpx_run<'w>(&self, word: &'w [u8], fc: u32) -> ChpxRun<'w> { + pub(crate) fn chpx_run<'w>(&'w self, word: &'w [u8], fc: u32) -> ChpxRun<'w> { + if let Some(runs) = &self.translated { + return match Self::translated_at(runs, fc) { + Ok(r) => ChpxRun { grpprl: &r.grpprl, lo: r.lo, hi: r.hi }, + Err((lo, hi)) => ChpxRun { grpprl: &[], lo, hi }, + }; + } let none = ChpxRun { grpprl: &[], lo: 1, hi: 0 }; let Some((page, page_lo, page_hi)) = self.page(word, fc) else { return none }; let Some(&crun) = page.get(511) else { return none }; @@ -113,7 +156,13 @@ impl Bins { /// ([MS-DOC] PapxFkp: rgfc u32s — each the FC *following* a paragraph mark —, rgbx BX /// entries (1 offset byte + 12 PHE bytes), crun in the last byte; the PAPX itself is a /// PapxInFkp with the cb/cb' size quirk). Query with the FC of the character after the mark. - pub(crate) fn papx<'w>(&self, word: &'w [u8], fc: u32) -> (u16, &'w [u8]) { + pub(crate) fn papx<'w>(&'w self, word: &'w [u8], fc: u32) -> (u16, &'w [u8]) { + if let Some(runs) = &self.translated { + return match Self::translated_at(runs, fc) { + Ok(r) => (r.istd, &r.grpprl), + Err(_) => (0, &[]), + }; + } let Some((page, _, _)) = self.page(word, fc) else { return (0, &[]) }; let Some(&crun) = page.get(511) else { return (0, &[]) }; // BX is 13 bytes; cap crun at what fits before byte 511. diff --git a/crates/doc-binary/src/lib.rs b/crates/doc-binary/src/lib.rs index 386e753f5..b2d7f3729 100644 --- a/crates/doc-binary/src/lib.rs +++ b/crates/doc-binary/src/lib.rs @@ -6,6 +6,9 @@ //! like the other Craft parsers, is lenient by design (unknown structures are skipped, bad numbers //! fall back to defaults) and bounded (stream sizes, piece counts and property counts are //! capped), so hostile input yields an error or a best-effort document, never a panic. +//! +//! Word 6.0/95 files, which that specification does not cover, are read by `word6` from their +//! observed layout: their text, formatting and tables so far. #![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] // The code came from an app whose lints allow this. #![allow(clippy::type_complexity)] @@ -29,6 +32,7 @@ mod sections; mod sprm; mod stsh; mod table; +mod word6; use std::sync::Arc; @@ -220,6 +224,9 @@ fn read_limited(bytes: &[u8], password: Option<&str>, max_paragraphs: usize) -> if word.len() as u64 > MAX_STREAM { return Err(DocbinError::Limit(format!("WordDocument stream is {} bytes", word.len()))); } + if word6::is_word6(&word) { + return read_word6(&comp, &word, max_paragraphs); + } let enc = fib::peek_encryption(&word); // The FIB names the Table stream (0Table or 1Table); fall back to whichever exists, // since files written by other writers sometimes carry the wrong bit. @@ -431,6 +438,55 @@ fn read_limited(bytes: &[u8], password: Option<&str>, max_paragraphs: usize) -> Ok(doc) } +/// Read a Word 6.0/95 file ([`word6`]): its main text, with its character and paragraph +/// formatting and table rows translated to Word 97 sprms, through the same walk as a Word 97 file. +/// Its paragraph styles are read too; character styles and the other stories are not yet, and a +/// section mark reads as a page break. +fn read_word6(comp: &craft_cfb::CompoundFile<'_>, word: &[u8], max_paragraphs: usize) -> Result { + let fib = word6::Fib6::parse(word)?; + let content = word6::content(word, &fib)?; + let mut doc = Document::new(); + // Word opens these files in compatibility mode 11 too. + doc.settings.compat_mode = 11; + if let Some(core) = metadata::read(comp) { + doc.core = core; + } + doc.body.clear(); + // The paragraph styles, built as for Word 97 files from their translated formatting. + doc.styles = Arc::new(fmt::stylesheet(&content.styles, &content.fonts, stsh::StandardFonts::default())); + let sheet: StyleSheet = StyleSheet::clone(&doc.styles); + // As for Word 97 files: the text read can't exceed twice the stream, nor the absolute ceiling. + let budget = std::cell::Cell::new((word.len() as u64).saturating_mul(2).min(MAX_TOTAL_CHARS)); + let work = Work::new(MAX_WORK); + let paragraphs = ParaBudget { left: std::cell::Cell::new(max_paragraphs), exceeded: std::cell::Cell::new(false) }; + let mut pics = media::Pictures::for_input(word.len()); + let ctx = WalkCtx { + word, + data: &[], + pieces: &content.pieces, + chpx_bins: &content.chpx, + papx_bins: &content.papx, + fonts: &content.fonts, + raw_styles: &content.styles, + sheet: &sheet, + notes: &[], + marks: &[], + drawn: &[], + sect_ends: &[], + budget: &budget, + paragraphs: ¶graphs, + work: &work, + }; + let paras = walk(&ctx, 0, fib.ccp_text, &mut pics); + if paragraphs.exceeded.get() { + return Err(paragraph_limit(max_paragraphs)); + } + doc.body = table::assemble(paras); + doc.ensure_nonempty(); + work.log(); + Ok(doc) +} + fn paragraph_limit(max_paragraphs: usize) -> DocbinError { DocbinError::Limit(format!("the document has more than {max_paragraphs} paragraphs and table cells")) } @@ -648,6 +704,20 @@ fn walk<'a>(ctx: &WalkCtx<'a>, cp_start: u32, cp_end: u32, pics: &mut media::Pic // querying at the mark's own FC finds that paragraph's properties. let fc_mark = ctx.pieces.fc_of_cp(cp).unwrap_or(0); let (istd, papx) = ctx.papx_bins.papx(ctx.word, fc_mark); + // The Prm of the piece holding the mark adds its paragraph sprms after the PAPX's + // ([MS-DOC] §2.4.6.1, step 5), as Word 6/95's fast saves record a paragraph change. + let extra = prm_paragraph_sprms(ctx.pieces.piece_prm(cp), ctx.work); + let combined; + let papx: &[u8] = if extra.is_empty() { + papx + } else { + // Re-framed from its Prls, so that a pad byte at its end can't swallow the + // first appended opcode. + let mut v: Vec = sprm::iter(papx).flat_map(|p| p.op.to_le_bytes().into_iter().chain(p.operand.iter().copied())).collect(); + v.extend_from_slice(&extra); + combined = v; + &combined + }; if let Some(st) = ctx.raw_styles.get(istd as usize).filter(|s| !s.name.is_empty()) { pb.props.style = Some(fmt::style_id(&st.name)); } @@ -991,13 +1061,42 @@ struct FmtState<'a> { /// paragraph style stay the same, so a Prm grpprl (up to 64 KB) is interpreted once per /// formatting change, not once per character (#3692). struct FmtMemo<'a> { - /// The CHPX grpprl, compared by address: it is a slice of the WordDocument stream. + /// The CHPX grpprl, compared by address: a slice of the WordDocument stream, or of a Word 6.0/95 + /// file's translated runs, both of which stay put for the walk. chpx: &'a [u8], prm: u16, style: Option, props: CharProps, } +/// The paragraph and table sprms (sgc 1 and 5) of a piece's Prm, as Prls to append to a +/// paragraph's PAPX: a Prm0's single sprm, or those of the Prm1's grpprl. A Prm1 grpprl is +/// charged to `work`, as for [`apply_prm`]. +fn prm_paragraph_sprms(prm: PrmRef<'_>, work: &Work) -> Vec { + let mut out = Vec::new(); + match prm { + PrmRef::Sprm0 { isprm, val } => { + if let Some(op) = sprm::from_prm0(isprm) + && matches!(Prl { op, operand: &[] }.sgc(), 1 | 5) + { + out.extend_from_slice(&op.to_le_bytes()); + out.push(val); + } + } + PrmRef::Grpprl(Some(g)) => { + if !work.charge(g.len() as u64, Cost::Formatting) { + return out; + } + for prl in sprm::iter(g).filter(|p| matches!(p.sgc(), 1 | 5)) { + out.extend_from_slice(&prl.op.to_le_bytes()); + out.extend_from_slice(prl.operand); + } + } + PrmRef::None | PrmRef::Grpprl(None) => {} + } + out +} + /// Apply a piece's Prm (Prm0 short sprm or Prm1 grpprl) to character properties. A Prm1 /// grpprl is charged to `work`; once that is spent the grpprl is skipped. fn apply_prm(props: &mut CharProps, prm: PrmRef<'_>, fonts: &[String], base: &CharProps, work: &Work) { diff --git a/crates/doc-binary/src/piece.rs b/crates/doc-binary/src/piece.rs index 5f36a8198..701a4bc67 100644 --- a/crates/doc-binary/src/piece.rs +++ b/crates/doc-binary/src/piece.rs @@ -1,20 +1,71 @@ //! The piece table ([MS-DOC] `Clx`/`Pcdt`/`PlcPcd`): where every character position lives in //! the WordDocument stream, and how it is encoded. +//! +//! Word 6.0/95 files ([`crate::word6`]) keep the same `Clx` shape, but every piece's text is at a +//! plain byte offset: 8-bit text in the code page of its piece's or run's font (or the document's), +//! or UTF-16 in a file with the FIB's fExtChar flag. -use craft_codepage::cp1252; +use craft_codepage::{code_page_byte, cp1252, is_byte_code_page}; use crate::DocbinError; /// Most pieces any real file has; more means a corrupt `PlcPcd`. const MAX_PIECES: usize = 1_000_000; -/// One piece: text at FcCompressed `fc` (30-bit offset + compression bit). +/// How a piece's characters are stored. +#[derive(Clone, Copy, Debug, PartialEq)] +enum Text { + /// UTF-16LE at `fc`. + Utf16, + /// Windows-1252, one byte per character at `fc / 2` (Word 97 FcCompressed). + Compressed, + /// One byte per character at `fc` itself, in the code page [`PieceTable::code_page_at`] gives + /// that byte (Word 6.0/95). + Bytes, +} + +/// One piece: its text at `fc` (for Word 97, the 30-bit offset of FcCompressed). #[derive(Clone, Copy, Debug)] struct Pcd { fc: u32, - compressed: bool, - /// `Prm`: either a single sprm (Prm0) or an index into the Clx's Prc grpprls (Prm1). + text: Text, + /// `Prm`: either a single sprm (Prm0) or an index into the Clx's Prc grpprls (Prm1). In a Word + /// 6.0/95 table the Prcs hold translated grpprls ([`PieceTable::parse_word6`]), and a Prm0's + /// isprm is already the opcode [MS-DOC]'s isprm table maps. prm: u16, + /// For a Word 6.0/95 [`Text::Bytes`] piece whose Prm names a font: that font's code page, + /// which outranks the runs' ([`PieceTable::code_page_at`]). + code_page: Option, +} + +impl Pcd { + /// The byte offset of the character `step / 2` characters into the piece. `step` is twice + /// the CP offset, so one formula serves all three encodings. + fn fc_at(&self, step: u32) -> Option { + match self.text { + Text::Utf16 => self.fc.checked_add(step), + Text::Compressed => self.fc.checked_add(step)?.checked_div(2), + Text::Bytes => self.fc.checked_add(step / 2), + } + } +} + +/// The code page of the 8-bit text in the byte range `[lo, hi)` of the WordDocument stream. +#[derive(Clone, Copy, Debug, PartialEq)] +pub(crate) struct CodePageRun { + pub(crate) lo: u32, + pub(crate) hi: u32, + pub(crate) code_page: u16, +} + +/// How a Word 6.0/95 file's text is stored and decoded ([`crate::word6`]). +pub(crate) struct Word6Text { + /// UTF-16 rather than 8-bit (the FIB's fExtChar flag). + pub(crate) wide: bool, + /// The code page of 8-bit text that no piece or run gives one for. + pub(crate) default_code_page: u16, + /// The code page of each byte range, from its run's font (sorted by `lo`, not overlapping). + pub(crate) code_pages: Vec, } /// The parsed piece table plus the grpprls of the Prcs that Prm1 references. @@ -23,11 +74,58 @@ pub(crate) struct PieceTable { cps: Vec, pieces: Vec, prc_grpprls: Vec>, + /// For [`Text::Bytes`] pieces: the code page of each byte range, sorted by `lo`, not + /// overlapping. Bytes outside every range are in `default_code_page`. + code_pages: Vec, + default_code_page: u16, } impl PieceTable { /// Parse the `Clx` at `table[fc .. fc+lcb]`. pub(crate) fn parse(table: &[u8], fc: u32, lcb: u32) -> Result { + Self::parse_clx(table, fc, lcb, None) + } + + /// Parse a Word 6.0/95 `Clx` at `word[fc .. fc+lcb]`: the Word 97 shape, but every piece's text + /// is at its plain byte offset. `piece_code_page` gives the code page a piece's raw Prm + /// imposes (with the Clx's raw Prc grpprls to look Prm1 indices up in), if any; then each Prc + /// grpprl is replaced by `translate_prc`'s Word 97 sprms, which the walk applies to the piece. + pub(crate) fn parse_word6( + word: &[u8], + fc: u32, + lcb: u32, + text: Word6Text, + piece_code_page: impl Fn(u16, &[Vec]) -> Option, + translate_prc: impl Fn(&[u8]) -> Vec, + ) -> Result { + let mut t = Self::parse_clx(word, fc, lcb, Some(text.wide))?; + if !text.wide { + for p in &mut t.pieces { + p.code_page = piece_code_page(p.prm, &t.prc_grpprls); + } + } + t.prc_grpprls = t.prc_grpprls.iter().map(|g| translate_prc(g)).collect(); + t.code_pages = text.code_pages; + t.default_code_page = text.default_code_page; + Ok(t) + } + + /// One Word 6.0/95 piece: `ccp` characters from byte `fc` on. This covers a file that was not + /// fast-saved, which has no `Clx`. + pub(crate) fn contiguous_word6(fc: u32, ccp: u32, text: Word6Text) -> PieceTable { + let kind = if text.wide { Text::Utf16 } else { Text::Bytes }; + PieceTable { + cps: vec![0, ccp], + pieces: vec![Pcd { fc, text: kind, prm: 0, code_page: None }], + prc_grpprls: Vec::new(), + code_pages: text.code_pages, + default_code_page: text.default_code_page, + } + } + + /// `word6`: `None` for Word 97 and later, `Some(wide)` for a Word 6.0/95 file (whose raw Prms + /// and Prcs are kept for [`PieceTable::parse_word6`]). + fn parse_clx(table: &[u8], fc: u32, lcb: u32, word6: Option) -> Result { let fc = fc as usize; let lcb = lcb as usize; if lcb == 0 { @@ -95,9 +193,26 @@ impl PieceTable { return Err(DocbinError::Malformed("truncated aPcd".into())); }; let prm = plc.get(at + 4..at + 6).map(|b| u16::from_le_bytes([b[0], b[1]])).unwrap_or(0); - pieces.push(Pcd { fc: fc_raw & 0x3FFF_FFFF, compressed: fc_raw & 0x4000_0000 != 0, prm }); + pieces.push(match word6 { + // A plain byte offset, with no compression bit: 8-bit text, or UTF-16 in a file with + // fExtChar. The raw Prm is kept for `parse_word6`. + Some(wide) => Pcd { fc: fc_raw, text: if wide { Text::Utf16 } else { Text::Bytes }, prm, code_page: None }, + None => { + let text = if fc_raw & 0x4000_0000 != 0 { Text::Compressed } else { Text::Utf16 }; + Pcd { fc: fc_raw & 0x3FFF_FFFF, text, prm, code_page: None } + } + }); + } + Ok(PieceTable { cps, pieces, prc_grpprls, code_pages: Vec::new(), default_code_page: 1252 }) + } + + /// The code page of the 8-bit text byte at `fc` (in [`Text::Bytes`] pieces). + fn code_page_at(&self, fc: u32) -> u16 { + let i = self.code_pages.partition_point(|r| r.lo <= fc); + match i.checked_sub(1).and_then(|i| self.code_pages.get(i)) { + Some(r) if fc < r.hi => r.code_page, + _ => self.default_code_page, } - Ok(PieceTable { cps, pieces, prc_grpprls }) } /// The Prls this piece's `Prm` adds. Bit 0 (`fComplex`) discriminates: a Prm0 @@ -133,8 +248,7 @@ impl PieceTable { let p = self.piece_of(cp)?; let i = self.cps.partition_point(|&c| c <= cp).checked_sub(1)?; let cp0 = *self.cps.get(i)?; - let step = cp.checked_sub(cp0)?.checked_mul(2)?; - if p.compressed { p.fc.checked_add(step)?.checked_div(2) } else { p.fc.checked_add(step) } + p.fc_at(cp.checked_sub(cp0)?.checked_mul(2)?) } /// The piece holding `cp`, with the CP range it covers: for every CP in that range the @@ -167,13 +281,24 @@ impl PieceTable { let (Some(p), Some(&cp0)) = (self.pieces.get(i), self.cps.get(i)) else { return Err(u32::MAX) }; let Some(off) = cp.checked_sub(cp0) else { return Err(next) }; let off = off as usize; - if p.compressed { - // 8-bit, Windows-1252, one byte per character, stored at fc/2. - let at = ((p.fc / 2) as usize).checked_add(off); - return match at.and_then(|a| word.get(a)) { - Some(&b) => Ok((cp1252(b), 1)), - None => Err(next), - }; + match p.text { + Text::Compressed => { + // 8-bit, Windows-1252, one byte per character, stored at fc/2. + let at = ((p.fc / 2) as usize).checked_add(off); + return match at.and_then(|a| word.get(a)) { + Some(&b) => Ok((cp1252(b), 1)), + None => Err(next), + }; + } + Text::Bytes => { + // 8-bit, one byte per character at fc, in its run's code page. + let at = (p.fc as usize).checked_add(off); + return match at.and_then(|a| Some((u32::try_from(a).ok()?, *word.get(a)?))) { + Some((a, b)) => Ok((byte_char(p.code_page.unwrap_or_else(|| self.code_page_at(a)), b), 1)), + None => Err(next), + }; + } + Text::Utf16 => {} } // UTF-16LE, one 16-bit unit per character, stored at fc. let unit = |k: usize| -> Option { @@ -207,8 +332,7 @@ impl PieceSpan<'_> { /// [`PieceTable::fc_of_cp`] for a CP inside the span. pub(crate) fn fc_at(&self, cp: u32) -> Option { - let step = cp.checked_sub(self.cp_lo)?.checked_mul(2)?; - if self.pcd.compressed { self.pcd.fc.checked_add(step)?.checked_div(2) } else { self.pcd.fc.checked_add(step) } + self.pcd.fc_at(cp.checked_sub(self.cp_lo)?.checked_mul(2)?) } /// The raw Prm of the piece: two CPs with the same value get the same [`PrmRef`]. @@ -228,3 +352,23 @@ pub(crate) enum PrmRef<'a> { /// Prm1: a grpprl from a Prc (absent if out of range). Grpprl(Option<&'a [u8]>), } + +/// Windows' code page number for symbol fonts (CP_SYMBOL). +pub(crate) const SYMBOL_CODE_PAGE: u16 = 42; + +/// One byte of 8-bit text in `code_page`. The East Asian code pages store a character as a lead +/// and a trail byte, which one CP per byte can't represent, so their bytes above ASCII read as +/// U+FFFD rather than as unrelated Windows-1252 letters. +fn byte_char(code_page: u16, b: u8) -> char { + if code_page == SYMBOL_CODE_PAGE { + // Each character from the space up to its private-use code point, as Word reads a document + // marked as Symbol; control characters (paragraph marks, fields) stay themselves. + if b < 0x20 { char::from(b) } else { char::from_u32(0xF000 + u32::from(b)).unwrap_or('\u{FFFD}') } + } else if is_byte_code_page(code_page) { + code_page_byte(code_page, b) + } else if b.is_ascii() { + char::from(b) + } else { + '\u{FFFD}' + } +} diff --git a/crates/doc-binary/src/tests.rs b/crates/doc-binary/src/tests.rs index 80cef3257..ec5650695 100644 --- a/crates/doc-binary/src/tests.rs +++ b/crates/doc-binary/src/tests.rs @@ -10,6 +10,7 @@ use super::read; mod drawing; mod encrypted; mod metadata; +mod word6; /// Offset in the WordDocument stream where test text pieces are placed (past the FIB). const TEXT_FC: u32 = 0x600; @@ -176,12 +177,15 @@ fn wrong_magic_rejected() { } #[test] -fn word_6_95_rejected() { - let mut fib = fib_bytes(&FibSpec::default()); +fn word_6_95_magic_goes_to_the_word6_reader() { + // A Word 6.0/95 signature is no longer refused: the stream is read as that version (here an + // empty document, since a Word 97 FIB holds no Word 6 text length). The flags are cleared: + // the Word 97 default sets fExtChar, which the Word 6 reader refuses. + let mut fib = fib_bytes(&FibSpec { flags: 0, ..FibSpec::default() }); fib[0] = 0xDC; fib[1] = 0xA5; let f = cfb_file(&[("WordDocument", &fib)]); - assert!(matches!(read(&f), Err(super::DocbinError::NotWord(m)) if m.contains("6.0/95"))); + assert!(read(&f).is_ok()); } #[test] @@ -882,8 +886,6 @@ fn corpus_files_open() { let bytes = std::fs::read(&p).expect("read"); let doc = match super::read(&bytes) { Ok(d) => d, - // Word 6/95 files are a documented, intentional rejection. - Err(super::DocbinError::NotWord(m)) if m.contains("6.0/95") => continue, Err(e) => panic!("{}: {e}", p.display()), }; assert!(!doc.styles.styles.is_empty(), "{}: no styles", p.display()); @@ -2106,6 +2108,29 @@ fn one_section(text: &'static str, sepx: Vec, papx_entries: Vec<(u32, u16, V } } +#[test] +fn piece_prm_paragraph_formatting_applies_to_the_marks_paragraph() { + // [MS-DOC] §2.4.6.1 step 5: the Prm of the piece holding the paragraph mark adds its paragraph + // sprms after the PAPX's. Prm1 -> a Prc with sprmPDxaLeft80 (1440 twips) and sprmPJc80 + // (centred); Prm0 isprm 0x05 (sprmPJc) with operand 2 (right). + let prcs = vec![grpprl(&[(0x840F, &1440i16.to_le_bytes()), (0x2403, &[0x01])])]; + let para = |prm: u16| { + let spec = RichSpec { + text: "Text\r", + styles: vec![(1, 0x0FFF, "Normal", Vec::new(), Vec::new())], + fonts: vec!["Times New Roman"], + chpx_runs: vec![(0, Vec::new())], + papx_entries: vec![(4, 0, grpprl(&[(0x840F, &360i16.to_le_bytes())]))], + }; + let doc = read(&rich_doc_with(&spec, &RichExtra { prcs: prcs.clone(), prm, ..Default::default() })).expect("opens"); + let p = para0(&doc); + (p.props.indent_left, p.props.align) + }; + assert_eq!(para(0x0000), (Some(18.0), None), "no Prm: the PAPX's indent"); + assert_eq!(para(0x0001), (Some(72.0), Some(craft_wordmodel::props::Align::Center)), "Prm1: the Prc's sprms override the PAPX's"); + assert_eq!(para(0x020A), (Some(18.0), Some(craft_wordmodel::props::Align::Right)), "Prm0: sprmPJc right"); +} + #[test] fn piece_prm_character_formatting_applied() { // #458. Bit 0 of a Prm is fComplex ([MS-DOC] §2.9.215/§2.9.216): Prm0 = isprm in bits diff --git a/crates/doc-binary/src/tests/word6.rs b/crates/doc-binary/src/tests/word6.rs new file mode 100644 index 000000000..fa3f68a51 --- /dev/null +++ b/crates/doc-binary/src/tests/word6.rs @@ -0,0 +1,747 @@ +//! Word 6.0/95 fixtures, built in code. No public specification describes this version; these +//! build the layout observed in real files and described in `crate::word6`. + +use proptest::prelude::*; + +use super::{body_texts, cfb_file}; +use crate::{DocbinError, read}; + +/// Where the text starts: right after the FIB, as in every observed file. +const TEXT_AT: u32 = 0x300; + +/// One stylesheet entry as Word 6/95 stores it. +#[derive(Clone, Copy, Default)] +struct Std<'a> { + /// The built-in style id, or 0x0FFE for a user-defined style. + sti: u16, + /// 1 paragraph, 2 character. + sgc: u16, + /// The base style's index, or 0x0FFF. + base: u16, + name: &'a str, + /// Paragraph sprms (paragraph styles only). + papx: &'a [u8], + /// Character sprms. + chpx: &'a [u8], +} + +/// The stylesheet bytes for `styles`: the 14-byte header (count, fixed-part size 8, …), then per +/// style a size and the entry, or 0 for an empty slot. +fn stylesheet(styles: &[Option]) -> Vec { + let mut v = vec![14u8, 0]; + for w in [styles.len() as u16, 8, 1, 0x4B, 0x0F, 0, 0] { + v.extend_from_slice(&w.to_le_bytes()); + } + for (istd, st) in styles.iter().enumerate() { + let Some(st) = st else { + v.extend_from_slice(&[0, 0]); + continue; + }; + let cupx: u16 = if st.sgc == 1 { 2 } else { 1 }; + let mut e = Vec::new(); + for w in [st.sti, st.sgc | st.base << 4, cupx, 0] { + e.extend_from_slice(&w.to_le_bytes()); + } + e.push(st.name.len() as u8); + e.extend_from_slice(st.name.as_bytes()); + e.push(0); + let upx = |e: &mut Vec, bytes: &[u8]| { + if e.len() % 2 == 1 { + e.push(0); + } + e.extend_from_slice(&(bytes.len() as u16).to_le_bytes()); + e.extend_from_slice(bytes); + }; + if st.sgc == 1 { + let mut papx = (istd as u16).to_le_bytes().to_vec(); + papx.extend_from_slice(st.papx); + upx(&mut e, &papx); + } + upx(&mut e, st.chpx); + v.extend_from_slice(&(e.len() as u16).to_le_bytes()); + v.extend_from_slice(&e); + } + v +} + +/// What a test Word 6.0/95 file holds. Offsets in `runs` and `pieces` are into `text`. +#[derive(Clone, Default)] +struct Doc6<'a> { + nfib: u16, + flags: u16, + /// The stored 8-bit text. + text: &'a [u8], + /// (charset, name) per font index. + fonts: &'a [(u8, &'a str)], + /// The font table's entries as raw bytes instead of `fonts` (the size word is added). + raw_fonts: Option<&'a [u8]>, + /// Contiguous character runs (lo, hi, grpprl), all in one FKP page. + runs: &'a [(u32, u32, &'a [u8])], + /// Contiguous paragraph runs (lo, hi, grpprl), style 0, all in one FKP page. Each covers the + /// paragraph whose mark lies in it. + paras: &'a [(u32, u32, &'a [u8])], + /// For a fast-saved file: (length, offset) per piece, in CP order. + pieces: Option<&'a [(u32, u32)]>, + /// Each piece's Prm (0 when absent or shorter than `pieces`). + prms: &'a [u16], + /// The Clx's Prc grpprls, which a Prm1 indexes. + prcs: &'a [&'a [u8]], + /// The FIB's character set word at 0x14. + chse: u16, + /// The style of each paragraph run (0, Normal, when absent or shorter than `paras`). + para_styles: &'a [u16], + /// The stylesheet: one entry per style index, `None` for an empty slot. + styles: &'a [Option>], + /// The main text's length in characters, when it isn't the stored bytes' (16-bit text). + ccp: Option, +} + +fn put16(v: &mut [u8], at: usize, x: u16) { + v[at..at + 2].copy_from_slice(&x.to_le_bytes()); +} + +fn put32(v: &mut [u8], at: usize, x: u32) { + v[at..at + 4].copy_from_slice(&x.to_le_bytes()); +} + +/// Append one FKP page for `runs` and its bin table, and return the bin table's (fc, lcb). A +/// character page has a 1-byte entry per run and CHPXs (byte count + grpprl); a paragraph page +/// has 7-byte entries and PAPXs (word count + style 0 + grpprl padded to whole words). +fn fkp(w: &mut Vec, runs: &[(u32, u32, &[u8])], papx: bool, istds: &[u16]) -> (u32, u32) { + let (Some(first), Some(last)) = (runs.first(), runs.last()) else { return (0, 0) }; + w.resize(w.len().div_ceil(512) * 512, 0); + let pn = (w.len() / 512) as u16; + let crun = runs.len(); + let entry = if papx { 7 } else { 1 }; + let mut page = [0u8; 512]; + for (k, (lo, _, _)) in runs.iter().enumerate() { + put32(&mut page, 4 * k, TEXT_AT + lo); + } + put32(&mut page, 4 * crun, TEXT_AT + last.1); + // Records from the end of the page down, each at an even offset. + let mut top = 511usize; + for (k, (_, _, g)) in runs.iter().enumerate() { + let body: Vec = if papx { + let mut b = istds.get(k).copied().unwrap_or(0).to_le_bytes().to_vec(); + b.extend_from_slice(g); + if b.len() % 2 == 1 { + b.push(0); + } + b + } else if g.is_empty() { + continue; + } else { + g.to_vec() + }; + let at = (top - 1 - body.len()) & !1; + page[at] = if papx { (body.len() / 2) as u8 } else { body.len() as u8 }; + page[at + 1..at + 1 + body.len()].copy_from_slice(&body); + page[4 * (crun + 1) + entry * k] = (at / 2) as u8; + top = at; + } + page[511] = crun as u8; + w.extend_from_slice(&page); + let mut bte = Vec::new(); + bte.extend_from_slice(&(TEXT_AT + first.0).to_le_bytes()); + bte.extend_from_slice(&(TEXT_AT + last.1).to_le_bytes()); + bte.extend_from_slice(&pn.to_le_bytes()); + let at = w.len() as u32; + w.extend_from_slice(&bte); + (at, bte.len() as u32) +} + +/// The WordDocument stream: FIB, text, then the FKP pages, font table and piece table. +fn word_stream(d: &Doc6) -> Vec { + let mut w = vec![0u8; TEXT_AT as usize]; + w.extend_from_slice(d.text); + let ccp = match (d.ccp, d.pieces) { + (Some(ccp), _) => ccp, + (None, pieces) => match pieces { + Some(p) => p.iter().map(|(len, _)| len).sum(), + None => d.text.len() as u32, + }, + }; + let mut pairs: Vec<(usize, u32, u32)> = Vec::new(); + for (pair, runs, papx) in [(12, d.runs, false), (13, d.paras, true)] { + let (fc, lcb) = fkp(&mut w, runs, papx, d.para_styles); + if lcb > 0 { + pairs.push((pair, fc, lcb)); + } + } + if !d.styles.is_empty() { + let st = stylesheet(d.styles); + pairs.push((1, w.len() as u32, st.len() as u32)); + w.extend_from_slice(&st); + } + if !d.fonts.is_empty() || d.raw_fonts.is_some() { + let mut ffn = vec![0u8, 0]; + ffn.extend_from_slice(d.raw_fonts.unwrap_or_default()); + for (charset, name) in d.fonts { + let mut e = vec![0u8, 0x16, 0x90, 0x01, *charset, 0x00]; + e.extend_from_slice(name.as_bytes()); + e.push(0); + e[0] = (e.len() - 1) as u8; + ffn.extend_from_slice(&e); + } + let size = ffn.len() as u16; + put16(&mut ffn, 0, size); + pairs.push((15, w.len() as u32, ffn.len() as u32)); + w.extend_from_slice(&ffn); + } + if let Some(pieces) = d.pieces { + let mut plc = Vec::new(); + let mut cp = 0u32; + plc.extend_from_slice(&cp.to_le_bytes()); + for (len, _) in pieces { + cp += len; + plc.extend_from_slice(&cp.to_le_bytes()); + } + for (i, (_, off)) in pieces.iter().enumerate() { + plc.extend_from_slice(&[0, 0]); + plc.extend_from_slice(&(TEXT_AT + off).to_le_bytes()); + plc.extend_from_slice(&d.prms.get(i).copied().unwrap_or(0).to_le_bytes()); + } + let mut clx = Vec::new(); + for g in d.prcs { + clx.push(0x01); + clx.extend_from_slice(&(g.len() as u16).to_le_bytes()); + clx.extend_from_slice(g); + } + clx.push(0x02); + clx.extend_from_slice(&(plc.len() as u32).to_le_bytes()); + clx.extend_from_slice(&plc); + pairs.push((33, w.len() as u32, clx.len() as u32)); + w.extend_from_slice(&clx); + } + put16(&mut w, 0x00, 0xA5DC); + put16(&mut w, 0x02, if d.nfib == 0 { 0x0065 } else { d.nfib }); + put16(&mut w, 0x06, 0x0409); + put16(&mut w, 0x0A, d.flags | if d.pieces.is_some() { 0x0004 } else { 0 }); + put16(&mut w, 0x0C, 0x0065); + put16(&mut w, 0x14, d.chse); + put32(&mut w, 0x18, TEXT_AT); + put32(&mut w, 0x1C, TEXT_AT + d.text.len() as u32); + let len = w.len() as u32; + put32(&mut w, 0x20, len); + put32(&mut w, 0x34, ccp); + for (i, fc, lcb) in pairs { + put32(&mut w, 0x58 + i * 8, fc); + put32(&mut w, 0x58 + i * 8 + 4, lcb); + } + w +} + +/// A Word 6.0/95 file holding `d`: the WordDocument stream only, as the real ones have no Table +/// stream. +fn word6_file(d: &Doc6) -> Vec { + cfb_file(&[("WordDocument", &word_stream(d))]) +} + +fn texts(d: &Doc6) -> Vec { + body_texts(&read(&word6_file(d)).expect("a Word 6.0/95 file opens")) +} + +/// Fonts 0 (Western) and 1 (Cyrillic, charset 204), as the Russian files have. +const FONTS: &[(u8, &str)] = &[(0, "Times New Roman"), (204, "Times New Roman Cyr")]; + +#[test] +fn word_6_and_95_files_open_as_paragraphs() { + for nfib in [0x0065, 0x0068] { + let d = Doc6 { nfib, text: b"Hello\rWorld\r", ..Default::default() }; + assert_eq!(texts(&d), ["Hello", "World"], "nFib 0x{nfib:04X}"); + } +} + +#[test] +fn fast_saved_text_follows_the_piece_table() { + // Stored out of order; the pieces' fcs are plain byte offsets (Word 97 would halve them). + let d = Doc6 { text: b"World\rHello\r", pieces: Some(&[(6, 6), (6, 0)]), ..Default::default() }; + assert_eq!(texts(&d), ["Hello", "World"]); +} + +#[test] +fn fast_saved_file_without_a_piece_table_is_malformed() { + let d = Doc6 { flags: 0x0004, text: b"Hello\r", ..Default::default() }; + assert!(matches!(read(&word6_file(&d)), Err(DocbinError::Malformed(_)))); +} + +#[test] +fn a_runs_font_charset_chooses_its_code_page() { + // "Da " in font 0, then 0xC4 0xE0 ("Да" in Windows-1251) in the Cyrillic font 1. + let text = b"Da \xC4\xE0\r"; + let runs: &[(u32, u32, &[u8])] = &[(0, 3, &[]), (3, 5, &[0x5D, 0x01, 0x00]), (5, 6, &[])]; + let d = Doc6 { text, fonts: FONTS, runs, ..Default::default() }; + assert_eq!(texts(&d), ["Da Да"]); +} + +/// Text in an East Asian code page (a Japanese font, charset 128: Shift-JIS) is refused, not +/// read a byte at a time into replacement characters (review of wordcraft#6345). +#[test] +fn east_asian_text_is_refused_not_lost() { + let fonts: &[(u8, &str)] = &[(0, "Times New Roman"), (128, "MS Mincho")]; + let runs: &[(u32, u32, &[u8])] = &[(0, 2, &[0x5D, 0x01, 0x00]), (2, 3, &[])]; + let d = Doc6 { text: b"\x93\xfa\r", fonts, runs, ..Default::default() }; + assert!(matches!(read(&word6_file(&d)), Err(DocbinError::NotWord(e)) if e.contains("East Asian"))); + // The same file in a Western font still opens. + let d = Doc6 { text: b"\x93\xfa\r", fonts, runs: &[], ..Default::default() }; + assert_eq!(texts(&d).len(), 1); +} + +#[test] +fn the_font_sprm_is_found_after_other_sprms() { + // As in real files: bold-like toggles, a size, a picture anchor's length-prefixed sprm. + let text = b"\xC4\xE0\r"; + for g in [&[0x55, 0x81, 0x63, 0x18, 0x00, 0x5D, 0x01, 0x00][..], &[0x75, 0x01, 0x44, 0x04, 0, 0, 0, 0, 0x5D, 0x01, 0x00][..]] { + let runs: &[(u32, u32, &[u8])] = &[(0, 2, g), (2, 3, &[])]; + let d = Doc6 { text, fonts: FONTS, runs, ..Default::default() }; + assert_eq!(texts(&d), ["Да"], "grpprl {g:02X?}"); + } +} + +#[test] +fn the_last_font_sprm_of_a_run_chooses_its_code_page() { + // Font 0, then font 1 (Cyrillic): the later sprm is the one the formatting applies. + let runs: &[(u32, u32, &[u8])] = &[(0, 2, &[0x5D, 0x00, 0x00, 0x5D, 0x01, 0x00]), (2, 3, &[])]; + let d = Doc6 { text: b"\xC4\xE0\r", fonts: FONTS, runs, ..Default::default() }; + assert_eq!(texts(&d), ["Да"]); +} + +#[test] +fn an_unknown_sprm_leaves_font_0() { + // 0x99 has no known operand length, so the rest can't be framed: font 0 (Windows-1252). + let text = b"\xC4\xE0\r"; + let runs: &[(u32, u32, &[u8])] = &[(0, 2, &[0x99, 0x5D, 0x01, 0x00]), (2, 3, &[])]; + let d = Doc6 { text, fonts: FONTS, runs, ..Default::default() }; + assert_eq!(texts(&d), ["Äà"]); +} + +#[test] +fn the_toggle_sprms_before_the_font_are_framed() { + // 0x57 (strike, a one-byte toggle in [MS-DOC]'s Prm0 isprm table) before the font sprm: the + // font must still be found, or the Cyrillic text reads as Windows-1252. + let text = b"\xC4\xE0\r"; + for op in [0x57, 0x58, 0x59, 0x5C] { + let g = [op, 0x01, 0x5D, 0x01, 0x00]; + let runs: &[(u32, u32, &[u8])] = &[(0, 2, &g), (2, 3, &[])]; + let d = Doc6 { text, fonts: FONTS, runs, ..Default::default() }; + assert_eq!(texts(&d), ["Да"], "opcode 0x{op:02X}"); + } +} + +#[test] +fn a_font_entry_too_short_for_its_charset_ends_the_font_table() { + // Entry 0 says it is 3 bytes long, which can't hold the charset at offset 4. The bytes after + // it (the next entry, whose flags byte is 0xCC) must not be read as font 0's charset. + let raw: &[u8] = &[0x02, 0x16, 0x90, 0x05, 0xCC, 0x90, 0x01, 0xCC, 0x00, b'X', 0]; + let d = Doc6 { text: b"\xC4\xE0\r", raw_fonts: Some(raw), runs: &[(0, 3, &[])], ..Default::default() }; + assert_eq!(texts(&d), ["Äà"], "font 0 has no charset: Windows-1252"); +} + +#[test] +fn a_bin_table_naming_one_page_many_times_reads_it_once() { + // A hostile bin table: 1,000,000 entries, all naming one page of 50 Cyrillic runs. The text still + // reads right; that the page is read once is pinned in `word6::tests`. + let text: Vec = std::iter::repeat_n(0xC4u8, 50).chain(*b"\r").collect(); + let g: &[u8] = &[0x5D, 0x01, 0x00]; + let runs: Vec<(u32, u32, &[u8])> = (0..50).map(|i| (i, i + 1, g)).chain([(50, 51, &[][..])]).collect(); + let mut w = word_stream(&Doc6 { text: &text, fonts: FONTS, runs: &runs, ..Default::default() }); + let u32_at = |w: &[u8], at: usize| u32::from_le_bytes([w[at], w[at + 1], w[at + 2], w[at + 3]]); + let bte_fc = u32_at(&w, 0x58 + 12 * 8) as usize; + let pn = u16::from_le_bytes([w[bte_fc + 8], w[bte_fc + 9]]); + let n = 1_000_000usize; + let mut bte = Vec::with_capacity(4 * (n + 1) + 2 * n); + bte.extend_from_slice(&TEXT_AT.to_le_bytes()); + for _ in 0..n { + bte.extend_from_slice(&(TEXT_AT + 51).to_le_bytes()); + } + for _ in 0..n { + bte.extend_from_slice(&pn.to_le_bytes()); + } + let at = w.len() as u32; + w.extend_from_slice(&bte); + put32(&mut w, 0x58 + 12 * 8, at); + put32(&mut w, 0x58 + 12 * 8 + 4, bte.len() as u32); + let doc = read(&cfb_file(&[("WordDocument", &w)])).expect("opens"); + assert_eq!(body_texts(&doc), ["Đ”".repeat(50)]); +} + +/// `s` as UTF-16LE bytes. +fn utf16(s: &str) -> Vec { + s.encode_utf16().flat_map(u16::to_le_bytes).collect() +} + +#[test] +fn a_file_with_extended_characters_holds_utf16_text() { + // fExtChar (0x1000): Word reads the text as 16-bit units. + let text = utf16("Да ok\r"); + let d = Doc6 { flags: 0x1000, text: &text, ccp: Some(6), ..Default::default() }; + assert_eq!(texts(&d), ["Да ok"]); +} + +#[test] +fn utf16_text_stops_at_the_end_of_its_bytes() { + // Word reads no more 16-bit characters than the text's byte range (0x18..0x1C) holds, even + // when the character count says more. + let text = utf16("Hi\r"); + let d = Doc6 { flags: 0x1000, text: &text, ccp: Some(50), ..Default::default() }; + assert_eq!(texts(&d), ["Hi"]); +} + +#[test] +fn text_stops_at_the_end_of_its_bytes() { + // The same for 8-bit text: Word read 19 characters of a file whose count said 40. The font + // table after the text must not be read as body text. + let d = Doc6 { text: b"Hi\r", ccp: Some(50), fonts: FONTS, ..Default::default() }; + assert_eq!(texts(&d), ["Hi"]); +} + +#[test] +fn fast_saved_utf16_pieces_are_at_plain_byte_offsets() { + let text = utf16("World\rHello\r"); + let d = Doc6 { flags: 0x1000, text: &text, pieces: Some(&[(6, 12), (6, 0)]), ..Default::default() }; + assert_eq!(texts(&d), ["Hello", "World"]); +} + +#[test] +fn a_macintosh_document_reads_as_mac_roman() { + // FIB 0x14 = 0x0100: byte 0x8E is "Ă©" (Mac Roman), not "Ž" (Windows-1252), even in a run whose + // font is Cyrillic. + let runs: &[(u32, u32, &[u8])] = &[(0, 2, &[0x5D, 0x01, 0x00]), (2, 3, &[])]; + let d = Doc6 { chse: 0x0100, text: b"H\x8E\r", fonts: FONTS, runs, ..Default::default() }; + assert_eq!(texts(&d), ["HĂ©"]); +} + +#[test] +fn a_symbol_document_maps_characters_to_private_use() { + // FIB 0x14 = 0x0002: every character from the space up is U+F000 + byte; marks stay marks. + let d = Doc6 { chse: 0x0002, text: b"Hi\x8E\rA\r", ..Default::default() }; + assert_eq!(texts(&d), ["\u{F048}\u{F069}\u{F08E}", "\u{F041}"]); +} + +#[test] +fn another_character_set_leaves_only_ascii_readable() { + let d = Doc6 { chse: 0x0001, text: b"H\x8Ei\r", ..Default::default() }; + assert_eq!(texts(&d), ["H\u{FFFD}i"]); +} + +#[test] +fn a_prm_on_the_marks_piece_formats_its_paragraph() { + // As Word 95 fast-saves a 1" left indent on the first paragraph: the piece holding its mark + // has Prm1 -> Prc `11 a0 05` (left indent 1440), and its PAPX says nothing. Word shows 72 pt; + // the second paragraph keeps none. + let text = b"First\rSecond\r"; + let pieces: &[(u32, u32)] = &[(5, 0), (1, 5), (7, 6)]; + let d = Doc6 { text, pieces: Some(pieces), prms: &[0, 0x0001], prcs: &[&[0x11, 0xA0, 0x05]], ..Default::default() }; + let doc = read(&word6_file(&d)).expect("opens"); + let paras: Vec<_> = doc.body.iter().filter_map(|b| b.as_para()).collect(); + assert_eq!(paras.iter().map(|p| p.text.as_str()).collect::>(), ["First", "Second"]); + assert_eq!((paras[0].props.indent_left, paras[1].props.indent_left), (Some(72.0), None)); +} + +#[test] +fn a_pieces_prm_formats_its_text() { + // As Word 95 fast-saves formatting: piece 0 has Prm1 -> a Prc setting font 1 and bold, which + // its character run doesn't. Piece 0 shows in font 1, bold (and decodes in its code page); + // piece 1 keeps the run's plain formatting. + let text = b"\xC4\xE0\xC4\xE0\r"; + let d = + Doc6 { text, fonts: FONTS, pieces: Some(&[(2, 0), (3, 2)]), prms: &[0x0001], prcs: &[&[0x5D, 0x01, 0x00, 0x55, 0x01]], ..Default::default() }; + let doc = read(&word6_file(&d)).expect("opens"); + let p = first_para(&doc); + assert_eq!(p.text, "ДаÄà"); + let first = &p.runs.first().expect("a run").props; + assert_eq!((first.font.as_deref(), first.bold), (Some("Times New Roman Cyr"), Some(true))); + let last = &p.runs.last().expect("a run").props; + assert_eq!((last.font.as_deref(), last.bold), (None, None)); +} + +#[test] +fn a_font_in_a_pieces_prm_chooses_its_code_page() { + // As Word 95 fast-saves a font change: piece 0 ("\xC4\xE0") has Prm1 -> Prc 0 = sprm 0x5D font 1 + // (Cyrillic), while its character run names no font. Piece 1 keeps the run's font 0. + let text = b"\xC4\xE0\xC4\xE0\r"; + let d = Doc6 { text, fonts: FONTS, pieces: Some(&[(2, 0), (3, 2)]), prms: &[0x0001], prcs: &[&[0x5D, 0x01, 0x00]], ..Default::default() }; + assert_eq!(texts(&d), ["ДаÄà"]); +} + +#[test] +fn a_prm_index_past_the_prcs_is_ignored() { + let d = Doc6 { text: b"\xC4\xE0\r", fonts: FONTS, pieces: Some(&[(3, 0)]), prms: &[0x0011], prcs: &[&[0x5D, 0x01, 0x00]], ..Default::default() }; + assert_eq!(texts(&d), ["Äà"]); +} + +#[test] +fn password_protected_files_are_refused() { + let d = Doc6 { flags: 0x0100, text: b"secret\r", ..Default::default() }; + assert!(matches!(read(&word6_file(&d)), Err(DocbinError::Encryption(_)))); +} + +/// The first paragraph of `doc`'s body. +fn first_para(doc: &craft_wordmodel::Document) -> &craft_wordmodel::para::Paragraph { + doc.body.iter().find_map(|b| b.as_para()).expect("a paragraph") +} + +#[test] +fn character_formatting_is_read() { + // Bold on, italic as the opposite of the (unread, so plain) style, 14 pt, red, single + // underline, font 2: the Word 6 opcodes as observed. + let g: &[u8] = &[0x55, 0x01, 0x56, 0x81, 0x63, 0x1C, 0x00, 0x62, 0x06, 0x5E, 0x01, 0x5D, 0x02, 0x00]; + let runs: &[(u32, u32, &[u8])] = &[(0, 2, g), (2, 3, &[])]; + let fonts = &[(0, "Times New Roman"), (2, "Symbol"), (0, "Arial")]; + let doc = read(&word6_file(&Doc6 { text: b"Hi\r", fonts, runs, ..Default::default() })).expect("opens"); + let p = first_para(&doc); + let c = &p.runs.first().expect("a run").props; + assert_eq!((c.bold, c.italic, c.size), (Some(true), Some(true), Some(14.0))); + assert_eq!(c.color, Some(craft_wordmodel::props::TextColor::Rgb(craft_wordmodel::props::Rgb(0xFF, 0, 0)))); + assert_eq!(c.underline, Some(craft_wordmodel::props::Underline::Single)); + assert_eq!(c.font.as_deref(), Some("Arial")); +} + +#[test] +fn unseen_operand_values_are_left_unread() { + // Underline 3 and "multiple" line spacing were never seen in a real file, so they aren't + // guessed at. + let runs: &[(u32, u32, &[u8])] = &[(0, 3, &[0x5E, 0x03])]; + let paras: &[(u32, u32, &[u8])] = &[(0, 3, &[0x14, 0xF0, 0x00, 0x01, 0x00])]; + let doc = read(&word6_file(&Doc6 { text: b"Hi\r", runs, paras, ..Default::default() })).expect("opens"); + let p = first_para(&doc); + assert_eq!(p.runs.first().and_then(|r| r.props.underline), None); + assert_eq!(p.props.line_spacing, None); +} + +#[test] +fn an_unseen_alignment_value_is_left_unread() { + // Only 0-3 were seen; 4 would otherwise become "distribute" and 0xFF "left". + for jc in [0x04u8, 0xFF] { + let g = [0x05, jc]; + let paras: &[(u32, u32, &[u8])] = &[(0, 3, &g)]; + let doc = read(&word6_file(&Doc6 { text: b"Hi\r", paras, ..Default::default() })).expect("opens"); + assert_eq!(first_para(&doc).props.align, None, "jc {jc:#04x}"); + } +} + +#[test] +fn paragraph_formatting_is_read() { + // Justified, 0.5" left and 0.25" right indents, a 0.5" hanging first line, at least 27 pt. + let g: &[u8] = &[0x05, 0x03, 0x11, 0xD0, 0x02, 0x10, 0x68, 0x01, 0x13, 0x30, 0xFD, 0x14, 0x1C, 0x02, 0x00, 0x00]; + let paras: &[(u32, u32, &[u8])] = &[(0, 3, g), (3, 6, &[0x05, 0x01])]; + let doc = read(&word6_file(&Doc6 { text: b"Hi\rHo\r", paras, ..Default::default() })).expect("opens"); + let ps: Vec<_> = doc.body.iter().filter_map(|b| b.as_para()).collect(); + let p = &ps[0].props; + assert_eq!(p.align, Some(craft_wordmodel::props::Align::Justify)); + assert_eq!((p.indent_left, p.indent_right, p.indent_first), (Some(36.0), Some(18.0), Some(-36.0))); + assert_eq!(p.line_spacing, Some(craft_wordmodel::props::LineSpacing::AtLeast(27.0))); + assert_eq!(ps[1].props.align, Some(craft_wordmodel::props::Align::Center)); +} + +/// A Word 6 table definition sprm (0xBE): the cell edges, then a 10-byte descriptor per cell +/// whose first word holds the merge flags. +fn def_table(edges: &[i16], rgfs: &[u16]) -> Vec { + let mut rest = vec![rgfs.len() as u8]; + for e in edges { + rest.extend_from_slice(&e.to_le_bytes()); + } + for r in rgfs { + rest.extend_from_slice(&r.to_le_bytes()); + rest.extend_from_slice(&[0; 8]); + } + let mut v = vec![0xBE]; + v.extend_from_slice(&((rest.len() + 1) as u16).to_le_bytes()); + v.extend_from_slice(&rest); + v +} + +/// The rows of the first table in `doc`'s body, each as its cells' texts. +fn table_rows(doc: &craft_wordmodel::Document) -> Vec> { + let t = doc.body.iter().find_map(|b| match b.as_ref() { + craft_wordmodel::Block::Table(t) => Some(t), + _ => None, + }); + let t = t.expect("a table"); + t.rows + .iter() + .map(|r| r.cells.iter().map(|c| c.blocks.iter().filter_map(|b| b.as_para()).map(|p| p.text.clone()).collect::>().join("|")).collect()) + .collect() +} + +#[test] +fn table_rows_and_merged_cells_are_read() { + // Row 1: "A", "B". Row 2: "C" merged across both cells. In-table marks carry 0x18, the row + // ends 0x18, 0x19 and the table definition. + let text = b"A\x07B\x07\x07C\x07\x07\x07after\r"; + let mut row1 = vec![0x18, 1, 0x19, 1]; + row1.extend(def_table(&[0, 1440, 2880], &[0, 0])); + let mut row2 = vec![0x18, 1, 0x19, 1]; + row2.extend(def_table(&[0, 1440, 2880], &[0x0001, 0x0002])); + let cell: &[u8] = &[0x18, 1]; + let paras: &[(u32, u32, &[u8])] = &[(0, 2, cell), (2, 4, cell), (4, 5, &row1), (5, 7, cell), (7, 8, cell), (8, 9, &row2), (9, 15, &[])]; + let doc = read(&word6_file(&Doc6 { text, paras, ..Default::default() })).expect("opens"); + assert_eq!(table_rows(&doc), [vec!["A", "B"], vec!["C"]]); + assert_eq!(body_texts(&doc), ["after"], "the paragraph after the table stays outside it"); +} + +#[test] +fn several_merges_in_one_row_each_keep_their_cells() { + // Five cells merged as [A B] [C] [D E]: the second merge must not be shifted by the first. + let text = b"A\x07B\x07C\x07D\x07E\x07\x07"; + let mut row = vec![0x18, 1, 0x19, 1]; + row.extend(def_table(&[0, 100, 200, 300, 400, 500], &[0x0001, 0x0002, 0x0000, 0x0001, 0x0002])); + let cell: &[u8] = &[0x18, 1]; + let paras: &[(u32, u32, &[u8])] = &[(0, 2, cell), (2, 4, cell), (4, 6, cell), (6, 8, cell), (8, 10, cell), (10, 11, &row)]; + let doc = read(&word6_file(&Doc6 { text, paras, ..Default::default() })).expect("opens"); + assert_eq!(table_rows(&doc), [vec!["A|B", "C", "D|E"]]); +} + +#[test] +fn a_field_shows_its_cached_result() { + let d = Doc6 { text: b"Date \x13DATE \x1412/12/97\x15\r", ..Default::default() }; + let text = read(&word6_file(&d)).expect("opens").plain_text(craft_wordmodel::StoryRef::Body); + assert!(text.contains("Date 12/12/97"), "{text:?}"); + assert!(!text.contains("DATE"), "{text:?}"); +} + +/// A fixture's paragraph styles: Normal, an empty slot, then "Mine" (user-defined, centred, bold, on +/// Normal) and "Heading" stored under a local name with sti 1. +const STYLES: &[Option] = &[ + Some(Std { sti: 0, sgc: 1, base: 0x0FFF, name: "Standard", papx: &[], chpx: &[0x61, 0x09, 0x04] }), + None, + Some(Std { sti: 0x0FFE, sgc: 1, base: 0, name: "Mine", papx: &[0x05, 0x01], chpx: &[0x55, 0x01] }), + Some(Std { sti: 1, sgc: 1, base: 0, name: "Ăœberschrift 1", papx: &[0x08, 0x01], chpx: &[0x55, 0x81, 0x63, 0x20, 0x00] }), +]; + +/// Each paragraph's style name, through the document's style sheet. +fn style_names(doc: &craft_wordmodel::Document) -> Vec { + doc.body + .iter() + .filter_map(|b| b.as_para()) + .map(|p| p.props.style.as_deref().and_then(|id| doc.styles.get(id)).map_or(String::new(), |s| s.name.clone())) + .collect() +} + +#[test] +fn paragraphs_take_their_styles() { + let paras: &[(u32, u32, &[u8])] = &[(0, 2, &[]), (2, 4, &[]), (4, 6, &[])]; + let d = Doc6 { text: b"A\rB\rC\r", paras, para_styles: &[0, 2, 3], styles: STYLES, ..Default::default() }; + let doc = read(&word6_file(&d)).expect("opens"); + // A built-in style goes by its built-in name, whatever the file stores (as in Word). + assert_eq!(style_names(&doc), ["Normal", "Mine", "Heading 1"]); +} + +#[test] +fn styles_carry_their_formatting_and_base() { + let d = Doc6 { text: b"A\r", styles: STYLES, ..Default::default() }; + let doc = read(&word6_file(&d)).expect("opens"); + let mine = doc.styles.styles.iter().find(|s| s.name == "Mine").expect("Mine"); + assert_eq!(mine.para.align, Some(craft_wordmodel::props::Align::Center)); + assert_eq!(mine.chr.bold, Some(true)); + assert_eq!(mine.based_on.as_deref().and_then(|id| doc.styles.get(id)).map(|s| s.name.as_str()), Some("Normal")); + let heading = doc.styles.styles.iter().find(|s| s.name == "Heading 1").expect("Heading 1"); + assert_eq!((heading.para.keep_next, heading.chr.size), (Some(true), Some(16.0))); +} + +#[test] +fn bulleted_styles_frame_past_their_numbering() { + // 0x0D (1 byte) and 0x0C (a count byte and its bytes) before the indents, as in the corpus. + let mut papx = vec![0x05, 0x03, 0x0D, 0x0B, 0x11, 0x1C, 0x01, 0x13, 0xE4, 0xFE, 0x0C, 4, 0xFF, 0x01, 0x00, 0x08]; + papx.extend_from_slice(&[0x0E, 0x01, 0x2C, 0x00, 0x1D, 0x60, 0x1A, 0xF8, 0xFF, 0x25, 0x02, 0x10, 0x68, 0x01]); + let styles = &[Some(Std { sgc: 1, base: 0x0FFF, name: "Normal", papx: &papx, ..Default::default() })]; + let doc = read(&word6_file(&Doc6 { text: b"A\r", styles, ..Default::default() })).expect("opens"); + let normal = doc.styles.get("Normal").expect("Normal"); + assert_eq!((normal.para.indent_left, normal.para.indent_first, normal.para.indent_right), (Some(14.2), Some(-14.2), Some(18.0))); +} + +#[test] +fn a_runs_language_or_its_styles_chooses_the_code_page_without_a_font() { + // No font named anywhere: the style's language decides (Russian -> Windows-1251), as in Word. + let styles = &[Some(Std { sgc: 1, base: 0x0FFF, name: "Normal", chpx: &[0x61, 0x19, 0x04], ..Default::default() })]; + let d = Doc6 { text: b"\xC4\xE0\r", fonts: FONTS, styles, ..Default::default() }; + assert_eq!(texts(&d), ["Да"]); + // The run's own language outranks the style's. + let runs: &[(u32, u32, &[u8])] = &[(0, 3, &[0x61, 0x09, 0x04])]; + let d = Doc6 { text: b"\xC4\xE0\r", fonts: FONTS, styles, runs, ..Default::default() }; + assert_eq!(texts(&d), ["Äà"]); +} + +#[test] +fn font_0_is_not_a_fallback() { + // Font 0 is Cyrillic, but nothing names it and the style's language is English: Word reads + // Windows-1252. + let fonts = &[(204, "Times New Roman Cyr"), (0, "Arial")]; + let styles = &[Some(Std { sgc: 1, base: 0x0FFF, name: "Normal", chpx: &[0x61, 0x09, 0x04], ..Default::default() })]; + // A character run that names no font (where the reader used to take font 0). + let runs: &[(u32, u32, &[u8])] = &[(0, 3, &[])]; + let d = Doc6 { text: b"\xC4\xE0\r", fonts, styles, runs, ..Default::default() }; + assert_eq!(texts(&d), ["Äà"]); +} + +#[test] +fn a_font_named_in_the_style_chooses_the_code_page() { + // The base style names the Cyrillic font 1; the run and its own style name none. + let styles = &[ + Some(Std { sgc: 1, base: 0x0FFF, name: "Normal", chpx: &[0x5D, 0x01, 0x00], ..Default::default() }), + Some(Std { sti: 0x0FFE, sgc: 1, base: 0, name: "Child", ..Default::default() }), + ]; + let paras: &[(u32, u32, &[u8])] = &[(0, 3, &[])]; + let d = Doc6 { text: b"\xC4\xE0\r", fonts: FONTS, styles, paras, para_styles: &[1], ..Default::default() }; + assert_eq!(texts(&d), ["Да"]); +} + +#[test] +fn an_ansi_font_leaves_the_code_page_to_the_language() { + // Arial (ANSI) named on the run, the style's language Russian: Word reads Cyrillic. + let fonts = &[(0, "Times New Roman"), (0, "Arial")]; + let styles = &[Some(Std { sgc: 1, base: 0x0FFF, name: "Normal", chpx: &[0x61, 0x19, 0x04], ..Default::default() })]; + let runs: &[(u32, u32, &[u8])] = &[(0, 2, &[0x5D, 0x01, 0x00]), (2, 3, &[])]; + let d = Doc6 { text: b"\xC4\xE0\r", fonts, styles, runs, ..Default::default() }; + assert_eq!(texts(&d), ["Да"]); +} + +/// A fixture with every structure read: runs, fonts and a piece table. +fn full_stream() -> Vec { + let runs: &[(u32, u32, &[u8])] = &[(0, 6, &[0x55, 0x81]), (6, 9, &[0x5D, 0x01, 0x00]), (9, 12, &[])]; + let styles = &[ + Some(Std { sgc: 1, base: 0x0FFF, name: "Normal", chpx: &[0x61, 0x09, 0x04], ..Default::default() }), + None, + Some(Std { sti: 0x0FFE, sgc: 1, base: 0, name: "Mine", papx: &[0x05, 0x01], chpx: &[0x55, 0x01] }), + ]; + let paras: &[(u32, u32, &[u8])] = &[(0, 6, &[]), (6, 12, &[])]; + word_stream(&Doc6 { + text: b"Hello\r\xC4\xE0\rx\r\r", + fonts: FONTS, + runs, + paras, + para_styles: &[2, 0], + styles, + pieces: Some(&[(6, 0), (6, 6)]), + ..Default::default() + }) +} + +#[test] +fn the_full_fixture_reads() { + let f = cfb_file(&[("WordDocument", &full_stream())]); + let doc = read(&f).expect("opens"); + assert_eq!(body_texts(&doc), ["Hello", "Да", "x", ""]); + assert_eq!(style_names(&doc)[0], "Mine"); +} + +proptest! { + #![proptest_config(ProptestConfig::with_cases(256))] + + #[test] + fn corrupt_word6_streams_never_panic(edits in proptest::collection::vec((0..0x700usize, any::()), 1..16)) { + let mut w = full_stream(); + for (at, b) in edits { + if let Some(x) = w.get_mut(at) { + *x = b; + } + } + let _ = read(&cfb_file(&[("WordDocument", &w)])); + } + + #[test] + fn truncated_word6_streams_never_panic(n in 0..0x700usize) { + let w = full_stream(); + let _ = read(&cfb_file(&[("WordDocument", &w[..n.min(w.len())])])); + } +} diff --git a/crates/doc-binary/src/word6.rs b/crates/doc-binary/src/word6.rs new file mode 100644 index 000000000..e9fab2d74 --- /dev/null +++ b/crates/doc-binary/src/word6.rs @@ -0,0 +1,842 @@ +//! Word 6.0 and Word 95 `.doc` files (`wIdent` 0xA5DC, `nFib` 0x0065 and 0x0068). +//! +//! No public specification describes this version: [MS-DOC] covers Word 97 and later. Everything +//! here is observed, from the layout of real files checked against what Word shows when it opens +//! them (its text through COM, its formatting in the `.docx` Word saves from them), plus the Word +//! 97 structures in [MS-DOC] that these files visibly share. Nothing else is assumed. +//! +//! What is read: the main text, its paragraphs and fields, the code page of each run, the +//! character and paragraph formatting listed below, table rows, and the paragraph styles (with +//! their formatting). Character styles, sections, headers and footers, notes, pictures, borders, +//! shading and tabs are not read yet. +//! +//! The layout as observed: +//! +//! - A compound file with a `WordDocument` stream and **no Table stream**: the structures that +//! Word 97 keeps in `0Table`/`1Table` live in the WordDocument stream itself. +//! - The FIB is one fixed block. The flags at 0x0A have the Word 97 meaning for bit 2 (fast-saved, +//! set exactly in the files that carry a piece table) and, as far as can be told, bit 8 +//! (encrypted). The text starts at the u32 at 0x18, and the main text's length in characters is +//! the u32 at 0x34. +//! - From 0x58, (u32 fc, u32 lcb) pairs, in the order of Word 97's `FibRgFcLcb97`. Index 12 is the +//! character bin table, 13 the paragraph bin table, 15 the font table and 33 the piece table. +//! - Text is one byte per character, in the code page of the font of its run. The FIB's language +//! does not choose it: files marked English hold Russian and Czech text that Word shows as such. +//! - Both bin tables hold 2-byte page numbers (Word 97: 4 bytes). The FKP pages have the Word 97 +//! layout, except that a paragraph page's entries are 7 bytes (an offset byte and 6 more; Word +//! 97: 13), and a PAPX is a count of 2-byte words followed by exactly that many words: the +//! 2-byte style index, then the grpprl (padded with one 0 byte to whole words). +//! - Sprms have **1-byte opcodes**. Those with a 1-byte operand match the opcodes [MS-DOC]'s +//! `Prm0` table lists for the same properties (bold 0x55, colour 0x62, in-table 0x18…); the +//! others were identified against Word. See [`chp`] and [`pap`]. +//! - A font table entry is a size byte (the entry's length − 1), a flags byte, a 2-byte weight, the +//! charset byte, one more byte, then the name in 8-bit characters. +//! - The stylesheet (pair 1) has the Word 97 shape with smaller parts: a u16 header size (14), +//! then a header of that many bytes whose first two words are the style count and the size of +//! a style's fixed part (8, Word 97: 10); +//! then per style a u16 size (0: an empty slot) and that many bytes: the fixed part (sti, then +//! kind and base style, then UPX count and next style, then a size), the name (a length byte, +//! 8-bit characters, a 0), and the UPXs, each a u16 size and its bytes, padded to even. A +//! paragraph style's first UPX is its style index and paragraph sprms, the second its character +//! sprms; a character style has one, its character sprms. + +use craft_codepage::charset_code_page; + +use crate::DocbinError; +use crate::fkp::{Bins, TranslatedRun}; +use crate::piece::{CodePageRun, PieceTable, SYMBOL_CODE_PAGE, Word6Text}; +use crate::stsh::RawStyle; + +/// `wIdent` of a Word 6.0/95 file. +const W_IDENT: u16 = 0xA5DC; +const OFF_FLAGS: usize = 0x0A; +/// Fast-saved: the text is in pieces listed by a piece table. +const F_COMPLEX: u16 = 0x0004; +/// Password-protected (Word 97's bit; no unprotected file sets it). +const F_ENCRYPTED: u16 = 0x0100; +/// Word 97's `fExtChar` bit. Word reads a Word 6.0/95 file that sets it as UTF-16 text (each +/// piece's fc a plain byte offset), and so does this reader. +const F_EXT_CHAR: u16 = 0x1000; +/// The document's character set (u16): 0 Windows, 0x0100 Macintosh (all 8-bit text is Mac Roman, +/// whatever the fonts say), 0x0002 Symbol (every byte a private-use character). Word reads the +/// bytes above ASCII of any other value as U+FFFD. +const OFF_CHSE: usize = 0x14; +const OFF_FC_MAC: usize = 0x1C; +/// Bin table entries read, as for Word 97 (`fkp::MAX_BINS`). +const MAX_BINS: usize = 100_000; +/// FKP runs read (character or paragraph): far more than any real file's, and few enough to stay small. +const MAX_RUNS: usize = 1_000_000; +const OFF_FC_MIN: usize = 0x18; +const OFF_CCP_TEXT: usize = 0x34; +const OFF_PAIRS: usize = 0x58; +/// Pairs read: up to and including the piece table's. +const PAIRS: usize = 34; +const PAIR_STSH: usize = 1; +const PAIR_BTE_CHPX: usize = 12; +const PAIR_BTE_PAPX: usize = 13; +const PAIR_FONTS: usize = 15; +const PAIR_CLX: usize = 33; +/// More characters than Word's own documents get near: the FIB is corrupt (as for Word 97). +const MAX_CCP: u32 = 100_000_000; +/// The font table's entries: more than any real file has. +const MAX_FONTS: usize = 0x1_0000; +/// FKP page size in bytes. +const PAGE: usize = 512; +/// Size of a paragraph FKP entry (offset byte + 6). +const PAPX_BX: usize = 7; +/// Most cells a row's table definition is read for (the model's own cap is 63). +const MAX_CELLS: usize = 63; +/// Styles read: style indexes stop below 0x0FFF ("no style"). +const MAX_STYLES: usize = 0x0FFE; +/// Base-style links followed when a style inherits a font or language. +const MAX_STYLE_DEPTH: usize = 16; + +/// Character sprms read, by Word 6.0/95 opcode. Each one was matched against Word's own reading of +/// the corpus files (the `.docx` it saves from them). +mod chp { + /// Bold, a toggle (0 off, 1 on, 0x80 as the style, 0x81 the opposite of the style). + pub(super) const BOLD: u8 = 0x55; + /// Italic, a toggle. + pub(super) const ITALIC: u8 = 0x56; + /// Small capitals, a toggle. + pub(super) const SMALL_CAPS: u8 = 0x5A; + /// All capitals, a toggle. + pub(super) const CAPS: u8 = 0x5B; + /// The font: a 2-byte index into the font table. + pub(super) const FONT: u8 = 0x5D; + /// Underline: 0 none, 1 single (the only values seen). + pub(super) const UNDERLINE: u8 = 0x5E; + /// Colour: an index into Word's 16-colour palette, as Word 97's `Ico`. + pub(super) const COLOR: u8 = 0x62; + /// Size in half-points (2 bytes). + pub(super) const SIZE: u8 = 0x63; + /// The language (2 bytes, a Windows LCID): with no non-ANSI font, it gives the code page. + pub(super) const LID: u8 = 0x61; +} + +/// Paragraph and table sprms read, by Word 6.0/95 opcode, matched like [`chp`]'s. +mod pap { + /// Alignment: 0 left, 1 centred, 2 right, 3 justified. + pub(super) const JC: u8 = 0x05; + /// Keep with the next paragraph (1 byte; on the heading styles H1-H6, which Word shows with + /// keepNext; [MS-DOC]'s isprm table has sprmPFKeepFollow at 0x08). + pub(super) const KEEP_NEXT: u8 = 0x08; + /// Right indent in twips (signed, 2 bytes). + pub(super) const RIGHT: u8 = 0x10; + /// Left indent in twips. + pub(super) const LEFT: u8 = 0x11; + /// First-line indent in twips, relative to the left one (negative: hanging). + pub(super) const FIRST: u8 = 0x13; + /// Line spacing: the 4-byte Word 97 `LSPD` (dyaLine, fMultLinespace). + pub(super) const LINE: u8 = 0x14; + /// The paragraph is in a table. + pub(super) const IN_TABLE: u8 = 0x18; + /// The mark ends a table row. + pub(super) const TTP: u8 = 0x19; + /// Row alignment (2 bytes): 1 centred. + pub(super) const ROW_JC: u8 = 0xB6; + /// Row height in twips (signed, 2 bytes): positive at least, negative exactly. + pub(super) const ROW_HEIGHT: u8 = 0xBD; + /// The row's cells: u16 count, cell count, (count + 1) i16 cell edges, then a 10-byte + /// descriptor per cell whose first word's bit 0 starts a horizontal merge and bit 1 continues + /// it. + pub(super) const DEF_TABLE: u8 = 0xBE; +} + +/// Word 97 opcodes the translations produce ([MS-DOC] §2.6). +mod w97 { + pub(super) const C_BOLD: u16 = 0x0835; + pub(super) const C_ITALIC: u16 = 0x0836; + pub(super) const C_SMALL_CAPS: u16 = 0x083A; + pub(super) const C_CAPS: u16 = 0x083B; + pub(super) const C_KUL: u16 = 0x2A3E; + pub(super) const C_ICO: u16 = 0x2A42; + pub(super) const C_HPS: u16 = 0x4A43; + pub(super) const C_FTC0: u16 = 0x4A4F; + pub(super) const P_JC80: u16 = 0x2403; + pub(super) const P_KEEP_FOLLOW: u16 = 0x2406; + pub(super) const P_DXA_RIGHT80: u16 = 0x840E; + pub(super) const P_DXA_LEFT80: u16 = 0x840F; + pub(super) const P_DXA_LEFT1_80: u16 = 0x8411; + pub(super) const P_DYA_LINE: u16 = 0x6412; + pub(super) const P_IN_TABLE: u16 = 0x2416; + pub(super) const P_TTP: u16 = 0x2417; + pub(super) const T_JC: u16 = 0x5400; + pub(super) const T_ROW_HEIGHT: u16 = 0x9407; + pub(super) const T_DEF_TABLE: u16 = 0xD608; + pub(super) const T_MERGE: u16 = 0x5624; +} + +/// Is this WordDocument stream a Word 6.0/95 one? +pub(crate) fn is_word6(word: &[u8]) -> bool { + word.get(0..2) == Some(&W_IDENT.to_le_bytes()[..]) +} + +/// The fields of a Word 6.0/95 FIB that are read. +#[derive(Clone, Debug)] +pub(crate) struct Fib6 { + complex: bool, + /// Text is UTF-16 (fExtChar). + wide: bool, + chse: u16, + fc_min: u32, + fc_mac: u32, + /// Characters in the main text. + pub(crate) ccp_text: u32, + pairs: [(u32, u32); PAIRS], +} + +impl Fib6 { + pub(crate) fn parse(word: &[u8]) -> Result { + let u16_at = |at: usize| word.get(at..at + 2).map(|b| u16::from_le_bytes([b[0], b[1]])); + let u32_at = |at: usize| word.get(at..at + 4).map(|b| u32::from_le_bytes([b[0], b[1], b[2], b[3]])); + if word.len() < OFF_PAIRS + PAIRS * 8 { + return Err(DocbinError::NotWord(format!("Word 6.0/95 WordDocument stream is {} bytes", word.len()))); + } + let flags = u16_at(OFF_FLAGS).unwrap_or(0); + if flags & F_ENCRYPTED != 0 { + return Err(DocbinError::Encryption("password-protected Word 6.0/95 files can't be opened".into())); + } + let ccp_text = u32_at(OFF_CCP_TEXT).unwrap_or(0); + if ccp_text > MAX_CCP { + return Err(DocbinError::Limit(format!("document claims {ccp_text} characters"))); + } + let mut pairs = [(0, 0); PAIRS]; + for (i, p) in pairs.iter_mut().enumerate() { + let at = OFF_PAIRS + i * 8; + *p = (u32_at(at).unwrap_or(0), u32_at(at + 4).unwrap_or(0)); + } + Ok(Fib6 { + complex: flags & F_COMPLEX != 0, + wide: flags & F_EXT_CHAR != 0, + chse: u16_at(OFF_CHSE).unwrap_or(0), + fc_min: u32_at(OFF_FC_MIN).unwrap_or(0), + fc_mac: u32_at(OFF_FC_MAC).unwrap_or(0), + ccp_text, + pairs, + }) + } + + /// The (fc, lcb) pair at index `i`, if it points at anything. + fn pair(&self, i: usize) -> Option<(u32, u32)> { + self.pairs.get(i).copied().filter(|&(_, lcb)| lcb > 0) + } +} + +/// Everything the walk needs from a Word 6.0/95 file besides its text stream. +pub(crate) struct Content { + pub(crate) pieces: PieceTable, + /// Character formatting, translated to Word 97 grpprls. + pub(crate) chpx: Bins, + /// Paragraph formatting and table rows, translated to Word 97 grpprls. + pub(crate) papx: Bins, + /// Font names by index (the walk resolves translated font sprms against them). + pub(crate) fonts: Vec, + /// The styles by index, their formatting translated to Word 97 grpprls. + pub(crate) styles: Vec, +} + +/// Read the text, formatting and fonts of a Word 6.0/95 file. +/// +/// 8-bit text decodes, as Word does, in the code page of: the document's character set when it +/// isn't Windows ([`OFF_CHSE`]); else the font a piece's Prm names (as Word 95 writes when a fast +/// save changes a font) when its charset has a code page; else, the same for the font of the +/// character run, or the one its paragraph style (or a base of it) names; else the language of +/// the run, or of its paragraph style; else Windows-1252. A font is only one named somewhere: +/// font 0 is not a fallback, and a font with the ANSI charset leaves the choice to the language. +pub(crate) fn content(word: &[u8], fib: &Fib6) -> Result { + let fonts = fib.pair(PAIR_FONTS).map(|(fc, lcb)| font_table(word, fc, lcb)).unwrap_or_default(); + let chp_runs = fib.pair(PAIR_BTE_CHPX).map(|(fc, lcb)| fkp_runs(word, fc, lcb, false)).unwrap_or_default(); + let pap_runs = fib.pair(PAIR_BTE_PAPX).map(|(fc, lcb)| fkp_runs(word, fc, lcb, true)).unwrap_or_default(); + let (styles, facts) = fib.pair(PAIR_STSH).map(|(fc, lcb)| stylesheet(word, fc, lcb)).unwrap_or_default(); + let windows = fib.chse == 0; + let default_code_page = match fib.chse { + // Text that no run covers is formatted by the Normal style. + 0 => text_code_page(&fonts, &facts, 0, None, None), + 0x0100 => MAC_ROMAN, + 0x0002 => SYMBOL_CODE_PAGE, + // Not a code page: bytes above ASCII read as U+FFFD. + _ => 0, + }; + let code_pages = if windows { code_page_runs(&chp_runs, &pap_runs, &facts, &fonts, default_code_page) } else { Vec::new() }; + // shortcut: East Asian (lead and trail byte) text is refused, as all Word 6.0/95 files were + // before; read it a character at a time when the piece walk can decode byte pairs. + let east_asian = || DocbinError::NotWord("Word 6.0/95 text in an East Asian code page is not read yet".into()); + if craft_codepage::is_multibyte_code_page(default_code_page) || code_pages.iter().any(|r| craft_codepage::is_multibyte_code_page(r.code_page)) { + return Err(east_asian()); + } + let multibyte_piece = std::cell::Cell::new(false); + let text = Word6Text { wide: fib.wide, default_code_page, code_pages }; + let pieces = if fib.complex { + let (fc, lcb) = fib.pair(PAIR_CLX).ok_or_else(|| DocbinError::Malformed("the fast-saved file has no piece table".into()))?; + // A Prm1's Prc may name the piece's font (sprm 0x5D); a Prm0 can't, its operand being one + // byte. + let piece_code_page = |prm: u16, prcs: &[Vec]| -> Option { + if !windows || prm & 1 == 0 { + return None; + } + let cp = last_font(prcs.get(usize::from(prm >> 1))?).and_then(|f| font_code_page(&fonts, f)); + if cp.is_some_and(craft_codepage::is_multibyte_code_page) { + multibyte_piece.set(true); + } + cp + }; + // The Prcs then carry Word 97 sprms, as the CHPXs and PAPXs do: the walk applies their + // character sprms to the piece's text and their paragraph sprms to the paragraph whose mark + // the piece holds. + let table = PieceTable::parse_word6(word, fc, lcb, text, piece_code_page, translate_prc)?; + if multibyte_piece.get() { + return Err(east_asian()); + } + table + } else { + // Word reads no more characters than the text's byte range (fcMin..fcMac) holds, one byte + // each or two for UTF-16, whatever the character count says. + let unit = if fib.wide { 2 } else { 1 }; + let ccp = fib.ccp_text.min(fib.fc_mac.saturating_sub(fib.fc_min) / unit); + PieceTable::contiguous_word6(fib.fc_min, ccp, text) + }; + let chpx = Bins::translated(chp_runs.iter().map(|r| TranslatedRun { lo: r.lo, hi: r.hi, istd: 0, grpprl: translate_chp(r.grpprl) }).collect()); + let papx = + Bins::translated(pap_runs.iter().map(|r| TranslatedRun { lo: r.lo, hi: r.hi, istd: r.istd, grpprl: translate_pap(r.grpprl) }).collect()); + Ok(Content { pieces, chpx, papx, fonts: fonts.into_iter().map(|f| f.name).collect(), styles }) +} + +/// Mac OS Roman, the 8-bit text of a Macintosh Word document. +const MAC_ROMAN: u16 = craft_codepage::MAC_ROMAN_CODE_PAGE; + +/// The code page of font `font`'s charset, if it has one (not for ANSI, default or symbol +/// charsets, whose text Word decodes by its language). +fn font_code_page(fonts: &[Font], font: u16) -> Option { + fonts.get(usize::from(font)).and_then(|f| charset_code_page(i32::from(f.charset))) +} + +/// Windows' ANSI code page for the language `lid` (an LCID), which Word uses for 8-bit text in a +/// font without a charset of its own. Russian (1251) and English (1252) were checked against Word; +/// the rest is the standard Windows assignment. +fn lid_code_page(lid: u16) -> u16 { + match lid & 0x03FF { + // Serbian in Cyrillic (0x0C1A, 0x1C1A); Croatian and Serbian in Latin otherwise. + 0x1A => match lid { + 0x0C1A | 0x1C1A => 1251, + _ => 1250, + }, + 0x02 | 0x19 | 0x22 | 0x23 | 0x2F | 0x3F | 0x40 | 0x44 | 0x50 => 1251, + 0x05 | 0x0E | 0x15 | 0x18 | 0x1B | 0x1C | 0x24 => 1250, + 0x08 => 1253, + 0x1F => 1254, + 0x0D => 1255, + 0x01 | 0x20 | 0x29 => 1256, + 0x25..=0x27 => 1257, + 0x2A => 1258, + 0x1E => 874, + 0x11 => 932, + 0x12 => 949, + 0x04 => match lid { + 0x0804 | 0x1004 => 936, + _ => 950, + }, + _ => 1252, + } +} + +/// A style's own font and language sprms and its base, for inheritance ([`StyleFacts::resolve`]). +#[derive(Clone, Copy, Debug, Default)] +struct StyleFacts { + base: u16, + font: Option, + lid: Option, +} + +impl StyleFacts { + /// Style `istd`'s font and language, each from the style itself or the nearest base that has + /// one. + fn resolve(facts: &[StyleFacts], istd: u16) -> (Option, Option) { + let (mut font, mut lid) = (None, None); + let mut at = istd; + for _ in 0..MAX_STYLE_DEPTH { + let Some(f) = facts.get(usize::from(at)) else { break }; + font = font.or(f.font); + lid = lid.or(f.lid); + if font.is_some() && lid.is_some() { + break; + } + at = f.base; + } + (font, lid) + } +} + +/// The styles of the stylesheet at `word[fc..fc+lcb]`, by index (an empty slot is a default +/// style, which keeps the indexes aligned), with their formatting translated, and each style's +/// own font and language. Reading stops at the first entry that runs past the table. +fn stylesheet(word: &[u8], fc: u32, lcb: u32) -> (Vec, Vec) { + let Some(st) = (fc as usize).checked_add(lcb as usize).and_then(|end| word.get(fc as usize..end)) else { return Default::default() }; + let u16_at = |b: &[u8], at: usize| b.get(at..at + 2).map(|x| u16::from_le_bytes([x[0], x[1]])); + let Some(cb_header) = u16_at(st, 0).map(usize::from) else { return Default::default() }; + let (Some(cstd), Some(cb_base)) = (u16_at(st, 2), u16_at(st, 4)) else { return Default::default() }; + let cb_base = usize::from(cb_base); + let (mut styles, mut facts) = (Vec::new(), Vec::new()); + let mut at = 2 + cb_header; + for _ in 0..usize::from(cstd).min(MAX_STYLES) { + let Some(cb) = u16_at(st, at).map(usize::from) else { break }; + let Some(std) = st.get(at + 2..at + 2 + cb) else { break }; + at += 2 + cb; + let (style, fact) = style(std, cb_base).unwrap_or_else(|| (RawStyle::default(), StyleFacts { base: 0x0FFF, ..Default::default() })); + styles.push(style); + facts.push(fact); + } + (styles, facts) +} + +/// The names of the built-in styles by sti: the "Index within built-in styles" of the reserved +/// style names in [MS-OE376] (Part 4 §2.7.3.9), which [MS-DOC]'s StdfBase says sti is. 0x0FFE +/// (user-defined) and anything past the table keep the name stored in the file. +const BUILT_IN_NAMES: [&str; 107] = [ + "Normal", + "heading 1", + "heading 2", + "heading 3", + "heading 4", + "heading 5", + "heading 6", + "heading 7", + "heading 8", + "heading 9", + "index 1", + "index 2", + "index 3", + "index 4", + "index 5", + "index 6", + "index 7", + "index 8", + "index 9", + "toc 1", + "toc 2", + "toc 3", + "toc 4", + "toc 5", + "toc 6", + "toc 7", + "toc 8", + "toc 9", + "Normal Indent", + "footnote text", + "annotation text", + "header", + "footer", + "index heading", + "caption", + "table of figures", + "envelope address", + "envelope return", + "footnote reference", + "annotation reference", + "line number", + "page number", + "endnote reference", + "endnote text", + "table of authorities", + "macro", + "toa heading", + "List", + "List Bullet", + "List Number", + "List 2", + "List 3", + "List 4", + "List 5", + "List Bullet 2", + "List Bullet 3", + "List Bullet 4", + "List Bullet 5", + "List Number 2", + "List Number 3", + "List Number 4", + "List Number 5", + "Title", + "Closing", + "Signature", + "Default Paragraph Font", + "Body Text", + "Body Text Indent", + "List Continue", + "List Continue 2", + "List Continue 3", + "List Continue 4", + "List Continue 5", + "Message Header", + "Subtitle", + "Salutation", + "Date", + "Body Text First Indent", + "Body Text First Indent 2", + "Note Heading", + "Body Text 2", + "Body Text 3", + "Body Text Indent 2", + "Body Text Indent 3", + "Block Text", + "Hyperlink", + "FollowedHyperlink", + "Strong", + "Emphasis", + "Document Map", + "Plain Text", + "E-mail Signature", + "HTML Top of Form", + "HTML Bottom of Form", + "Normal (Web)", + "HTML Acronym", + "HTML Address", + "HTML Cite", + "HTML Code", + "HTML Definition", + "HTML Keyboard", + "HTML Preformatted", + "HTML Sample", + "HTML Typewriter", + "HTML Variable", + "Normal Table", + "annotation subject", +]; + +/// One style entry (`None` for an empty slot or one too short for its fixed part and name). +fn style(std: &[u8], cb_base: usize) -> Option<(RawStyle, StyleFacts)> { + let u16_at = |at: usize| std.get(at..at + 2).map(|x| u16::from_le_bytes([x[0], x[1]])); + if cb_base < 8 { + return None; + } + let (sti, kind_base, upx_next) = (u16_at(0)? & 0x0FFF, u16_at(2)?, u16_at(4)?); + let (stk, istd_base, cupx) = ((kind_base & 0x000F) as u8, kind_base >> 4, usize::from(upx_next & 0x000F)); + let len = usize::from(*std.get(cb_base)?); + // A built-in style goes by its built-in name, as in Word, whatever name the file stores + // ("Default" or "Đ�Đ±Ñ‹Ñ‡Đ½Ñ‹Đ¹" for Normal). Other names are 8-bit, as Word shows them + // (Windows-1252: names it couldn't store read back with '?'). + let stored = std.get(cb_base + 1..cb_base + 1 + len)?; + let name: String = match BUILT_IN_NAMES.get(usize::from(sti)) { + Some(built_in) => (*built_in).to_owned(), + None => stored.iter().map(|&b| craft_codepage::cp1252(b)).collect(), + }; + let mut upxs: Vec<&[u8]> = Vec::new(); + let mut j = cb_base + 1 + len + 1; + for _ in 0..cupx.min(3) { + j += j % 2; + let Some(n) = u16_at(j).map(usize::from) else { break }; + let Some(upx) = std.get(j + 2..j + 2 + n) else { break }; + upxs.push(upx); + j += 2 + n; + } + // A paragraph style: style index + paragraph sprms, then character sprms. A character style: + // character sprms. + let (papx, chpx) = match stk { + 1 => (upxs.first().and_then(|u| u.get(2..)).unwrap_or_default(), upxs.get(1).copied().unwrap_or_default()), + _ => (&[][..], upxs.first().copied().unwrap_or_default()), + }; + let fact = StyleFacts { base: istd_base, font: last_font(chpx), lid: last_lid(chpx) }; + Some((RawStyle { stk, istd_base, name, papx: translate_pap(papx), chpx: translate_chp(chpx) }, fact)) +} + +/// One font table entry. +struct Font { + name: String, + charset: u8, +} + +/// The fonts of the font table at `word[fc..fc+lcb]`, by index. A truncated entry ends the table. +fn font_table(word: &[u8], fc: u32, lcb: u32) -> Vec { + let Some(table) = (fc as usize).checked_add(lcb as usize).and_then(|end| word.get(fc as usize..end)) else { return Vec::new() }; + // The table starts with its own size, which bounds the entries along with `lcb`. + let size = table.get(0..2).map_or(0, |b| usize::from(u16::from_le_bytes([b[0], b[1]]))); + let table = table.get(..size.min(table.len())).unwrap_or_default(); + let mut fonts = Vec::new(); + let mut at = 2usize; + while let Some(&cb_m1) = table.get(at) { + let next = at + usize::from(cb_m1) + 1; + // The entry must lie inside the table and hold its charset (one too short is corrupt). + let Some(entry) = table.get(at..next) else { break }; + let (Some(&charset), Some(name)) = (entry.get(4), entry.get(6..)) else { break }; + if fonts.len() >= MAX_FONTS { + break; + } + // An 8-bit name; font names are ASCII in every file seen, other bytes read as Latin-1. + let name = name.split(|&b| b == 0).next().unwrap_or_default().iter().map(|&b| char::from(b)).collect(); + fonts.push(Font { name, charset }); + at = next; + } + fonts +} + +/// One FKP run as stored: its FC range, the paragraph style (PAPX runs) and its Word 6 grpprl. +struct Run<'w> { + lo: u32, + hi: u32, + istd: u16, + grpprl: &'w [u8], +} + +/// The page numbers of a bin table with `n` entries ((n+1) u32 FCs, then n u16 page numbers), +/// each once, however many entries name it: a page's runs carry their own FCs, so reading it +/// again would only repeat them. The entries read are capped as for Word 97. +fn distinct_pages(plc: &[u8], n: usize) -> Vec { + let mut pns: Vec = (0..n.min(MAX_BINS)) + .filter_map(|i| plc.get(4 * (n + 1) + 2 * i..4 * (n + 1) + 2 * i + 2).map(|b| usize::from(u16::from_le_bytes([b[0], b[1]])))) + .collect(); + pns.sort_unstable(); + pns.dedup(); + pns +} + +/// Every run of the FKP pages a bin table at `word[fc..fc+lcb]` lists, each page read once and +/// at most [`MAX_RUNS`] runs in all. Unreadable pages are skipped. +fn fkp_runs(word: &[u8], fc: u32, lcb: u32, papx: bool) -> Vec> { + let Some(plc) = (fc as usize).checked_add(lcb as usize).and_then(|end| word.get(fc as usize..end)) else { return Vec::new() }; + let Some(n) = plc.len().checked_sub(4).filter(|r| r % 6 == 0).map(|r| r / 6) else { return Vec::new() }; + let entry = if papx { PAPX_BX } else { 1 }; + let mut runs = Vec::new(); + for pn in distinct_pages(plc, n) { + if runs.len() >= MAX_RUNS { + break; + } + let Some(page) = pn.checked_mul(PAGE).and_then(|at| word.get(at..at.checked_add(PAGE)?)) else { continue }; + let crun = usize::from(page.get(PAGE - 1).copied().unwrap_or(0)); + // The FCs and the entries must fit before the count byte. + if 4 * (crun + 1) + entry * crun >= PAGE { + continue; + } + let fc_at = |k: usize| page.get(4 * k..4 * k + 4).map(|b| u32::from_le_bytes([b[0], b[1], b[2], b[3]])); + for k in 0..crun { + let (Some(lo), Some(hi)) = (fc_at(k), fc_at(k + 1)) else { break }; + let off = usize::from(page.get(4 * (crun + 1) + entry * k).copied().unwrap_or(0)) * 2; + let (istd, grpprl) = match (off, papx) { + (0, _) => (0, &[][..]), + // A CHPX: a byte count, then the grpprl. + (_, false) => (0, page.get(off).and_then(|&cb| page.get(off + 1..off + 1 + usize::from(cb))).unwrap_or_default()), + // A PAPX: a count of 2-byte words, then the style index and the grpprl. + (_, true) => { + let body = page.get(off).and_then(|&cw| page.get(off + 1..off + 1 + 2 * usize::from(cw))).unwrap_or_default(); + let istd = body.get(0..2).map_or(0, |b| u16::from_le_bytes([b[0], b[1]])); + (istd, body.get(2..).unwrap_or_default()) + } + }; + runs.push(Run { lo, hi, istd, grpprl }); + } + } + runs +} + +/// The sprms of a Word 6.0/95 grpprl as (opcode, operand). Reading stops at an opcode whose +/// operand length is not known, since the rest can't be framed, and at a lone trailing 0 (the +/// padding of a PAPX to whole words). +fn sprms(grpprl: &[u8]) -> impl Iterator { + let mut rest = grpprl; + std::iter::from_fn(move || { + let (&op, after) = rest.split_first()?; + let len = match op { + // 0x55-0x5C are the character toggles of [MS-DOC]'s Prm0 isprm table (bold, italic, + // strike, outline, shadow, small caps, caps, hidden), all one byte; 0x55, 0x56, 0x5A + // and 0x5B are also seen in real files. + // 0x0D, 0x0E, 0x1D, 0x25 and 0x2C are one-byte paragraph sprms of [MS-DOC]'s isprm table + // too (numbering level, no line numbers, frame anchoring, wrapping, no hyphenation). + 0x05 | 0x08 | 0x0D | 0x0E | 0x18 | 0x19 | 0x1D | 0x25 | 0x2C | 0x33 | 0x55..=0x5C | 0x5E | 0x62 | 0x68 | 0x75 | 0xB7 | 0xB9 => 1, + // 0x1A/0x1B: a frame's position; 0x6B: kerning in half-points (H1 has 0x0024; Word + // shows kern=36). + 0x10 | 0x11 | 0x13 | 0x15 | 0x16 | 0x1A | 0x1B | 0x26 | 0x27 | 0x28 | 0x29 | 0x2F | 0x50 | 0x5D | 0x60 | 0x61 | 0x63 | 0x6B | 0xB6 + | 0xB8 | 0xBD | 0xC0 => 2, + 0x14 => 4, + // A count byte, then that many bytes (0x0C: a bulleted style's numbering, 52 bytes). + 0x0C | 0x0F | 0x44 | 0xBF => 1 + usize::from(*after.first()?), + // A 2-byte count, then one byte fewer than it says (as Word 97's sprmTDefTable). + 0xBE => 2 + usize::from(u16::from_le_bytes([*after.first()?, *after.get(1)?])).saturating_sub(1), + _ => return None, + }; + let operand = after.get(..len)?; + rest = after.get(len..).unwrap_or_default(); + Some((op, operand)) + }) +} + +/// Append a Word 97 Prl: opcode, then operand. +fn prl(out: &mut Vec, op: u16, operand: &[u8]) { + out.extend_from_slice(&op.to_le_bytes()); + out.extend_from_slice(operand); +} + +/// A Word 6.0/95 CHPX grpprl as the Word 97 sprms it stands for. Sprms not read are left out. +fn translate_chp(grpprl: &[u8]) -> Vec { + let mut out = Vec::new(); + for (op, v) in sprms(grpprl) { + match op { + chp::BOLD => prl(&mut out, w97::C_BOLD, v), + chp::ITALIC => prl(&mut out, w97::C_ITALIC, v), + chp::SMALL_CAPS => prl(&mut out, w97::C_SMALL_CAPS, v), + chp::CAPS => prl(&mut out, w97::C_CAPS, v), + // Only none and single were seen; any other value is left unread. + chp::UNDERLINE if matches!(v, [0] | [1]) => prl(&mut out, w97::C_KUL, v), + chp::COLOR => prl(&mut out, w97::C_ICO, v), + chp::SIZE => prl(&mut out, w97::C_HPS, v), + chp::FONT => prl(&mut out, w97::C_FTC0, v), + _ => {} + } + } + out +} + +/// A Word 6.0/95 Prc grpprl as the Word 97 sprms it stands for: character and paragraph sprms +/// alike (Word 95 records a paragraph change before a fast save as a Prc on the mark's piece). +fn translate_prc(grpprl: &[u8]) -> Vec { + [translate_chp(grpprl), translate_pap(grpprl)].concat() +} + +/// A Word 6.0/95 PAPX grpprl as the Word 97 sprms it stands for, table rows included. Sprms not +/// read are left out. +fn translate_pap(grpprl: &[u8]) -> Vec { + let mut out = Vec::new(); + for (op, v) in sprms(grpprl) { + match op { + // Only 0-3 were seen: another value is left unread. + pap::JC if matches!(v, [0..=3]) => prl(&mut out, w97::P_JC80, v), + pap::KEEP_NEXT => prl(&mut out, w97::P_KEEP_FOLLOW, v), + pap::RIGHT => prl(&mut out, w97::P_DXA_RIGHT80, v), + pap::LEFT => prl(&mut out, w97::P_DXA_LEFT80, v), + pap::FIRST => prl(&mut out, w97::P_DXA_LEFT1_80, v), + // Only "at least" spacing was seen (a positive height, not a multiple): the other + // forms are left unread. + pap::LINE => { + if let [l0, l1, 0, 0] = v + && i16::from_le_bytes([*l0, *l1]) > 0 + { + prl(&mut out, w97::P_DYA_LINE, v); + } + } + pap::IN_TABLE => prl(&mut out, w97::P_IN_TABLE, v), + pap::TTP => prl(&mut out, w97::P_TTP, v), + pap::ROW_JC => prl(&mut out, w97::T_JC, v), + pap::ROW_HEIGHT => prl(&mut out, w97::T_ROW_HEIGHT, v), + pap::DEF_TABLE => def_table(&mut out, v), + _ => {} + } + } + out +} + +/// A Word 6.0/95 table definition as a Word 97 `sprmTDefTable` holding the cell edges (the only +/// part the table reader uses) and one `sprmTMerge` per horizontal merge. +fn def_table(out: &mut Vec, v: &[u8]) { + let Some(&itc) = v.get(2) else { return }; + let itc = usize::from(itc).min(MAX_CELLS); + let Some(edges) = v.get(3..3 + 2 * (itc + 1)) else { return }; + // Word 97 operand: cb (u16: the bytes after it, plus one), the cell count, the edges. + let cb = (1 + edges.len() + 1) as u16; + out.extend_from_slice(&w97::T_DEF_TABLE.to_le_bytes()); + out.extend_from_slice(&cb.to_le_bytes()); + out.push(itc as u8); + out.extend_from_slice(edges); + let cells = v.get(3 + edges.len()..).unwrap_or_default(); + let mut merges: Vec<(usize, usize)> = Vec::new(); + let mut start: Option = None; + for i in 0..=itc { + let rgf = cells.get(10 * i..10 * i + 2).map_or(0, |b| u16::from_le_bytes([b[0], b[1]])); + let continues = i < itc && rgf & 0x0002 != 0 && rgf & 0x0001 == 0; + if continues { + continue; + } + // A merge runs from its first cell through every cell that continues it. + if let Some(a) = start.take() + && i > a + 1 + { + merges.push((a, i)); + } + if i < itc && rgf & 0x0001 != 0 { + start = Some(i); + } + } + // Right to left: the table reader folds each merge's cells into its first one as it goes, so + // a merge further right must be applied before one to its left moves its cells. + for (a, b) in merges.into_iter().rev() { + prl(out, w97::T_MERGE, &[a as u8, b as u8]); + } +} + +/// The code page of text in paragraph style `istd` whose run names `run_font` and `run_lid`: the +/// run's font, else the style's (nearest in its base chain), when that font's charset has a code +/// page; else the run's language, else the style's; else Windows-1252. +fn text_code_page(fonts: &[Font], facts: &[StyleFacts], istd: u16, run_font: Option, run_lid: Option) -> u16 { + let (style_font, style_lid) = StyleFacts::resolve(facts, istd); + run_font.or(style_font).and_then(|f| font_code_page(fonts, f)).or_else(|| run_lid.or(style_lid).map(lid_code_page)).unwrap_or(1252) +} + +/// The language a grpprl names (sprm 0x61), if any: the last one. +fn last_lid(grpprl: &[u8]) -> Option { + sprms(grpprl).filter(|(op, _)| *op == chp::LID).filter_map(|(_, v)| Some(u16::from_le_bytes([*v.first()?, *v.get(1)?]))).last() +} + +/// The byte ranges whose code page is not the document default `default`, cut at every character and paragraph run +/// boundary, each stretch in [`text_code_page`]'s. Sorted by start and not overlapping. +fn code_page_runs(chp: &[Run], pap: &[Run], facts: &[StyleFacts], fonts: &[Font], default: u16) -> Vec { + let sorted = |runs: &[Run]| -> Vec<(u32, u32, usize)> { + let mut v: Vec<(u32, u32, usize)> = runs.iter().enumerate().filter(|(_, r)| r.lo < r.hi).map(|(i, r)| (r.lo, r.hi, i)).collect(); + v.sort_unstable(); + v + }; + let (chp_sorted, pap_sorted) = (sorted(chp), sorted(pap)); + let covering = |v: &[(u32, u32, usize)], fc: u32| -> Option { + let k = v.partition_point(|&(lo, _, _)| lo <= fc).checked_sub(1)?; + v.get(k).filter(|&&(_, hi, _)| fc < hi).map(|&(_, _, i)| i) + }; + let mut cuts: Vec = chp_sorted.iter().chain(pap_sorted.iter()).flat_map(|&(lo, hi, _)| [lo, hi]).collect(); + cuts.sort_unstable(); + cuts.dedup(); + let mut out: Vec = Vec::new(); + for w in cuts.windows(2) { + let (&lo, &hi) = (w.first().unwrap_or(&0), w.get(1).unwrap_or(&0)); + let run = covering(&chp_sorted, lo).and_then(|i| chp.get(i)); + let para = covering(&pap_sorted, lo).and_then(|i| pap.get(i)); + if run.is_none() && para.is_none() { + continue; + } + let run_font = run.and_then(|r| last_font(r.grpprl)); + let run_lid = run.and_then(|r| last_lid(r.grpprl)); + let code_page = text_code_page(fonts, facts, para.map_or(0, |p| p.istd), run_font, run_lid); + if code_page == default { + continue; + } + match out.last_mut() { + Some(last) if last.hi == lo && last.code_page == code_page => last.hi = hi, + _ => out.push(CodePageRun { lo, hi, code_page }), + } + } + out +} + +/// The font a CHPX or Prc grpprl names (sprm 0x5D), if any: the last one, as a later sprm +/// overrides an earlier one when the formatting is applied (a font before an opcode of unknown +/// length stands). +fn last_font(grpprl: &[u8]) -> Option { + sprms(grpprl).filter(|(op, _)| *op == chp::FONT).filter_map(|(_, v)| Some(u16::from_le_bytes([*v.first()?, *v.get(1)?]))).last() +} + +#[cfg(test)] +mod tests { + #[test] + fn a_page_named_by_many_bin_entries_is_read_once() { + // A hostile bin table: a million entries all naming page 7, and one naming page 3. + let n = 1_000_001usize; + let mut plc = vec![0u8; 4 * (n + 1)]; + for i in 0..n { + plc.extend_from_slice(&(if i == 5 { 3u16 } else { 7u16 }).to_le_bytes()); + } + assert_eq!(super::distinct_pages(&plc, n), [3, 7]); + } + + #[test] + fn bin_entries_past_the_cap_are_not_read() { + let n = super::MAX_BINS + 1; + let mut plc = vec![0u8; 4 * (n + 1)]; + for i in 0..n { + plc.extend_from_slice(&(if i == n - 1 { 9u16 } else { 7u16 }).to_le_bytes()); + } + assert_eq!(super::distinct_pages(&plc, n), [7]); + } +} diff --git a/crates/docpath/tests/long_path.rs b/crates/docpath/tests/long_path.rs new file mode 100644 index 000000000..2ecf08ba1 --- /dev/null +++ b/crates/docpath/tests/long_path.rs @@ -0,0 +1,29 @@ +//! Paths long enough that a per-level recursive descent would exhaust the stack. Parsing is +//! iterative — `resolve` and `classify` reach `parse_share` at a fixed depth — so these come +//! back; a refusal is fine, a crash is not. + +use craft_docpath::{classify, resolve}; + +/// Half a megabyte of components under a UNC share: one component per level for a parser that +/// recursed on each separator. +#[test] +fn a_half_megabyte_unc_path_comes_back() { + let mut path = String::from(r"\\?\UNC\host\share"); + for _ in 0..(1 << 18) { + path.push_str(r"\d"); + } + let _ = resolve(&path, None); + let _ = classify(&path); +} + +/// Sixteen thousand nested verbatim prefixes: each level was a candidate recursive call when the +/// UNC prefixes were first reviewed. +#[test] +fn nested_verbatim_prefixes_come_back() { + let mut path = String::new(); + for _ in 0..(1 << 14) { + path.push_str(r"\\?\UNC\"); + } + let _ = resolve(&path, None); + let _ = classify(&path); +} diff --git a/crates/document-bundle/Cargo.toml b/crates/document-bundle/Cargo.toml new file mode 100644 index 000000000..705ac7db5 --- /dev/null +++ b/crates/document-bundle/Cargo.toml @@ -0,0 +1,17 @@ +[package] +name = "craft-document-bundle" +description = "A reusable JSON document bundle envelope over ZIP: metadata, previews, asset entries and legacy inline asset extraction." +version.workspace = true +edition.workspace = true +license.workspace = true +rust-version.workspace = true +repository.workspace = true + +[lints] +workspace = true + +[dependencies] +craft-zip = { workspace = true } +serde_json = { workspace = true } +thiserror = { workspace = true } +zip = { version = "9", default-features = false, features = ["deflate-flate2"] } diff --git a/crates/document-bundle/src/lib.rs b/crates/document-bundle/src/lib.rs new file mode 100644 index 000000000..70a960a1f --- /dev/null +++ b/crates/document-bundle/src/lib.rs @@ -0,0 +1,457 @@ +#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] +//! A JSON document bundle envelope over ZIP. +//! +//! Applications keep their document models. This crate owns the reusable container mechanics: +//! ordered mimetype, metadata, JSON model, previews, embedded asset paths, compression policy, +//! bounded reads and extraction of legacy base64 asset maps. Entry names and MIME type are +//! supplied by the format using the bundle. + +use std::io::{Cursor, Read, Seek, Write}; + +use craft_zip::{Archive, Limits}; +use serde_json::Value; + +const MAX_ENTRIES: u64 = 100_000; +const JSON_LIMIT: u64 = 256 << 20; +const ASSET_LIMIT: u64 = 1 << 30; +const EXPANSION_LIMIT: u64 = 4 << 30; +/// Largest proxy preview read from a bundle; larger entries are skipped. +pub const PROXY_LIMIT: u64 = 1 << 20; + +/// Per-entry and total-expansion limits for an untrusted document bundle. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct ReadLimits { + /// Maximum expanded size of metadata and document JSON. + pub json_bytes: u64, + /// Maximum expanded size of one compressed asset. Stored assets are bounded by archive size. + pub compressed_asset_bytes: u64, + /// Maximum expansion beyond the bytes in the source archive. + pub expansion_bytes: u64, +} + +impl Default for ReadLimits { + fn default() -> Self { + Self { json_bytes: JSON_LIMIT, compressed_asset_bytes: ASSET_LIMIT, expansion_bytes: EXPANSION_LIMIT } + } +} + +/// Names and identity used by one JSON document bundle format. +#[derive(Clone, Copy, Debug)] +pub struct Spec<'a> { + /// Stored first entry and its payload. + pub mimetype_entry: &'a str, + pub mimetype: &'a str, + /// Metadata and document JSON member names. + pub metadata_entry: &'a str, + pub document_entry: &'a str, + /// Optional preview entry name. + pub preview_entry: &'a str, + /// Directory prefix for embedded assets, without a trailing slash. + pub assets_directory: &'a str, + /// Directory prefix for proxy previews of linked assets, without a trailing slash. + pub proxies_directory: &'a str, + /// Metadata value written in the `format` field. + pub format_name: &'a str, + /// Old single-file JSON field containing base64 assets keyed by decimal id. + pub legacy_assets_field: Option<&'a str>, +} + +/// An embedded asset passed to [`write`]. +#[derive(Clone, Copy, Debug)] +pub struct Asset<'a> { + /// Decimal asset id used in the archive entry path. + pub id: &'a str, + /// MIME type used to select the extension and compression policy. + pub mime: &'a str, + /// File bytes retained byte-for-byte. + pub data: &'a [u8], +} + +/// A document and embedded payloads returned by [`read`]. +#[derive(Clone, Debug, PartialEq)] +pub struct Package { + /// Parsed model JSON, after extracting legacy inline assets. + pub document: Value, + /// Format version declared by metadata, or 1 when absent. + pub version: u64, + /// Embedded asset bytes keyed by decimal asset id. + pub assets: Vec<(String, Vec)>, + /// Proxy preview bytes (JPEG or PNG) keyed by decimal asset id, for ids the document lists. + pub proxies: Vec<(String, Vec)>, +} + +/// Why a document bundle could not be read or written. +#[derive(Clone, Debug, PartialEq, Eq, thiserror::Error)] +pub enum Error { + /// The bytes are not a bundle or legacy JSON document. + #[error("not a document bundle: {0}")] + NotBundle(String), + /// The bundle requires a newer reader. + #[error("document format {0} is newer than this reader supports")] + TooNew(u64), + /// A member could not be written. + #[error("could not write document bundle: {0}")] + Write(String), + /// A member exceeded a read limit or could not be read. + #[error("could not read document bundle: {0}")] + Read(String), +} + +fn extension(mime: &str) -> &'static str { + match mime { + "image/png" => "png", + "image/jpeg" => "jpg", + "image/gif" => "gif", + "image/webp" => "webp", + "image/tiff" => "tif", + "application/pdf" => "pdf", + "application/postscript" => "eps", + "image/svg+xml" => "svg", + _ => "bin", + } +} + +/// Whether `bytes` are a ZIP bundle that contains the configured document member. +pub fn has_document(bytes: &[u8], spec: Spec<'_>) -> bool { + let Ok(archive) = Archive::open(bytes, Limits { max_entries: MAX_ENTRIES, max_total_bytes: 0 }) else { return false }; + archive.contains(spec.document_entry) +} + +/// Whether `prefix` starts with the configured stored mimetype entry. +pub fn starts_with_mimetype(prefix: &[u8], spec: Spec<'_>) -> bool { + craft_zip::stored_mimetype(prefix).is_some_and(|mime| mime == spec.mimetype.as_bytes()) +} + +/// Refuses every write and seek after the first failure of the wrapped writer. Dropping a +/// `ZipWriter` after an error finalizes the archive, which must not touch a cancelled target. +struct Latch { + inner: W, + failed: bool, +} + +impl Latch { + fn latched() -> std::io::Error { + std::io::Error::other("writer already failed") + } +} + +impl Write for Latch { + fn write(&mut self, buf: &[u8]) -> std::io::Result { + if self.failed { + return Err(Self::latched()); + } + self.inner.write(buf).inspect_err(|_| self.failed = true) + } + fn flush(&mut self) -> std::io::Result<()> { + if self.failed { + return Err(Self::latched()); + } + self.inner.flush().inspect_err(|_| self.failed = true) + } +} + +impl Seek for Latch { + fn seek(&mut self, pos: std::io::SeekFrom) -> std::io::Result { + if self.failed { + return Err(Self::latched()); + } + self.inner.seek(pos).inspect_err(|_| self.failed = true) + } +} + +fn write_err(e: impl std::fmt::Display) -> Error { + Error::Write(e.to_string()) +} + +/// Write a ZIP document bundle containing the serialized document, metadata, preview, assets and +/// proxy previews. +pub fn write( + document: &[u8], + version: u32, + generator: &str, + preview: Option<&[u8]>, + assets: &[Asset<'_>], + proxies: &[Asset<'_>], + spec: Spec<'_>, +) -> Result, Error> { + let mut out = Cursor::new(Vec::new()); + write_to(&mut out, document, version, generator, preview, assets, proxies, spec)?; + Ok(out.into_inner()) +} + +/// Like [`write`], streaming the archive into `out`. A writer that fails (for example to cancel +/// a long save) stops the write at the next member or buffer and the error is returned. +#[allow(clippy::too_many_arguments)] +pub fn write_to( + out: W, + document: &[u8], + version: u32, + generator: &str, + preview: Option<&[u8]>, + assets: &[Asset<'_>], + proxies: &[Asset<'_>], + spec: Spec<'_>, +) -> Result<(), Error> { + use zip::CompressionMethod::{Deflated, Stored}; + let mut zip = zip::ZipWriter::new(Latch { inner: out, failed: false }); + let mut add = |name: &str, data: &[u8], method: zip::CompressionMethod| -> Result<(), Error> { + let options = zip::write::SimpleFileOptions::default() + .compression_method(method) + .compression_level(if method == Deflated { Some(6) } else { None }) + .last_modified_time(zip::DateTime::default()) + .large_file(u64::try_from(data.len()).unwrap_or(u64::MAX) >= u64::from(u32::MAX) / 2); + zip.start_file(name, options).map_err(|e| Error::Write(format!("{name}: {e}")))?; + zip.write_all(data).map_err(|e| Error::Write(format!("{name}: {e}"))) + }; + add(spec.mimetype_entry, spec.mimetype.as_bytes(), Stored)?; + let metadata = serde_json::json!({"format":spec.format_name, "version":version, "generator":generator}); + add(spec.metadata_entry, &serde_json::to_vec(&metadata).map_err(write_err)?, Deflated)?; + if let Some(preview) = preview { + add(spec.preview_entry, preview, Stored)?; + } + add(spec.document_entry, document, Deflated)?; + for asset in assets.iter().filter(|asset| !asset.data.is_empty()) { + let path = format!("{}/{}.{}", spec.assets_directory, asset.id, extension(asset.mime)); + let method = if matches!(asset.mime, "image/png" | "image/jpeg" | "image/gif" | "image/webp") { Stored } else { Deflated }; + add(&path, asset.data, method)?; + } + for proxy in proxies.iter().filter(|proxy| !proxy.data.is_empty()) { + let ext = if proxy.mime == "image/jpeg" { "jpg" } else { "png" }; + add(&format!("{}/{}.{ext}", spec.proxies_directory, proxy.id), proxy.data, Stored)?; + } + zip.finish().map(drop).map_err(write_err) +} + +/// Read a ZIP bundle or legacy JSON document within bounded member sizes. +/// +/// Members under the configured assets directory whose id does not occur in the document's +/// `assets` object are never decompressed. Unreferenced data cannot consume the caller's memory. +pub fn read(bytes: &[u8], supported_version: u32, spec: Spec<'_>) -> Result { + read_with_limits(bytes, supported_version, spec, ReadLimits::default()) +} + +/// Like [`read`], with caller-selected bounded-read limits. +pub fn read_with_limits(bytes: &[u8], supported_version: u32, spec: Spec<'_>, limits: ReadLimits) -> Result { + if !bytes.starts_with(b"PK") { + return read_legacy(bytes, spec); + } + let source_size = u64::try_from(bytes.len()).unwrap_or(u64::MAX); + let mut archive = Archive::open(bytes, Limits { max_entries: MAX_ENTRIES, max_total_bytes: source_size.saturating_add(limits.expansion_bytes) }) + .map_err(|e| Error::NotBundle(e.to_string()))?; + if archive.has_duplicate_names() { + return Err(Error::NotBundle("the archive has duplicate member names".into())); + } + let version = archive + .read(spec.metadata_entry, limits.json_bytes) + .map_err(|e| Error::Read(e.to_string()))? + .and_then(|metadata| serde_json::from_slice::(&metadata).ok()) + .and_then(|metadata| metadata.get("version").and_then(Value::as_u64)) + .unwrap_or(1); + if version > u64::from(supported_version) { + return Err(Error::TooNew(version)); + } + let document_bytes = archive + .read(spec.document_entry, limits.json_bytes) + .map_err(|e| Error::Read(e.to_string()))? + .ok_or_else(|| Error::NotBundle(format!("missing {}", spec.document_entry)))?; + let document: Value = serde_json::from_slice(&document_bytes).map_err(|e| Error::NotBundle(e.to_string()))?; + let wanted: std::collections::HashSet = + document.get("assets").and_then(Value::as_object).into_iter().flat_map(|assets| assets.keys().cloned()).collect(); + let prefix = format!("{}/", spec.assets_directory); + let mut embedded = Vec::new(); + for path in archive.names().map(|name| name.into_owned()).collect::>() { + let Some(member) = path.strip_prefix(&prefix) else { continue }; + let Some(id) = member.split('.').next() else { continue }; + if !wanted.contains(id) || id.parse::().is_err() { + continue; + } + let stored = archive.is_stored(&path).unwrap_or(false); + let limit = if stored { source_size } else { limits.compressed_asset_bytes }; + let data = archive.read(&path, limit).map_err(|e| Error::Read(e.to_string()))?; + if let Some(data) = data { + embedded.push((id.to_owned(), data)); + } + } + let proxies = read_proxies(&mut archive, &wanted, spec); + Ok(Package { document, version, assets: embedded, proxies }) +} + +/// Proxy previews for known asset ids. Only `/.jpg|png` is accepted; entries over +/// [`PROXY_LIMIT`] or that fail to read are skipped. +fn read_proxies(archive: &mut Archive<'_>, wanted: &std::collections::HashSet, spec: Spec<'_>) -> Vec<(String, Vec)> { + let prefix = format!("{}/", spec.proxies_directory); + let mut proxies = Vec::new(); + for path in archive.names().map(|name| name.into_owned()).collect::>() { + let Some(member) = path.strip_prefix(&prefix) else { continue }; + let Some((id, ext)) = member.split_once('.') else { continue }; + if !matches!(ext, "jpg" | "png") || !wanted.contains(id) || id.parse::().is_err() { + continue; + } + if archive.declared_size(&path).is_none_or(|size| size > PROXY_LIMIT) { + continue; + } + let Some(size) = archive.declared_size(&path) else { continue }; + if archive.budget().remaining() < size { + break; + } + match archive.read(&path, PROXY_LIMIT) { + Ok(Some(data)) => proxies.push((id.to_owned(), data)), + Ok(None) => {} + // A failed read may have decompressed up to `size` bytes without being charged. + Err(_) => { + if archive.budget_mut().charge(size).is_err() { + break; + } + } + } + } + proxies +} + +/// Read a preview with a fixed maximum size without decompressing the full bundle. +pub fn read_preview(bytes: &[u8], preview_entry: &str) -> Option> { + const PREVIEW_LIMIT: u64 = 4 << 20; + let mut archive = Archive::open(bytes, Limits { max_entries: MAX_ENTRIES, max_total_bytes: PREVIEW_LIMIT }).ok()?; + archive.read(preview_entry, PREVIEW_LIMIT).ok().flatten() +} + +/// Read a bounded preview from a seekable file without loading the full bundle into memory. +pub fn read_preview_from(reader: R, preview_entry: &str) -> Option> { + const PREVIEW_LIMIT: u64 = 4 << 20; + let mut zip = zip::ZipArchive::new(reader).ok()?; + let entry = zip.by_name(preview_entry).ok()?; + if entry.size() > PREVIEW_LIMIT { + return None; + } + let mut preview = Vec::new(); + entry.take(PREVIEW_LIMIT.saturating_add(1)).read_to_end(&mut preview).ok()?; + (u64::try_from(preview.len()).ok()? <= PREVIEW_LIMIT).then_some(preview) +} + +fn read_legacy(bytes: &[u8], spec: Spec<'_>) -> Result { + let mut document: Value = serde_json::from_slice(bytes).map_err(|e| Error::NotBundle(e.to_string()))?; + let mut assets = Vec::new(); + let mut total = 0u64; + if let Some(field) = spec.legacy_assets_field { + if let Some(inline) = document.get(field).and_then(Value::as_object) { + for (id, value) in inline { + if id.parse::().is_ok() + && let Some(encoded) = value.as_str() + { + let decoded_size = u64::try_from(encoded.len()).unwrap_or(u64::MAX).saturating_mul(3) / 4; + if decoded_size > ASSET_LIMIT { + return Err(Error::Read(format!("legacy asset {id} exceeds the {ASSET_LIMIT}-byte limit"))); + } + total = total.saturating_add(decoded_size); + if total > EXPANSION_LIMIT { + return Err(Error::Read(format!("legacy assets exceed the {EXPANSION_LIMIT}-byte total limit"))); + } + assets.push((id.clone(), decode_base64(encoded))); + } + } + } + if let Some(object) = document.as_object_mut() { + object.remove(field); + } + } + Ok(Package { document, version: 1, assets, proxies: Vec::new() }) +} + +fn decode_base64(input: &str) -> Vec { + let mut out = Vec::with_capacity(input.len().saturating_mul(3) / 4); + let (mut buffer, mut bits) = (0u32, 0u32); + for byte in input.bytes() { + let value = match byte { + b'A'..=b'Z' => byte - b'A', + b'a'..=b'z' => byte - b'a' + 26, + b'0'..=b'9' => byte - b'0' + 52, + b'+' | b'-' => 62, + b'/' | b'_' => 63, + _ => continue, + }; + buffer = (buffer << 6) | u32::from(value); + bits += 6; + if bits >= 8 { + bits -= 8; + out.push((buffer >> bits) as u8); + } + } + out +} + +#[cfg(test)] +mod tests { + use super::*; + + const SPEC: Spec<'static> = Spec { + mimetype_entry: "mimetype", + mimetype: "application/vnd.test+zip", + metadata_entry: "meta.json", + document_entry: "document.json", + preview_entry: "preview.png", + assets_directory: "assets", + proxies_directory: "proxies", + format_name: "test", + legacy_assets_field: Some("assetData"), + }; + + #[test] + fn writes_and_reads_bundle_members() { + let payload = [1u8, 2, 3]; + let bytes = write(br#"{"assets":{"7":{}}}"#, 4, "Test 0.1", Some(b"png"), &[Asset { id: "7", mime: "image/png", data: &payload }], &[], SPEC) + .unwrap(); + assert!(has_document(&bytes, SPEC)); + assert!(starts_with_mimetype(&bytes, SPEC)); + let loaded = read(&bytes, 4, SPEC).unwrap(); + assert_eq!(loaded.document, serde_json::json!({"assets":{"7":{}}})); + assert_eq!(loaded.version, 4); + assert_eq!(loaded.assets, [("7".to_owned(), payload.to_vec())]); + assert_eq!(read_preview(&bytes, SPEC.preview_entry).as_deref(), Some(&b"png"[..])); + } + + #[test] + fn extracts_legacy_inline_assets() { + let loaded = read(br#"{"assetData":{"7":"AQID"},"name":"x"}"#, 4, SPEC).unwrap(); + assert_eq!(loaded.document, serde_json::json!({"name":"x"})); + assert_eq!(loaded.assets, [("7".to_owned(), vec![1, 2, 3])]); + } + + #[test] + fn refuses_newer_versions() { + let bytes = write(b"{}", 9, "future", None, &[], &[], SPEC).unwrap(); + assert_eq!(read(&bytes, 4, SPEC), Err(Error::TooNew(9))); + } + + #[test] + fn stored_assets_use_source_size_and_deflated_assets_use_their_limit() { + let bytes = vec![7u8; 128]; + let document = br#"{"assets":{"7":{}}}"#; + let limits = ReadLimits { json_bytes: 1 << 20, compressed_asset_bytes: 8, expansion_bytes: 1 << 20 }; + let stored = write(document, 4, "test", None, &[Asset { id: "7", mime: "image/png", data: &bytes }], &[], SPEC).unwrap(); + let package = read_with_limits(&stored, 4, SPEC, limits).unwrap(); + assert_eq!(package.assets, [("7".to_owned(), bytes.clone())]); + + let compressed = write(document, 4, "test", None, &[Asset { id: "7", mime: "application/octet-stream", data: &bytes }], &[], SPEC).unwrap(); + assert!(read_with_limits(&compressed, 4, SPEC, limits).is_err()); + } + + #[test] + fn failed_proxy_reads_are_charged_and_stop_reading_when_the_budget_is_spent() { + let pad = " ".repeat(40_000); + let document = format!(r#"{{"assets":{{"1":{{}},"2":{{}},"3":{{}},"4":{{}}}},"pad":"{pad}"}}"#); + let blobs: Vec> = (1u8..=4).map(|n| vec![n.wrapping_mul(50); 1000]).collect(); + let ids = ["1", "2", "3", "4"]; + let proxies: Vec> = ids.iter().zip(&blobs).map(|(id, data)| Asset { id, mime: "image/png", data }).collect(); + let mut bytes = write(document.as_bytes(), 4, "t", None, &[], &proxies, SPEC).unwrap(); + // Corrupt proxies 1 to 3 so their reads fail their checksum. + for blob in &blobs[..3] { + let at = bytes.windows(blob.len()).position(|w| w == blob.as_slice()).unwrap(); + bytes[at + 500] ^= 0xFF; + } + let source = bytes.len() as u64; + let all = read(&bytes, 4, SPEC).unwrap(); + assert_eq!(all.proxies.len(), 1); + let tight = ReadLimits { json_bytes: 1 << 20, compressed_asset_bytes: 1 << 20, expansion_bytes: 40_100 + 2_500 - source }; + assert!(read_with_limits(&bytes, 4, SPEC, tight).unwrap().proxies.is_empty()); + } +} diff --git a/crates/docx/src/chart_spec.rs b/crates/docx/src/chart_spec.rs index e31363903..a090c5c96 100644 --- a/crates/docx/src/chart_spec.rs +++ b/crates/docx/src/chart_spec.rs @@ -6,7 +6,7 @@ //! [`spec_of`] turns a chart part back into the model only when writing that model again gives //! the same markup, so a chart with anything the model leaves out stays as it was read. -use craft_wordmodel::chart::{ChartSeries, ChartSpec, ChartType, LegendPos, MAX_POINTS, MAX_SERIES}; +use craft_wordmodel::chart::{ChartAxis, ChartSeries, ChartSpec, ChartType, LegendPos, MAX_POINTS, MAX_SERIES}; use craft_wordmodel::graphic::GraphicItem; use craft_wordmodel::props::Rgb; @@ -102,11 +102,24 @@ fn bounded(spec: &ChartSpec) -> ChartSpec { .series .iter() .take(MAX_SERIES) - .map(|s| ChartSeries { name: s.name.clone(), values: s.values.iter().take(MAX_POINTS).copied().collect() }) + .map(|s| ChartSeries { + name: s.name.clone(), + values: s.values.iter().take(MAX_POINTS).copied().collect(), + fill: s.fill, + no_fill: s.no_fill, + line_w: s.line_w, + no_line: s.no_line, + markers: s.markers, + smooth: s.smooth, + labels: s.labels, + fmt: s.fmt.clone(), + }) .collect(), title: spec.title.clone(), legend: spec.legend, data_labels: spec.data_labels, + cat_axis: spec.cat_axis.clone(), + val_axis: spec.val_axis.clone(), } .sanitized() } @@ -185,17 +198,19 @@ fn plot(w: &mut W, spec: &ChartSpec) { match k { ChartType::Pie | ChartType::Doughnut => {} ChartType::Scatter => { - value_axis(w, CAT_AX, VAL_AX, "b", false, "midCat"); - value_axis(w, VAL_AX, CAT_AX, "l", true, "midCat"); + value_axis(w, CAT_AX, VAL_AX, "b", "midCat", &spec.cat_axis); + value_axis(w, VAL_AX, CAT_AX, "l", "midCat", &spec.val_axis); } _ => { - category_axis(w, if horizontal { "l" } else { "b" }); - value_axis(w, VAL_AX, CAT_AX, if horizontal { "b" } else { "l" }, true, "between"); + category_axis(w, if horizontal { "l" } else { "b" }, &spec.cat_axis); + value_axis(w, VAL_AX, CAT_AX, if horizontal { "b" } else { "l" }, "between", &spec.val_axis); } } } -/// One `c:ser` (ECMA-376 element order: idx, order, tx, spPr, marker, cat/xVal, val/yVal, smooth). +/// One `c:ser` (ECMA-376 element order: idx, order, tx, spPr, marker, invertIfNegative, dLbls, +/// cat/xVal, val/yVal, smooth): the series' formatting (fill, line, markers, its own labels) sits +/// in `c:spPr`, `c:marker` and `c:dLbls`; the values carry their number format. fn series(w: &mut W, spec: &ChartSpec, i: usize, s: &ChartSeries) { let k = spec.kind; let i = i.to_string(); @@ -207,25 +222,26 @@ fn series(w: &mut W, spec: &ChartSpec, i: usize, s: &ChartSeries) { w.leaf("c:v", &[], &s.name); w.close("c:tx"); } - match k { - // Markers only: the series has no line. - ChartType::Scatter => { - w.open("c:spPr", &[]); - w.open("a:ln", &[("w", "19050")]); - w.empty("a:noFill", &[]); - w.close("a:ln"); - w.close("c:spPr"); - } - ChartType::Column | ChartType::ColumnStacked | ChartType::Bar | ChartType::BarStacked => { - w.empty("c:invertIfNegative", &[("val", "0")]); - } - ChartType::Line => { - w.open("c:marker", &[]); - w.empty("c:symbol", &[("val", "none")]); - w.close("c:marker"); - } - _ => {} + series_props(w, s); + series_marker(w, k, s); + if matches!(k, ChartType::Column | ChartType::ColumnStacked | ChartType::Bar | ChartType::BarStacked) { + w.empty("c:invertIfNegative", &[("val", "0")]); } + if let Some(on) = s.labels { + w.open("c:dLbls", &[]); + for (name, show) in [ + ("c:showLegendKey", false), + ("c:showVal", on), + ("c:showCatName", false), + ("c:showSerName", false), + ("c:showPercent", false), + ("c:showBubbleSize", false), + ] { + w.empty(name, &[("val", if show { "1" } else { "0" })]); + } + w.close("c:dLbls"); + } + let code = s.fmt.as_deref().unwrap_or("General"); if k == ChartType::Scatter { let xs: Option>> = spec .categories @@ -234,27 +250,85 @@ fn series(w: &mut W, spec: &ChartSpec, i: usize, s: &ChartSeries) { .collect(); w.open("c:xVal", &[]); match xs { - Some(xs) => num_lit(w, &xs), + Some(xs) => num_lit(w, &xs, "General"), None => str_lit(w, &spec.categories), } w.close("c:xVal"); w.open("c:yVal", &[]); - num_lit(w, &s.values); + num_lit(w, &s.values, code); w.close("c:yVal"); } else { w.open("c:cat", &[]); str_lit(w, &spec.categories); w.close("c:cat"); w.open("c:val", &[]); - num_lit(w, &s.values); + num_lit(w, &s.values, code); w.close("c:val"); } if matches!(k, ChartType::Line | ChartType::Scatter) { - w.empty("c:smooth", &[("val", "0")]); + w.empty("c:smooth", &[("val", if s.smooth { "1" } else { "0" })]); } w.close("c:ser"); } +/// `c:spPr`: the series' fill (`a:solidFill` / `a:noFill`) and line (`a:ln`), written when the +/// model sets them. `a:ln` without `w` is the default width, `w` a fixed one, and `a:noFill` +/// inside it "no line" (a scatter series keeps its markers with no line). +fn series_props(w: &mut W, s: &ChartSeries) { + let fill_set = s.no_fill || s.fill.is_some(); + let ln_set = s.no_line || s.line_w.is_some(); + if !(fill_set || ln_set) { + return; + } + w.open("c:spPr", &[]); + if fill_set { + // "No fill" wins over a stored colour, so the pair reads back as it was written. + match &s.fill { + Some(c) if !s.no_fill => { + w.open("a:solidFill", &[]); + w.empty("a:srgbClr", &[("val", &c.hex())]); + w.close("a:solidFill"); + } + _ => w.empty("a:noFill", &[]), + } + } + if ln_set { + let w_attr = s.line_w.map(emu).unwrap_or_default(); + let attrs = [("w", w_attr.as_str())]; + w.open("a:ln", if w_attr.is_empty() { &[] } else { &attrs }); + if s.no_line { + w.empty("a:noFill", &[]); + } + w.close("a:ln"); + } + w.close("c:spPr"); +} + +/// `c:marker`: line and scatter series say which symbol each point gets; other types have none. +fn series_marker(w: &mut W, k: ChartType, s: &ChartSeries) { + match k { + ChartType::Line => { + w.open("c:marker", &[]); + w.empty("c:symbol", &[("val", if s.markers == Some(true) { "circle" } else { "none" })]); + w.close("c:marker"); + } + // Unset: no `c:marker`, so Word picks its automatic symbol for each series. + ChartType::Scatter => { + if let Some(on) = s.markers { + w.open("c:marker", &[]); + w.empty("c:symbol", &[("val", if on { "circle" } else { "none" })]); + w.close("c:marker"); + } + } + _ => {} + } +} + +/// A line width in points as the `w` attribute DrawingML wants (EMU). +fn emu(pt: f32) -> String { + ((pt * 12700.0).round().max(0.0) as i64).to_string() +} + fn str_lit(w: &mut W, items: &[String]) { w.open("c:strLit", &[]); w.empty("c:ptCount", &[("val", &items.len().to_string())]); @@ -266,9 +340,9 @@ fn str_lit(w: &mut W, items: &[String]) { w.close("c:strLit"); } -fn num_lit(w: &mut W, values: &[Option]) { +fn num_lit(w: &mut W, values: &[Option], code: &str) { w.open("c:numLit", &[]); - w.leaf("c:formatCode", &[], "General"); + w.leaf("c:formatCode", &[], code); w.empty("c:ptCount", &[("val", &values.len().to_string())]); for (i, v) in values.iter().enumerate() { if let Some(v) = v.filter(|v| v.is_finite()) { @@ -285,17 +359,21 @@ fn num(v: f64) -> String { if v == 0.0 { "0".to_string() } else { format!("{v}") } } -fn category_axis(w: &mut W, pos: &str) { +fn category_axis(w: &mut W, pos: &str, ax: &ChartAxis) { w.open("c:catAx", &[]); w.empty("c:axId", &[("val", CAT_AX)]); - w.open("c:scaling", &[]); - w.empty("c:orientation", &[("val", "minMax")]); - w.close("c:scaling"); + scaling(w, ax); w.empty("c:delete", &[("val", "0")]); w.empty("c:axPos", &[("val", pos)]); + if ax.gridlines { + w.empty("c:majorGridlines", &[]); + } + axis_title(w, ax); + num_fmt(w, ax); w.empty("c:majorTickMark", &[("val", "none")]); w.empty("c:minorTickMark", &[("val", "none")]); - w.empty("c:tickLblPos", &[("val", "nextTo")]); + w.empty("c:tickLblPos", &[("val", if ax.labels { "nextTo" } else { "none" })]); + axis_line(w, ax); w.empty("c:crossAx", &[("val", VAL_AX)]); w.empty("c:crosses", &[("val", "autoZero")]); w.empty("c:auto", &[("val", "1")]); @@ -305,34 +383,81 @@ fn category_axis(w: &mut W, pos: &str) { w.close("c:catAx"); } -/// A value axis: with gridlines and no axis line (`grid`), else with its line and no gridlines. -fn value_axis(w: &mut W, id: &str, cross: &str, pos: &str, grid: bool, between: &str) { +/// A value axis (scatter x axes included): its scale, gridlines, title, labels and line, as the +/// model sets them. +fn value_axis(w: &mut W, id: &str, cross: &str, pos: &str, between: &str, ax: &ChartAxis) { w.open("c:valAx", &[]); w.empty("c:axId", &[("val", id)]); - w.open("c:scaling", &[]); - w.empty("c:orientation", &[("val", "minMax")]); - w.close("c:scaling"); + scaling(w, ax); w.empty("c:delete", &[("val", "0")]); w.empty("c:axPos", &[("val", pos)]); - if grid { + if ax.gridlines { w.empty("c:majorGridlines", &[]); } + axis_title(w, ax); + num_fmt(w, ax); w.empty("c:majorTickMark", &[("val", "none")]); w.empty("c:minorTickMark", &[("val", "none")]); - w.empty("c:tickLblPos", &[("val", "nextTo")]); - if grid { - w.open("c:spPr", &[]); - w.open("a:ln", &[]); - w.empty("a:noFill", &[]); - w.close("a:ln"); - w.close("c:spPr"); - } + w.empty("c:tickLblPos", &[("val", if ax.labels { "nextTo" } else { "none" })]); + axis_line(w, ax); w.empty("c:crossAx", &[("val", cross)]); w.empty("c:crosses", &[("val", "autoZero")]); w.empty("c:crossBetween", &[("val", between)]); w.close("c:valAx"); } +/// `c:scaling`: the orientation, then the fixed bounds when the model sets them. +fn scaling(w: &mut W, ax: &ChartAxis) { + w.open("c:scaling", &[]); + w.empty("c:orientation", &[("val", "minMax")]); + for (name, v) in [("c:max", ax.max), ("c:min", ax.min)] { + if let Some(v) = v.filter(|v| v.is_finite()) { + w.empty(name, &[("val", &num(v))]); + } + } + w.close("c:scaling"); +} + +/// The axis title, in the same rich-text shape the chart title uses. +fn axis_title(w: &mut W, ax: &ChartAxis) { + let Some(t) = &ax.title else { return }; + w.open("c:title", &[]); + w.open("c:tx", &[]); + w.open("c:rich", &[]); + w.empty("a:bodyPr", &[]); + w.empty("a:lstStyle", &[]); + w.open("a:p", &[]); + w.open("a:r", &[]); + w.leaf("a:t", &[], t); + w.close("a:r"); + w.close("a:p"); + w.close("c:rich"); + w.close("c:tx"); + w.empty("c:overlay", &[("val", "0")]); + w.close("c:title"); +} + +/// Tick-label number format: the model's code, unlinked from the source, else the default. +fn num_fmt(w: &mut W, ax: &ChartAxis) { + match &ax.fmt { + Some(code) => w.empty("c:numFmt", &[("formatCode", code), ("sourceLinked", "0")]), + None => w.empty("c:numFmt", &[("formatCode", "General"), ("sourceLinked", "1")]), + } +} + +/// `c:spPr`: an axis line only when the model keeps it (the default writes nothing, which reads +/// as a line); `a:noFill` when the axis line is off. +fn axis_line(w: &mut W, ax: &ChartAxis) { + if ax.line { + return; + } + w.open("c:spPr", &[]); + w.open("a:ln", &[]); + w.empty("a:noFill", &[]); + w.close("a:ln"); + w.close("c:spPr"); +} + /// The chart a chart part holds, when the part is exactly what [`chart_xml`] writes for it (so /// editing and saving it loses nothing). `None` for any other chart. pub(crate) fn spec_of(space: &El) -> Option { @@ -365,15 +490,40 @@ pub(crate) fn spec_of(space: &El) -> Option { let categories = sers.first().and_then(|s| s.child(if scatter { "c:xVal" } else { "c:cat" })).map(literal).unwrap_or_default(); let series = sers .iter() - .map(|s| ChartSeries { - name: s.child("c:tx").and_then(|t| t.child("c:v")).map(|v| v.text()).unwrap_or_default(), - values: s - .child(if scatter { "c:yVal" } else { "c:val" }) - .map(literal) - .unwrap_or_default() - .iter() - .map(|v| v.parse::().ok().filter(|v| v.is_finite())) - .collect(), + .map(|s| { + let ln = s.child("c:spPr").and_then(|p| p.child("a:ln")); + let symbol = s.child("c:marker").and_then(|m| m.child("c:symbol")).and_then(|m| m.attr("val")); + ChartSeries { + name: s.child("c:tx").and_then(|t| t.child("c:v")).map(|v| v.text()).unwrap_or_default(), + values: s + .child(if scatter { "c:yVal" } else { "c:val" }) + .map(literal) + .unwrap_or_default() + .iter() + .map(|v| v.parse::().ok().filter(|v| v.is_finite())) + .collect(), + fill: s + .child("c:spPr") + .and_then(|p| p.child("a:solidFill")) + .and_then(|f| f.child("a:srgbClr")) + .and_then(|c| c.attr("val")) + .and_then(Rgb::parse), + no_fill: s.child("c:spPr").and_then(|p| p.child("a:noFill")).is_some(), + line_w: ln.and_then(|l| l.attr("w")).and_then(|w| w.parse::().ok()).filter(|w| w.is_finite()).map(|w| (w / 12700.0) as f32), + no_line: ln.and_then(|l| l.child("a:noFill")).is_some(), + // Line series show no markers by default: only a setting that differs from that + // belongs in the model. A scatter series keeps what the file says, so a missing + // `c:marker` (Word's automatic symbols) stays unset and is not written back. + markers: symbol.map(|v| v != "none").filter(|m| scatter || *m), + smooth: s.child("c:smooth").and_then(|v| v.attr("val")).is_some_and(|v| v == "1"), + labels: s.child("c:dLbls").and_then(|d| d.child("c:showVal")).and_then(|v| v.attr("val")).map(|v| v == "1"), + fmt: s + .child(if scatter { "c:yVal" } else { "c:val" }) + .and_then(|v| v.els().find(|e| e.name == "c:numLit")) + .and_then(|l| l.child("c:formatCode")) + .map(|f| f.text()) + .filter(|f| *f != "General"), + } }) .collect(); let title = chart.child("c:title").and_then(|t| t.find("a:t")).map(|t| t.text()); @@ -384,12 +534,48 @@ pub(crate) fn spec_of(space: &El) -> Option { _ => LegendPos::Right, }); let data_labels = ty.child("c:dLbls").and_then(|d| d.child("c:showVal")).and_then(|v| v.attr("val")) == Some("1"); - let spec = ChartSpec { kind, categories, series, title, legend, data_labels }.sanitized(); - // Editable only when nothing would be lost: writing the model again gives the same part. - let again = xml::parse(chart_xml(&spec).as_bytes()).ok()?; + let (cat_a, val_a) = crate::read::chart::chart_axes(plot, scatter); + let axis = |a: crate::read::chart::Axis| ChartAxis { + title: a.title, + min: a.min, + max: a.max, + fmt: a.fmt, + gridlines: a.grid, + labels: a.labels, + line: a.line, + }; + let spec = ChartSpec { kind, categories, series, title, legend, data_labels, cat_axis: axis(cat_a), val_axis: axis(val_a) }.sanitized(); + // Editable only when nothing would be lost: writing the model again gives the same part, as + // this writer or the earlier one wrote it. + let mut again = xml::parse(chart_xml(&spec).as_bytes()).ok()?; + if same(space, &again, 0) { + return Some(spec); + } + as_previous_writer(&mut again); same(space, &again, 0).then_some(spec) } +/// `part` (what [`chart_xml`] writes) as the earlier WordCraft writer wrote the same model: it +/// had no axis `c:numFmt`. Only that element goes, and only with the value its absence reads as +/// (General, linked to the source), so a part that matches still holds nothing the model leaves out. +fn as_previous_writer(part: &mut El) { + fn els_mut(e: &mut El) -> impl Iterator { + e.kids.iter_mut().filter_map(|n| match n { + Node::El(e) => Some(e), + Node::Text(_) => None, + }) + } + fn drop_els(e: &mut El, drop: impl Fn(&El) -> bool) { + e.kids.retain(|n| !matches!(n, Node::El(c) if drop(c))); + } + let Some(plot) = els_mut(part).find(|e| e.name == "c:chart").and_then(|c| els_mut(c).find(|e| e.name == "c:plotArea")) else { return }; + for e in els_mut(plot) { + if matches!(e.name.as_str(), "c:catAx" | "c:valAx") { + drop_els(e, |c| c.name == "c:numFmt" && c.attr("sourceLinked") == Some("1")); + } + } +} + /// The points of a `c:strLit` / `c:numLit` (or a cache) as texts by index, `ptCount` long. fn literal(e: &El) -> Vec { let Some(lit) = e.els().find(|c| matches!(c.name.as_str(), "c:strLit" | "c:numLit")) else { return Vec::new() }; @@ -461,10 +647,11 @@ mod tests { let spec = ChartSpec { kind: ChartType::Column, categories: (0..4500).map(|i| i.to_string()).collect(), - series: (0..300).map(|i| ChartSeries { name: format!("S{i}"), values: vec![Some(1.0); 4500] }).collect(), + series: (0..300).map(|i| ChartSeries { name: format!("S{i}"), values: vec![Some(1.0); 4500], ..Default::default() }).collect(), title: None, legend: None, data_labels: false, + ..Default::default() }; let x = chart_xml(&spec); let series = x.matches("").count(); @@ -491,6 +678,33 @@ mod tests { assert_eq!(spec_of(&parse(&plain)), Some(plain)); } + #[test] + fn no_fill_wins_over_a_stored_colour() { + let mut spec = ChartSpec::sample(ChartType::Column); + spec.series[0].fill = Some(Rgb(255, 0, 0)); + spec.series[0].no_fill = true; + let x = chart_xml(&spec); + assert!(!x.contains("FF0000"), "colour written next to no_fill"); + let back = spec_of(&parse(&spec)).expect("stays editable"); + assert!(back.series[0].no_fill); + assert_eq!(back.series[0].fill, None); + } + + #[test] + fn an_axis_format_is_limited_in_characters_not_bytes() { + use craft_wordmodel::chart::MAX_FMT; + let cjk = format!("0\"{}\"", "\u{5143}".repeat(90)); + let ascii = format!("0\"{}\"", "x".repeat(90)); + for fmt in [cjk, ascii] { + assert!(fmt.chars().count() <= MAX_FMT); + let mut spec = ChartSpec::sample(ChartType::Column); + spec.val_axis.fmt = Some(fmt.clone()); + let spec = spec.sanitized(); + assert_eq!(spec.val_axis.fmt.as_deref(), Some(fmt.as_str())); + assert_eq!(spec_of(&parse(&spec)).as_ref(), Some(&spec), "{} chars", fmt.chars().count()); + } + } + #[test] fn a_chart_with_more_than_the_model_stays_read_only() { let spec = ChartSpec::sample(ChartType::Column); @@ -500,6 +714,61 @@ mod tests { assert_eq!(spec_of(&xml::parse(x.as_bytes()).unwrap()), None); } + /// Chart parts as the previous WordCraft chart writer saved them (its output for + /// `ChartSpec::sample`): no axis `c:numFmt`, and scatter series with no `c:marker`. + const PREVIOUS_COLUMN: &str = r#"Series 1Q1Q2Q3Q4General3.24.13.65Series 2Q1Q2Q3Q4General22.73.93.3Series 3Q1Q2Q3Q4General1.52.22.43.8"#; + const PREVIOUS_SCATTER: &str = r#"Series 1General1234General2.13.434.6Series 2General1234General1.21.92.83.1"#; + + #[test] + fn charts_saved_by_the_previous_writer_stay_editable() { + for (x, kind) in [(PREVIOUS_COLUMN, ChartType::Column), (PREVIOUS_SCATTER, ChartType::Scatter)] { + let spec = spec_of(&xml::parse(x.as_bytes()).unwrap()).unwrap_or_else(|| panic!("{kind:?}: the earlier part is read-only")); + assert_eq!((spec.kind, spec.series.len()), (kind, if kind == ChartType::Scatter { 2 } else { 3 })); + assert_eq!((spec.cat_axis.fmt.as_deref(), spec.val_axis.fmt.as_deref()), (None, None), "{kind:?}"); + assert!(spec.series.iter().all(|s| s.markers.is_none()), "{kind:?}: markers unset"); + // Saved again, it is in the current shape and still reads back as the same model. + assert_eq!(spec_of(&parse(&spec)).as_ref(), Some(&spec), "{kind:?}"); + } + // What neither writer writes still keeps such a part read-only: an axis format linked to + // the source that isn't General, a scatter marker other than the default, a marker size. + let edited = |x: &str, from: &str, to: &str| { + assert!(x.contains(from), "{from}"); + spec_of(&xml::parse(x.replacen(from, to, 1).as_bytes()).unwrap()) + }; + let ax = r#""#; + assert_eq!(edited(PREVIOUS_COLUMN, ax, &format!(r#"{ax}"#)), None); + let ln = r#""#; + assert_eq!(edited(PREVIOUS_SCATTER, ln, &format!(r#"{ln}"#)), None); + assert_eq!(edited(PREVIOUS_SCATTER, ln, &format!(r#"{ln}"#)), None); + assert_eq!(edited(PREVIOUS_COLUMN, "", ""), None); + } + + /// A scatter series with no `c:marker` (Word's automatic symbols) stays without one after an + /// unrelated edit and a save; an explicit circle or "none" stays explicit. + #[test] + fn scatter_series_without_a_marker_stay_without_after_a_save() { + let marker_of = |x: &El| -> Vec> { + let sc = x.child("c:chart").unwrap().child("c:plotArea").unwrap().child("c:scatterChart").unwrap(); + sc.children("c:ser").map(|s| s.child("c:marker").map(|m| m.child("c:symbol").unwrap().attr("val").unwrap().to_string())).collect() + }; + let mut spec = spec_of(&xml::parse(PREVIOUS_SCATTER.as_bytes()).unwrap()).unwrap(); + assert_eq!(marker_of(&xml::parse(PREVIOUS_SCATTER.as_bytes()).unwrap()), [None, None]); + spec.title = Some("Edited".into()); + spec.series[0].name = "Renamed".into(); + let saved = parse(&spec); + assert_eq!(marker_of(&saved), [None, None]); + let again = spec_of(&saved).unwrap(); + assert_eq!(again, spec); + assert!(again.series.iter().all(|s| s.markers.is_none())); + // Explicit choices are kept as written. + spec.series[0].markers = Some(true); + spec.series[1].markers = Some(false); + let saved = parse(&spec); + assert_eq!(marker_of(&saved), [Some("circle".to_string()), Some("none".to_string())]); + let again = spec_of(&saved).unwrap(); + assert_eq!((again.series[0].markers, again.series[1].markers), (Some(true), Some(false))); + } + #[test] fn elements_come_in_schema_order() { let mut spec = ChartSpec::sample(ChartType::ColumnStacked); @@ -517,6 +786,11 @@ mod tests { ["c:barDir", "c:grouping", "c:varyColors", "c:ser", "c:ser", "c:ser", "c:dLbls", "c:gapWidth", "c:overlap", "c:axId", "c:axId"] ); assert_eq!(names(bar.child("c:ser").unwrap()), ["c:idx", "c:order", "c:tx", "c:invertIfNegative", "c:cat", "c:val"]); + // A series with its own labels keeps `c:invertIfNegative` ahead of `c:dLbls` (CT_BarSer). + spec.series[0].labels = Some(true); + let labelled = parse(&spec); + let ser = labelled.child("c:chart").unwrap().child("c:plotArea").unwrap().child("c:barChart").unwrap().child("c:ser").unwrap(); + assert_eq!(names(ser), ["c:idx", "c:order", "c:tx", "c:invertIfNegative", "c:dLbls", "c:cat", "c:val"]); let val_ax = plot.child("c:valAx").unwrap(); assert_eq!( names(val_ax), @@ -526,6 +800,7 @@ mod tests { "c:delete", "c:axPos", "c:majorGridlines", + "c:numFmt", "c:majorTickMark", "c:minorTickMark", "c:tickLblPos", @@ -535,13 +810,35 @@ mod tests { "c:crossBetween" ] ); + let cat_ax = plot.child("c:catAx").unwrap(); + assert_eq!( + names(cat_ax), + [ + "c:axId", + "c:scaling", + "c:delete", + "c:axPos", + "c:numFmt", + "c:majorTickMark", + "c:minorTickMark", + "c:tickLblPos", + "c:crossAx", + "c:crosses", + "c:auto", + "c:lblAlgn", + "c:lblOffset", + "c:noMultiLvlLbl" + ] + ); let line = parse(&ChartSpec { data_labels: true, ..ChartSpec::sample(ChartType::Line) }); let lc = line.child("c:chart").unwrap().child("c:plotArea").unwrap().child("c:lineChart").unwrap(); assert_eq!(names(lc), ["c:grouping", "c:varyColors", "c:ser", "c:ser", "c:ser", "c:dLbls", "c:marker", "c:axId", "c:axId"]); assert_eq!(names(lc.child("c:ser").unwrap()), ["c:idx", "c:order", "c:tx", "c:marker", "c:cat", "c:val", "c:smooth"]); - let sc = parse(&ChartSpec::sample(ChartType::Scatter)); + let mut scatter = ChartSpec::sample(ChartType::Scatter); + scatter.series.iter_mut().for_each(|s| s.markers = Some(true)); + let sc = parse(&scatter); let sc = sc.child("c:chart").unwrap().child("c:plotArea").unwrap().child("c:scatterChart").unwrap(); - assert_eq!(names(sc.child("c:ser").unwrap()), ["c:idx", "c:order", "c:tx", "c:spPr", "c:xVal", "c:yVal", "c:smooth"]); + assert_eq!(names(sc.child("c:ser").unwrap()), ["c:idx", "c:order", "c:tx", "c:spPr", "c:marker", "c:xVal", "c:yVal", "c:smooth"]); let dn = parse(&ChartSpec::sample(ChartType::Doughnut)); let dn = dn.child("c:chart").unwrap().child("c:plotArea").unwrap(); assert_eq!(names(dn), ["c:layout", "c:doughnutChart"]); diff --git a/crates/docx/src/read.rs b/crates/docx/src/read.rs index 3e6ea5e61..0f95b4f76 100644 --- a/crates/docx/src/read.rs +++ b/crates/docx/src/read.rs @@ -119,6 +119,10 @@ pub(crate) struct Reader<'p> { pub theme_fills: shape_style::ThemeFills, /// The fill of the group being read (what its members' `a:grpFill` stands for). pub group_fill: Option, + /// The package has no default table style (no styles part, or none marked `w:default`): + /// Word then gives tables no cell margins rather than its Normal Table's 0.08". Set by + /// `read_styles`, which runs before any story with tables is read. + pub no_default_table_style: bool, } /// The package path that internal relationship `id` points at, if it has type `kind` (an `rt` @@ -193,11 +197,31 @@ fn read_opened_in(pkg: &Package, chunks: ChunkState) -> Result<(Document, usize, theme_fills: Default::default(), group_fill: None, chunks, + no_default_table_style: true, }; r.doc.missing_pictures.main = main.clone(); r.pc.major_font = r.doc.settings.major_font.clone(); r.pc.minor_font = r.doc.settings.minor_font.clone(); - let part = |t: &str| rels.by_type(t).filter(|x| !x.external).map(|x| x.target.clone()); + let part = |t: &str| { + let related = rels.by_type(t).filter(|x| !x.external).map(|x| x.target.clone()).filter(|p| pkg.get(p).is_some()); + related.or_else(|| { + // Older DesignCraft imports read these two conventional parts directly. Keep opening + // those files when a producer omitted the document relationship; the bounded package + // reader still applies the same entry and XML limits. An explicit usable relationship + // always wins, including when it points outside the conventional `word/` paths. + if !main.eq_ignore_ascii_case("word/document.xml") { + return None; + } + let fallback = match t { + rt::STYLES => "word/styles.xml", + rt::NUMBERING => "word/numbering.xml", + rt::FOOTNOTES => "word/footnotes.xml", + rt::ENDNOTES => "word/endnotes.xml", + _ => return None, + }; + pkg.get(fallback).map(|_| fallback.to_string()) + }) + }; // The font table and the fonts embedded through it ride along as opaque parts. if let Some(p) = part(rt::FONT_TABLE) { diff --git a/crates/docx/src/read/chart.rs b/crates/docx/src/read/chart.rs index a0087a3bb..5b02603d6 100644 --- a/crates/docx/src/read/chart.rs +++ b/crates/docx/src/read/chart.rs @@ -29,7 +29,9 @@ pub(super) use canvas::text_w; use canvas::{Rect, clip_text, legend, text}; #[cfg(test)] use format::{civil, excel_stamp, format_value}; -use model::{Axis, Grouping, Kind, Pos, Series}; +pub(crate) use model::Axis; +use model::{Grouping, Kind, Pos, Series}; +pub(crate) use parse::chart_axes; #[cfg(test)] use parse::{axis, category_text, labels, numbers}; use parse::{read_series, text_size, title_of}; diff --git a/crates/docx/src/read/chart/canvas.rs b/crates/docx/src/read/chart/canvas.rs index a8280e4ba..ab3fbf410 100644 --- a/crates/docx/src/read/chart/canvas.rs +++ b/crates/docx/src/read/chart/canvas.rs @@ -41,8 +41,13 @@ pub(super) fn widest(items: &[String], size: f32) -> f32 { items.iter().map(|s| text_w(s, size)).fold(0.0, f32::max) } +/// Text in `rect`; `angle` degrees clockwise turns it about the rect's centre (axis titles). pub(super) fn text(out: &mut Vec, rect: Rect, s: &str, size: f32, align: TextAlign, bold: bool) { - out.push(GraphicItem::Text { rect: rect.arr(), text: s.to_string(), size, color: TEXT, bold, align, font: None }); + text_angle(out, rect, s, size, align, bold, 0.0); +} + +pub(super) fn text_angle(out: &mut Vec, rect: Rect, s: &str, size: f32, align: TextAlign, bold: bool, angle: f32) { + out.push(GraphicItem::Text { rect: rect.arr(), text: s.to_string(), size, color: TEXT, bold, align, font: None, angle }); } /// Text centred on `cx` with its top at `y`. diff --git a/crates/docx/src/read/chart/model.rs b/crates/docx/src/read/chart/model.rs index 71ee7040f..13da1d654 100644 --- a/crates/docx/src/read/chart/model.rs +++ b/crates/docx/src/read/chart/model.rs @@ -7,6 +7,33 @@ pub(super) struct DataLabel { pub(super) y: f32, pub(super) text: String, pub(super) inside: bool, + /// Where a label that is not `inside` sits relative to (`x`, `y`). + pub(super) pos: LabelPos, + /// Distance from (`x`, `y`) to the label for `pos` other than `Center`, points (half the marker). + pub(super) off: f32, +} + +/// Where a data label sits beside its point (`c:dLblPos`: `t`, `b`, `l`, `r`, `ctr`). +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub(super) enum LabelPos { + Above, + Below, + Left, + Right, + Center, +} + +impl LabelPos { + pub(super) fn parse(v: &str) -> Option { + Some(match v.trim() { + "t" => LabelPos::Above, + "b" => LabelPos::Below, + "l" => LabelPos::Left, + "r" => LabelPos::Right, + "ctr" => LabelPos::Center, + _ => return None, + }) + } } #[derive(Clone, Copy, Debug, PartialEq, Eq)] @@ -52,16 +79,56 @@ pub(super) enum Pos { /// One value axis (or category axis) as the file states it. #[derive(Default)] -pub(super) struct Axis { - pub(super) deleted: bool, - pub(super) grid: bool, - pub(super) line: bool, +pub(crate) struct Axis { + pub(crate) deleted: bool, + pub(crate) grid: bool, + pub(crate) line: bool, /// `c:crossBetween val="midCat"`: categories sit on the tick marks, not between them. pub(super) mid: bool, - pub(super) min: Option, - pub(super) max: Option, + pub(crate) min: Option, + pub(crate) max: Option, /// Number format of the tick labels, when not linked to the source data. - pub(super) fmt: Option, + pub(crate) fmt: Option, + /// The axis title, when the file gives it one. + pub(crate) title: Option, + /// The title's `a:bodyPr/@rot` in degrees clockwise (`-5400000` is 90° counter-clockwise), + /// when the file gives one. + pub(crate) title_rot: Option, + /// Tick labels, as `c:tickLblPos` says (`none` hides them; a file that says nothing shows + /// them, as Word does). + pub(crate) labels: bool, +} + +/// The shape of a series' point markers (`ST_MarkerStyle`, ECMA-376 §21.2.3.28). +#[derive(Clone, Copy, Default, PartialEq, Eq, Debug)] +pub(super) enum Symbol { + #[default] + Circle, + Square, + Diamond, + Triangle, + X, + Plus, + Star, + Dash, + Dot, +} + +impl Symbol { + /// The symbol a `c:symbol` value names; anything unknown (`auto`, `picture`) is a circle. + pub(super) fn parse(name: &str) -> Symbol { + match name { + "square" => Symbol::Square, + "diamond" => Symbol::Diamond, + "triangle" => Symbol::Triangle, + "x" => Symbol::X, + "plus" => Symbol::Plus, + "star" => Symbol::Star, + "dash" => Symbol::Dash, + "dot" => Symbol::Dot, + _ => Symbol::Circle, + } + } } pub(super) struct Series { @@ -70,9 +137,21 @@ pub(super) struct Series { /// when missing or not a valid number. pub(super) idx: Option, pub(super) color: Option, - pub(super) line_w: f32, + /// Fixed line width in points, `None`: the chart's default. + pub(super) line_w: Option, /// Whether the series is drawn as a line (scatter series without one get markers). pub(super) line: bool, + /// The series' own fill: `Some` colour, `None` colour with `no_fill` set, else the chart's + /// accent for the series. + pub(super) no_fill: bool, + /// Markers on the series' points, when the file says. + pub(super) markers: Option, + /// The marker shape (`c:marker/c:symbol`; a circle when missing or unknown) and its size in + /// points (`c:marker/c:size`), when the file gives one. + pub(super) symbol: Symbol, + pub(super) marker_size: Option, + /// Draw the series' line as a smooth curve (`c:smooth`). + pub(super) smooth: bool, pub(super) values: Vec>, /// Scatter x values (index + 1 when the file has none). pub(super) xs: Vec>, @@ -83,6 +162,13 @@ pub(super) struct Series { pub(super) x_fmt: Option, /// Data labels on or off for this series (`c:dLbls` of its own), when it says. pub(super) labels: Option, + /// The labels' number format (`c:dLbls/c:numFmt`): `None` when not given (the chart type's + /// applies), `Some(None)` when linked to the source, else `Some(Some(code))`. + pub(super) label_fmt: Option>, + /// The data labels' `c:dLblPos`, when the file gives a known one. + pub(super) label_pos: Option, + /// The marker's own fill (`c:marker/c:spPr`); without it markers take the automatic colour. + pub(super) marker_color: Option, /// The values as the file has them, for data labels (`values` become shares in a 100% /// stacked chart); empty when the series shows no labels. pub(super) raw: Vec>, diff --git a/crates/docx/src/read/chart/parse.rs b/crates/docx/src/read/chart/parse.rs index 3c09fcfe9..522d50b3b 100644 --- a/crates/docx/src/read/chart/parse.rs +++ b/crates/docx/src/read/chart/parse.rs @@ -1,7 +1,7 @@ use super::format::{date_text, excel_stamp, format_value, is_date_code}; -use super::model::{Axis, Grouping, Kind, Pos, Series}; +use super::model::{Axis, Grouping, Kind, LabelPos, Pos, Series, Symbol}; use super::plot::label; -use super::{Chart, El, FONT, LINE_W, MAX_CELLS, MAX_POINTS, MAX_RADAR_SERIES, MAX_SERIES, MAX_SPOKES, Rgb, fill_colour, int, measure, slot}; +use super::{Chart, El, FONT, MAX_CELLS, MAX_POINTS, MAX_RADAR_SERIES, MAX_SERIES, MAX_SPOKES, Rgb, fill_colour, int, measure, slot}; impl Chart { pub(super) fn parse(space: &El, theme: &[Rgb]) -> Option { @@ -39,10 +39,15 @@ impl Chart { cats.resize(n, String::new()); let radar_style = ty.child("c:radarStyle").and_then(|r| r.attr("val")); let type_labels = show_values(ty.child("c:dLbls")); + let type_fmt = label_format(ty.child("c:dLbls")); for s in &mut series { s.values.resize(n, None); s.xs.resize(n, None); s.labels = Some(s.labels.or(type_labels).unwrap_or(false)); + // Absent (`None`) inherits the chart type's; explicitly source-linked (`Some(None)`) does not. + if s.label_fmt.is_none() { + s.label_fmt.clone_from(&type_fmt); + } if s.labels == Some(true) { s.raw = s.values.clone(); } @@ -98,6 +103,13 @@ impl Chart { s.and_then(|s| s.color).unwrap_or_else(|| self.accent(s.and_then(|s| s.idx).unwrap_or(j))) } + /// Colour of series `j`'s markers: the marker's own fill, else the automatic colour of the + /// series index (not the series' explicit line colour, as Word draws them). + pub(super) fn marker_colour(&self, j: usize) -> Rgb { + let s = self.series.get(j); + s.and_then(|s| s.marker_color).unwrap_or_else(|| self.accent(s.and_then(|s| s.idx).unwrap_or(j))) + } + /// Colour of point `i` of series `j`: its `c:dPt` override, else the varied accent (pie) or /// the series colour. pub(super) fn point_colour(&self, j: usize, i: usize) -> Rgb { @@ -149,7 +161,7 @@ fn to_percent(series: &mut [Series], n: usize) { } /// The category axis (scatter: the horizontal value axis) and the value axis of a plot area. -fn chart_axes(plot: &El, scatter: bool) -> (Axis, Axis) { +pub(crate) fn chart_axes(plot: &El, scatter: bool) -> (Axis, Axis) { let vals_ax: Vec<&El> = plot.children("c:valAx").collect(); let horiz_pos = |a: &&El| matches!(a.child("c:axPos").and_then(|p| p.attr("val")), Some("b" | "t")); let (val_el, x_el) = if scatter { @@ -180,6 +192,17 @@ fn show_values(d: Option<&El>) -> Option { d.child("c:showVal").map(|v| v.attr("val").is_none_or(truthy)) } +/// The number format of data labels (`c:dLbls/c:numFmt`): `None` when the labels say nothing, +/// `Some(None)` when they are linked to the source (so a chart-level format does not apply), +/// `Some(Some(code))` for their own format code. +fn label_format(d: Option<&El>) -> Option> { + let f = d?.child("c:numFmt")?; + if f.attr("sourceLinked").is_some_and(truthy) { + return Some(None); + } + f.attr("formatCode").filter(|f| f.chars().count() <= craft_wordmodel::chart::MAX_FMT).map(|f| Some(f.to_string())) +} + fn legend_pos(chart: &El) -> Option { chart.child("c:legend").map(|l| match l.child("c:legendPos").and_then(|p| p.attr("val")).unwrap_or("r") { "l" => Pos::Left, @@ -206,9 +229,24 @@ impl Series { let sp = ser.child("c:spPr"); let ln = sp.and_then(|p| p.child("a:ln")); let marker = ser.child("c:marker").and_then(|m| m.child("c:spPr")).and_then(|p| fill_colour(p, theme)); + let marker_color = marker; let color = sp.and_then(|p| fill_colour(p, theme)).or_else(|| ln.and_then(|l| fill_colour(l, theme))).or(marker); - let line_w = ln.and_then(|l| l.attr("w")).and_then(|v| measure(v, 12_700.0)).map_or(LINE_W, |w| w.clamp(0.0, 20.0)); + let no_fill = sp.and_then(|p| p.child("a:noFill")).is_some(); + let line_w = ln.and_then(|l| l.attr("w")).and_then(|v| measure(v, 12_700.0)).map(|w| w.clamp(0.0, 20.0)); let line = ln.is_none_or(|l| l.child("a:noFill").is_none()); + // `c:marker/c:symbol`: which symbol each point gets (`none` for no markers). Scatter + // series show markers unless the file says not; line series only when it says so. + let symbol = ser.child("c:marker").and_then(|m| m.child("c:symbol")).and_then(|s| s.attr("val")).map(str::to_string); + let markers = symbol.as_deref().map(|s| s != "none").or_else(|| scatter.then_some(true)); + let symbol = symbol.as_deref().map_or(Symbol::Circle, Symbol::parse); + let marker_size = ser + .child("c:marker") + .and_then(|m| m.child("c:size")) + .and_then(|s| s.attr("val")) + .and_then(|v| v.trim().parse::().ok()) + .filter(|v| v.is_finite()) + .map(|v| v.clamp(2.0, 72.0)); + let smooth = ser.child("c:smooth").and_then(|s| s.attr("val")).is_some_and(truthy); let (values, fmt) = numbers(ser.child(if scatter { "c:yVal" } else { "c:val" }), cap); let (mut xs, x_fmt) = if scatter { numbers(ser.child("c:xVal"), cap) } else { (Vec::new(), None) }; if scatter && xs.is_empty() { @@ -223,7 +261,30 @@ impl Series { } } let labels = show_values(ser.child("c:dLbls")); - Series { name, idx, color, line_w, line, values, xs, points, fmt, x_fmt, labels, raw: Vec::new() } + let label_fmt = label_format(ser.child("c:dLbls")); + let label_pos = ser.child("c:dLbls").and_then(|d| d.child("c:dLblPos")).and_then(|p| p.attr("val")).and_then(LabelPos::parse); + Series { + name, + idx, + color, + line_w, + line, + no_fill, + markers, + symbol, + marker_size, + smooth, + values, + xs, + points, + fmt, + x_fmt, + labels, + label_fmt, + label_pos, + marker_color, + raw: Vec::new(), + } } } @@ -290,7 +351,7 @@ fn pts(cache: &El, cap: usize) -> Vec> { } pub(super) fn axis(a: Option<&El>) -> Axis { - let Some(a) = a else { return Axis { deleted: true, ..Default::default() } }; + let Some(a) = a else { return Axis { deleted: true, labels: true, line: true, ..Default::default() } }; let scaling = a.child("c:scaling"); let num = |n: &str| scaling.and_then(|s| s.child(n)).and_then(|e| e.attr("val")).and_then(|v| v.trim().parse::().ok()).filter(|v| v.is_finite()); @@ -305,8 +366,16 @@ pub(super) fn axis(a: Option<&El>) -> Axis { .child("c:numFmt") .filter(|f| f.attr("sourceLinked") != Some("1")) .and_then(|f| f.attr("formatCode")) - .filter(|f| f.len() <= MAX_FORMAT_CODE) + .filter(|f| f.chars().count() <= craft_wordmodel::chart::MAX_FMT) .map(str::to_string), + title: a.child("c:title").and_then(title_text), + title_rot: a + .child("c:title") + .and_then(|t| t.find("a:bodyPr")) + .and_then(|b| b.attr("rot")) + .and_then(|v| v.trim().parse::().ok()) + .map(|v| (v as f64 / 60_000.0).clamp(-360.0, 360.0) as f32), + labels: a.child("c:tickLblPos").and_then(|p| p.attr("val")) != Some("none"), } } diff --git a/crates/docx/src/read/chart/plot.rs b/crates/docx/src/read/chart/plot.rs index 652ee79a8..22d05b554 100644 --- a/crates/docx/src/read/chart/plot.rs +++ b/crates/docx/src/read/chart/plot.rs @@ -1,9 +1,12 @@ -use super::canvas::{Rect, fill, rect_segs, runs, seg, stroke, text, text_at, text_w, widest, wrap}; +use super::canvas::{Rect, clip_text, fill, rect_segs, runs, seg, stroke, text, text_angle, text_at, text_w, widest, wrap}; use super::format::format_value; -use super::model::{DataLabel, Grouping, Kind, Series}; +use super::model::{DataLabel, Grouping, Kind, LabelPos, Series, Symbol}; use super::parse::category_text; use super::polar::{draw_data_labels, label_of}; -use super::{AXIS, Chart, El, GAP, GRID, GraphicItem, LINE, MAX_LABEL, MAX_TICKS, PathSeg, Rgb, ShapeKind, TextAlign, fill_colour, measure}; +use super::{AXIS, Chart, El, GAP, GRID, GraphicItem, LINE, LINE_W, MAX_LABEL, MAX_TICKS, PathSeg, Rgb, ShapeKind, TextAlign, fill_colour, measure}; + +/// Line width of the stroked marker symbols (x, plus, star), points. +const MARKER_STROKE: f32 = 1.5; pub(super) fn label(s: &str) -> String { s.chars().take(MAX_LABEL).collect() @@ -142,6 +145,8 @@ impl Axes { /// The room around the plot for axis labels, and the category labels as they'll be drawn. struct Margins { left: f32, + /// Part of `left` taken by the rotated category-axis title (bars): the labels start after it. + title: f32, top: f32, bottom: f32, /// Category labels in lines (none for scatter charts or a deleted category axis). @@ -151,22 +156,25 @@ struct Margins { } /// The margins the axis labels need around the plot in `area`. Category labels under a column -/// chart wrap to the band; labels that still don't fit show every k-th category. +/// chart wrap to the band; labels that still don't fit show every k-th category. Hidden tick +/// labels take no room; an axis title adds one line on its side. fn axis_margins(c: &Chart, ax: &Axes, area: Rect) -> Margins { let fs = c.text_size; let lh = fs * LINE; let scatter = c.kind == Kind::Scatter; let (cat_on, val_on) = (!c.cat.deleted, !c.val.deleted); + let (cat_labels, val_labels) = (cat_on && c.cat.labels, val_on && c.val.labels); + let title = title_left(c, cat_on, val_on, area.w); let left = if c.horizontal { - if cat_on { widest(&c.cats, fs).min(area.w * 0.4) + GAP } else { 0.0 } - } else if val_on { + if cat_labels { widest(&c.cats, fs).min(area.w * 0.4) + GAP } else { 0.0 } + } else if val_labels { ax.vticks.iter().map(|(_, s)| text_w(s, fs)).fold(0.0, f32::max) + GAP } else { 0.0 - }; + } + title; let band_w = (area.w - left).max(1.0) / c.n as f32; - let wrapped: Vec> = if cat_on && !scatter { - let width = if c.horizontal { left - GAP } else { band_w }; + let wrapped: Vec> = if cat_labels && !scatter { + let width = if c.horizontal { left - title - GAP } else { band_w }; c.cats.iter().map(|s| wrap(s, width, fs)).collect() } else { Vec::new() @@ -174,15 +182,48 @@ fn axis_margins(c: &Chart, ax: &Axes, area: Rect) -> Margins { let widest_line = wrapped.iter().flatten().map(|s| text_w(s, fs)).fold(0.0, f32::max); let skip = ((widest_line / band_w).ceil() as usize).max(1); let label_lines = wrapped.iter().map(Vec::len).max().unwrap_or(1) as f32; - let bottom = if cat_on && !c.horizontal { + let bottom = (if cat_labels && !c.horizontal { if scatter { lh + GAP } else { label_lines * lh + GAP } - } else if c.horizontal && val_on { + } else if c.horizontal && val_labels { lh + GAP } else { 0.0 - }; - let top = if !c.horizontal && val_on { lh / 2.0 } else { 0.0 }; - Margins { left, top, bottom, wrapped, skip } + }) + title_bottom(c, cat_on, val_on, lh); + let top = if !c.horizontal && val_labels { lh / 2.0 } else { 0.0 }; + Margins { left, title, top, bottom, wrapped, skip } +} + +/// Room for the vertical axis title on the plot's left: one line when the axis that sits there +/// (the value axis for columns, the category axis for bars) has a title. +/// A horizontal title (`rot="0"`) gets a strip as wide as its wrapped lines, at most a quarter of +/// `area_w`, so the plot keeps its room. +fn title_left(c: &Chart, cat_on: bool, val_on: bool, area_w: f32) -> f32 { + let side = if c.horizontal { cat_on && c.cat.title.is_some() } else { val_on && c.val.title.is_some() }; + if !side { + return 0.0; + } + match flat_title_lines(c, area_w) { + Some(lines) => widest(&lines, c.text_size).min((area_w * 0.25).max(0.0)) + GAP, + None => c.text_size * LINE, + } +} + +/// The wrapped lines of the left axis's title when the file keeps it horizontal (`rot="0"`). +fn flat_title_lines(c: &Chart, area_w: f32) -> Option> { + let left = if c.horizontal { &c.cat } else { &c.val }; + let r = left.title_rot?.rem_euclid(360.0); + if !(r < 1.0 || r > 359.0) { + return None; + } + let width = if area_w.is_finite() { (area_w * 0.18).max(1.0) } else { 1.0 }; + left.title.as_ref().map(|t| wrap(t, width, c.text_size)) +} + +/// Room for the horizontal axis title under the plot: one line when the axis that sits there +/// (the category axis for columns, the value axis for bars) has a title. +fn title_bottom(c: &Chart, cat_on: bool, val_on: bool, lh: f32) -> f32 { + let side = if c.horizontal { val_on && c.val.title.is_some() } else { cat_on && c.cat.title.is_some() }; + if side { lh } else { 0.0 } } /// A bar, column, line, area or scatter chart in `area`: gridlines, series, axes, then labels. @@ -200,6 +241,9 @@ pub(super) fn cartesian(out: &mut Vec, c: &Chart, area: Rect) { if c.val.grid { draw_grid(out, &g, &ax.vticks); } + if c.cat.grid { + draw_cat_grid(out, &g, &ax.xticks, c.kind); + } let mut labels = Vec::new(); match c.kind { Kind::Bar => bars(out, c, &g, &mut labels), @@ -210,12 +254,49 @@ pub(super) fn cartesian(out: &mut Vec, c: &Chart, area: Rect) { } draw_axes(out, c, &g); draw_data_labels(out, &labels, c.text_size, g.horiz && c.kind == Kind::Bar); - if !c.val.deleted { - draw_value_labels(out, c, &g, &ax.vticks, Rect { w: m.left, ..area }); + if !c.val.deleted && c.val.labels { + let room = title_left(c, !c.cat.deleted, true, area.w); + draw_value_labels(out, c, &g, &ax.vticks, Rect { x: area.x + room, w: m.left - room, ..area }); } - if !c.cat.deleted { + if !c.cat.deleted && c.cat.labels { draw_category_labels(out, c, &g, &ax.xticks, &m, area); } + draw_axis_titles(out, c, &g, area); +} + +/// The axis titles, one line: the title of the vertical axis (the value axis for columns, the +/// category axis for bars) is turned 90° on the plot's left; the horizontal axis's title sits +/// under the plot. +fn draw_axis_titles(out: &mut Vec, c: &Chart, g: &Frame, area: Rect) { + let lh = c.text_size * LINE; + // The axis drawn along the plot's left (the value axis for columns, the category axis for + // bars) gets its title rotated there; the other axis's title sits under the plot. + let (left, bottom) = if c.horizontal { (&c.cat, &c.val) } else { (&c.val, &c.cat) }; + // Word turns the vertical axis's title counter-clockwise (reading upwards) unless the file + // turns it the other way or keeps it horizontal (`rot="0"`, wrapped in its own strip); any + // other rotation would not fit the strip and is read as counter-clockwise. + let up = if left.title_rot.is_some_and(|r| (r.rem_euclid(360.0) - 90.0).abs() < 1.0) { 90.0 } else { -90.0 }; + if !left.deleted + && let Some(lines) = flat_title_lines(c, area.w) + { + let strip = (g.plot.x - area.x - GAP).max(1.0); + let top = g.plot.y + g.plot.h / 2.0 - lines.len() as f32 * lh / 2.0; + for (k, l) in lines.iter().enumerate() { + let l = clip_text(l, strip, c.text_size); + text_angle(out, Rect { x: area.x, y: top + k as f32 * lh, w: strip, h: lh }, &l, c.text_size, TextAlign::Center, false, 0.0); + } + } else if !left.deleted + && let Some(t) = &left.title + { + let t = clip_text(t, g.plot.h, c.text_size); + text_angle(out, Rect { x: area.x, y: g.plot.y, w: lh, h: g.plot.h }, &t, c.text_size, TextAlign::Center, false, up); + } + if !bottom.deleted + && let Some(t) = &bottom.title + { + let t = clip_text(t, g.plot.w, c.text_size); + text_angle(out, Rect { x: g.plot.x, y: area.bottom() - lh, w: g.plot.w, h: lh }, &t, c.text_size, TextAlign::Center, false, 0.0); + } } /// Major gridlines at the value ticks. @@ -228,6 +309,31 @@ fn draw_grid(out: &mut Vec, g: &Frame, ticks: &[(f64, String)]) { } } +/// Gridlines along the category axis: across the plot at each category's centre (where line and +/// area points sit, on the tick marks when the axis crosses "midCat"), at the x ticks of a +/// scatter chart. +fn draw_cat_grid(out: &mut Vec, g: &Frame, xticks: &[(f64, String)], kind: Kind) { + let plot = g.plot; + if kind == Kind::Scatter { + for (t, _) in xticks { + let p = g.scat_x(*t); + stroke(out, seg(p, plot.y, p, plot.bottom()), GRID, 0.75); + } + return; + } + for i in 0..g.n { + let segs = if g.horiz { + let p = g.band_start(i) + g.band() / 2.0; + seg(plot.x, p, plot.right(), p) + } else { + // Bars stay in their bands; only line and area points move onto the ticks. + let p = if matches!(kind, Kind::Line | Kind::Area) { g.cat_x(i) } else { g.band_start(i) + g.band() / 2.0 }; + seg(p, plot.y, p, plot.bottom()) + }; + stroke(out, segs, GRID, 0.75); + } +} + /// The category axis line, the plot area's outline and the value axis line, as the file shows them. fn draw_axes(out: &mut Vec, c: &Chart, g: &Frame) { let plot = g.plot; @@ -271,7 +377,7 @@ fn draw_category_labels(out: &mut Vec, c: &Chart, g: &Frame, xticks for (i, lines) in m.wrapped.iter().enumerate() { let top = g.band_start(i) + g.band() / 2.0 - lines.len() as f32 * lh / 2.0; for (k, s) in lines.iter().enumerate() { - text(out, Rect { x: area.x, y: top + k as f32 * lh, w: m.left - GAP, h: lh }, s, fs, TextAlign::Right, false); + text(out, Rect { x: area.x + m.title, y: top + k as f32 * lh, w: m.left - m.title - GAP, h: lh }, s, fs, TextAlign::Right, false); } } } else { @@ -361,7 +467,13 @@ fn bars(out: &mut Vec, c: &Chart, g: &Frame, labels: &mut Vec, c: &Chart, g: &Frame, labels: &mut Vec (if v >= 0.0 { p1 } else { p0 }, mid), (false, false) => (mid, if v >= 0.0 { p0 } else { p1 }), }; - labels.push(DataLabel { x, y, text, inside: stacked }); + labels.push(DataLabel { x, y, text, inside: stacked, pos: LabelPos::Above, off: 0.0 }); } } } @@ -401,13 +513,109 @@ fn lines(out: &mut Vec, c: &Chart, g: &Frame, labels: &mut Vec f32 { + if drawn { s.marker_size.filter(|v| v.is_finite()).unwrap_or(6.0).clamp(2.0, 72.0) / 2.0 } else { 0.0 } +} + +fn markers(out: &mut Vec, pts: impl Iterator, colour: Rgb, s: &Series) { + const MARKER: f32 = 6.0; + let size = s.marker_size.filter(|v| v.is_finite()).unwrap_or(MARKER).clamp(2.0, 72.0); + let h = size / 2.0; + let shape = |out: &mut Vec, rect: [f32; 4], kind: ShapeKind| { + out.push(GraphicItem::Shape { rect, kind, fill: Some(colour), stroke: None, stroke_width: 0.0 }); + }; + let mut strokes = Vec::new(); + for (x, y) in pts { + match s.symbol { + Symbol::Circle => shape(out, [x - h, y - h, size, size], ShapeKind::Ellipse), + Symbol::Square => shape(out, [x - h, y - h, size, size], ShapeKind::Rectangle), + Symbol::Dot => shape(out, [x - h / 2.0, y - h / 2.0, h, h], ShapeKind::Ellipse), + Symbol::Dash => shape(out, [x - h, y - size / 6.0, size, size / 3.0], ShapeKind::Rectangle), + Symbol::Diamond => fill( + out, + vec![PathSeg::Move(x, y - h), PathSeg::Line(x + h, y), PathSeg::Line(x, y + h), PathSeg::Line(x - h, y), PathSeg::Close], + colour, + ), + Symbol::Triangle => { + fill(out, vec![PathSeg::Move(x, y - h), PathSeg::Line(x + h, y + h), PathSeg::Line(x - h, y + h), PathSeg::Close], colour) + } + Symbol::X | Symbol::Plus | Symbol::Star => { + if s.symbol != Symbol::Plus { + strokes.extend(seg(x - h, y - h, x + h, y + h)); + strokes.extend(seg(x - h, y + h, x + h, y - h)); + } + if s.symbol != Symbol::X { + strokes.extend(seg(x - h, y, x + h, y)); + strokes.extend(seg(x, y - h, x, y + h)); + } + } + } + } + stroke(out, strokes, colour, MARKER_STROKE); +} + +/// A polyline through `pts`, a smooth curve (Catmull-Rom pieces) when `smooth` and there are at +/// least three points in a row; gaps (`None`) end a run. +fn curve(pts: impl Iterator>, smooth: bool) -> Vec { + if !smooth { + return runs(pts); + } + let pts: Vec> = pts.collect(); + let mut segs = Vec::new(); + let mut i = 0; + while i < pts.len() { + let Some(first) = pts[i] else { + i += 1; + continue; + }; + // Collect one run of points. + let mut run = vec![first]; + while let Some(p) = pts.get(i + 1).copied().flatten() { + run.push(p); + i += 1; + } + i += 1; + if run.len() < 3 { + segs.extend(runs(run.into_iter().map(Some))); + continue; + } + segs.push(PathSeg::Move(run[0].0, run[0].1)); + for k in 0..run.len() - 1 { + // Catmull-Rom: tangent at p1 from p0 to p2 (ends mirror their own tangent). + let p0 = run[k.max(1) - 1]; + let p1 = run[k]; + let p2 = run[k + 1]; + let p3 = run.get(k + 2).copied().unwrap_or(p2); + let c1 = (p1.0 + (p2.0 - p0.0) / 6.0, p1.1 + (p2.1 - p0.1) / 6.0); + let c2 = (p2.0 - (p3.0 - p1.0) / 6.0, p2.1 - (p3.1 - p1.1) / 6.0); + segs.push(PathSeg::Cubic(c1.0, c1.1, c2.0, c2.1, p2.0, p2.1)); + } + } + segs +} + fn areas(out: &mut Vec, c: &Chart, g: &Frame, labels: &mut Vec) { let mut base = vec![0.0f64; g.n]; for (j, s) in c.series.iter().enumerate().filter(|(_, s)| has_values(s)) { @@ -422,10 +630,12 @@ fn areas(out: &mut Vec, c: &Chart, g: &Frame, labels: &mut Vec, c: &Chart, g: &Frame, labels: &mut Vec, c: &Chart, g: &Frame, labels: &mut Vec) { - const MARKER: f32 = 6.0; for (j, s) in c.series.iter().enumerate() { let pts: Vec> = s.xs.iter() @@ -448,16 +657,21 @@ fn draw_scatter(out: &mut Vec, c: &Chart, g: &Frame, labels: &mut V .collect(); for (i, p) in pts.iter().enumerate() { if let (Some((x, y)), Some(text)) = (p, label_of(s, i)) { - labels.push(DataLabel { x: *x, y: *y - MARKER / 2.0, text, inside: false }); + labels.push(DataLabel { + x: *x, + y: *y, + text, + inside: false, + pos: s.label_pos.unwrap_or(LabelPos::Right), + off: marker_half(s, s.markers != Some(false)) + 2.0, + }); } } if s.line { - stroke(out, runs(pts.iter().copied()), c.series_colour(j), s.line_w); - continue; + stroke(out, curve(pts.iter().copied(), s.smooth), c.series_colour(j), s.line_w.unwrap_or(LINE_W)); } - for (x, y) in pts.iter().flatten() { - let rect = [x - MARKER / 2.0, y - MARKER / 2.0, MARKER, MARKER]; - out.push(GraphicItem::Shape { rect, kind: ShapeKind::Ellipse, fill: Some(c.series_colour(j)), stroke: None, stroke_width: 0.0 }); + if s.markers != Some(false) { + markers(out, pts.iter().flatten().copied(), c.marker_colour(j), s); } } } diff --git a/crates/docx/src/read/chart/polar.rs b/crates/docx/src/read/chart/polar.rs index 57ad10472..cf5e7452e 100644 --- a/crates/docx/src/read/chart/polar.rs +++ b/crates/docx/src/read/chart/polar.rs @@ -1,8 +1,8 @@ use super::canvas::{Rect, clip_text, fill, stroke, text, text_w, widest}; use super::format::format_value; -use super::model::{DataLabel, Kind, Series}; +use super::model::{DataLabel, Kind, LabelPos, Series}; use super::plot::{Axes, has_values}; -use super::{Chart, FRAC_PI_2, GAP, GRID, GraphicItem, LINE, PathSeg, Rgb, ShapeKind, TAU, TEXT, TextAlign}; +use super::{Chart, FRAC_PI_2, GAP, GRID, GraphicItem, LINE, LINE_W, PathSeg, Rgb, ShapeKind, TAU, TEXT, TextAlign}; pub(super) fn pie(out: &mut Vec, c: &Chart, area: Rect) { let Some(s) = c.series.first() else { return }; @@ -38,7 +38,7 @@ pub(super) fn pie(out: &mut Vec, c: &Chart, area: Rect) { // Halfway across the ring (pie: two thirds out from the centre). let at = if inner > 0.0 { (r + inner) / 2.0 } else { r * 0.65 }; let (x, y) = polar(cx, cy, at, (a + b) / 2.0); - labels.push(DataLabel { x: x as f32, y: y as f32, text, inside: true }); + labels.push(DataLabel { x: x as f32, y: y as f32, text, inside: true, pos: LabelPos::Above, off: 0.0 }); } a = b; } @@ -92,14 +92,14 @@ pub(super) fn radar(out: &mut Vec, c: &Chart, area: Rect) { if c.radar_fill { fill(out, segs.clone(), col); } - stroke(out, segs, col, s.line_w); + stroke(out, segs, col, s.line_w.unwrap_or(LINE_W)); for (i, p) in ring.iter().enumerate() { if c.radar_marks && s.values.get(i).copied().flatten().is_some() { let rect = [p.0 as f32 - MARKER / 2.0, p.1 as f32 - MARKER / 2.0, MARKER, MARKER]; out.push(GraphicItem::Shape { rect, kind: ShapeKind::Ellipse, fill: Some(col), stroke: None, stroke_width: 0.0 }); } if let Some(text) = label_of(s, i) { - labels.push(DataLabel { x: p.0 as f32, y: p.1 as f32, text, inside: false }); + labels.push(DataLabel { x: p.0 as f32, y: p.1 as f32, text, inside: false, pos: LabelPos::Above, off: 0.0 }); } } } @@ -134,13 +134,13 @@ pub(super) fn radar(out: &mut Vec, c: &Chart, area: Rect) { } /// The data label of point `i` of series `s`, when it shows one: the value as the file has it, -/// in the series' number format. +/// in the labels' own number format, else the series' one. pub(super) fn label_of(s: &Series, i: usize) -> Option { if s.labels != Some(true) { return None; } let v = s.raw.get(i).copied().flatten()?; - let fmt = s.fmt.as_deref().filter(|f| !f.eq_ignore_ascii_case("General")); + let fmt = s.label_fmt.as_ref().map_or(s.fmt.as_deref(), |f| f.as_deref().or(s.fmt.as_deref())).filter(|f| !f.eq_ignore_ascii_case("General")); Some(match fmt { Some(f) => format_value(v, f, 1.0), None => general_value(v), @@ -165,13 +165,22 @@ pub(super) fn draw_data_labels(out: &mut Vec, labels: &[DataLabel], let w = text_w(&l.text, fs) + 4.0; let (rect, align) = if l.inside { (Rect { x: l.x - w / 2.0, y: l.y - lh / 2.0, w, h: lh }, TextAlign::Center) + } else if l.off > 0.0 || l.pos != LabelPos::Above { + let o = l.off; + match l.pos { + LabelPos::Right => (Rect { x: l.x + o, y: l.y - lh / 2.0, w, h: lh }, TextAlign::Left), + LabelPos::Left => (Rect { x: l.x - o - w, y: l.y - lh / 2.0, w, h: lh }, TextAlign::Right), + LabelPos::Below => (Rect { x: l.x - w / 2.0, y: l.y + o, w, h: lh }, TextAlign::Center), + LabelPos::Center => (Rect { x: l.x - w / 2.0, y: l.y - lh / 2.0, w, h: lh }, TextAlign::Center), + LabelPos::Above => (Rect { x: l.x - w / 2.0, y: l.y - o - lh, w, h: lh }, TextAlign::Center), + } } else if beside_right { (Rect { x: l.x + GAP / 2.0, y: l.y - lh / 2.0, w, h: lh }, TextAlign::Left) } else { (Rect { x: l.x - w / 2.0, y: l.y - lh, w, h: lh }, TextAlign::Center) }; let color = if l.inside { Rgb::WHITE } else { TEXT }; - out.push(GraphicItem::Text { rect: rect.arr(), text: l.text.clone(), size: fs, color, bold: false, align, font: None }); + out.push(GraphicItem::Text { rect: rect.arr(), text: l.text.clone(), size: fs, color, bold: false, align, font: None, angle: 0.0 }); } } diff --git a/crates/docx/src/read/chart/tests.rs b/crates/docx/src/read/chart/tests.rs index 6e8bcfd82..ae80c028e 100644 --- a/crates/docx/src/read/chart/tests.rs +++ b/crates/docx/src/read/chart/tests.rs @@ -386,9 +386,13 @@ fn axis_format_code_length_is_capped() { let x = format!(r#""#); axis(Some(&crate::xml::parse(x.as_bytes()).unwrap())).fmt }; - let ok = format!("{}0", "x".repeat(254)); + use craft_wordmodel::chart::MAX_FMT; + // The limit is the model's, counted in characters (a CJK character is one). + let ok = format!("{}0", "x".repeat(MAX_FMT - 1)); assert_eq!(fmt(&ok), Some(ok.clone())); - assert_eq!(fmt(&format!("{}0", "x".repeat(300))), None); + let cjk = format!("{}0", "\u{5143}".repeat(MAX_FMT - 1)); + assert_eq!(fmt(&cjk), Some(cjk.clone())); + assert_eq!(fmt(&format!("{}0", "x".repeat(MAX_FMT))), None); } #[test] @@ -590,3 +594,308 @@ fn series_without_a_usable_c_idx_take_their_place() { let big = [renumbered(&a, &format!(r#""#, u32::MAX))]; assert_eq!(column_colours(&big).0.len(), 1); } + +#[test] +fn bar_category_labels_start_after_the_axis_title() { + let long = "A category label that is much too long for its column"; + let title = r#"Region"#; + let plot = format!( + r#"{}{title}"#, + ser(0, "S", Some("FF0000"), &[long, "Short"], &["4", "8"]), + ); + let items = chart_items(&space(&plot), &theme(), 300.0, 200.0); + let rect_of = |f: &dyn Fn(&str, &TextAlign) -> bool| -> Vec<[f32; 4]> { + items + .iter() + .filter_map(|it| if let GraphicItem::Text { rect, text, align, .. } = it { f(text, align).then_some(*rect) } else { None }) + .collect() + }; + let title_rect = rect_of(&|t, _| t == "Region"); + assert_eq!(title_rect.len(), 1); + let strip_end = title_rect[0][0] + title_rect[0][2]; + let labels = rect_of(&|t, a| matches!(a, TextAlign::Right) && t != "Region" && (long.contains(t) || t == "Short")); + assert!(!labels.is_empty()); + assert!(labels.iter().all(|r| r[0] >= strip_end - 1e-3), "labels {labels:?} overlap the title strip ending at {strip_end}"); + assert!(labels.iter().all(|r| r[2] <= 0.4 * 300.0), "{labels:?}"); +} + +#[test] +fn category_gridlines_follow_the_points_when_the_axis_crosses_mid() { + let line = |cross: &str| { + let plot = format!( + r#"{}"#, + ser(0, "L", Some("FF0000"), &["Q1", "Q2", "Q3"], &["1", "5", "3"]), + ); + let items = chart_items(&space(&plot), &theme(), 300.0, 200.0); + let grid: Vec = items + .iter() + .filter_map(|it| match it { + GraphicItem::Path { segs, stroke: Some(c), .. } if *c == GRID => match segs.first() { + Some(PathSeg::Move(x, _)) => Some(*x), + _ => None, + }, + _ => None, + }) + .collect(); + let pts: Vec = items + .iter() + .filter_map(|it| match it { + GraphicItem::Path { segs, stroke: Some(c), .. } if *c == Rgb(255, 0, 0) => Some( + segs.iter().filter_map(|s| if let PathSeg::Move(x, _) | PathSeg::Line(x, _) = s { Some(*x) } else { None }).collect::>(), + ), + _ => None, + }) + .next() + .unwrap_or_default(); + (grid, pts) + }; + for cross in ["midCat", "between"] { + let (grid, pts) = line(cross); + assert_eq!(grid.len(), 3, "{cross}"); + assert_eq!(pts.len(), 3, "{cross}"); + assert!(grid.iter().zip(&pts).all(|(g, p)| (g - p).abs() < 1e-3), "{cross}: grid {grid:?} points {pts:?}"); + } +} + +/// A line chart with one series; `extra` goes into the series (`c:marker`, `c:dLbls`), `axes` +/// into the value axis. +fn line_with(extra: &str, val_axis: &str) -> Vec { + let s = ser(0, "S", Some("FF0000"), &["Q1", "Q2", "Q3"], &["3.2", "4.1", "5"]).replace("", &format!("{extra}")); + let plot = format!( + r#"{s}{val_axis}"# + ); + chart_items(&space(&plot), &theme(), 300.0, 200.0) +} + +fn marker_xml(symbol: &str, size: u32) -> String { + format!(r#""#) +} + +#[test] +fn marker_symbols_are_drawn_as_their_shape() { + let shapes = |items: &[GraphicItem], kind: ShapeKind| { + items + .iter() + .filter(|it| matches!(it, GraphicItem::Shape { kind: k, rect, .. } if *k == kind && rect[2] > 3.0 && rect[2] < 20.0 && rect[3] > 1.0)) + .count() + }; + let paths = |items: &[GraphicItem], filled: bool| { + items + .iter() + .filter(|it| matches!(it, GraphicItem::Path { fill, segs, .. } if fill.is_some() == filled && segs.len() <= 12 && segs.len() >= 3)) + .count() + }; + // One marker per point, in the series' size. + let circle = line_with(&marker_xml("circle", 9), ""); + assert_eq!(shapes(&circle, ShapeKind::Ellipse), 3); + assert!(circle.iter().any(|it| matches!(it, GraphicItem::Shape { kind: ShapeKind::Ellipse, rect, .. } if (rect[2] - 9.0).abs() < 1e-3))); + let square = line_with(&marker_xml("square", 7), ""); + assert_eq!(shapes(&square, ShapeKind::Rectangle) - shapes(&circle, ShapeKind::Rectangle), 3, "a square per point"); + assert_eq!(shapes(&square, ShapeKind::Ellipse), 0, "no circles"); + // Diamonds and triangles are filled polygons; x, plus and star are strokes. + let base = paths(&line_with("", ""), true); + assert_eq!(paths(&line_with(&marker_xml("diamond", 7), ""), true) - base, 3); + assert_eq!(paths(&line_with(&marker_xml("triangle", 7), ""), true) - base, 3); + for symbol in ["x", "plus", "star"] { + let items = line_with(&marker_xml(symbol, 7), ""); + assert!(items.iter().any(|it| matches!(it, GraphicItem::Path { fill: None, stroke: Some(_), segs, .. } if segs.len() >= 6)), "{symbol}"); + assert_eq!(shapes(&items, ShapeKind::Ellipse), 0, "{symbol} is not a circle"); + } + // Dash and dot are small shapes; `none` and unknown names never crash (unknown: a circle). + assert_eq!(shapes(&line_with(&marker_xml("dash", 7), ""), ShapeKind::Rectangle) - shapes(&circle, ShapeKind::Rectangle), 3); + assert_eq!(shapes(&line_with(&marker_xml("dot", 7), ""), ShapeKind::Ellipse), 3); + assert_eq!(shapes(&line_with(&marker_xml("none", 7), ""), ShapeKind::Ellipse), 0); + assert_eq!(shapes(&line_with(&marker_xml("wibble", 7), ""), ShapeKind::Ellipse), 3); + // Hostile sizes are clamped. + for size in ["0", "99999999999", "-4", "x"] { + let items = line_with(&format!(r#""#), ""); + assert!(items.iter().all(|it| match it { + GraphicItem::Shape { rect, .. } => rect.iter().all(|v| v.is_finite()) && rect[2] <= 300.0, + _ => true, + })); + } +} + +#[test] +fn data_labels_use_their_own_format_code() { + let dl = |num: &str| format!(r#"{num}"#); + let shown = |items: &[GraphicItem]| -> Vec { + items.iter().filter_map(|it| if let GraphicItem::Text { text, .. } = it { Some(text.clone()) } else { None }).collect() + }; + let own = shown(&line_with(&dl(r#""#), "")); + assert!(own.contains(&"3.20".to_string()) && own.contains(&"5.00".to_string()), "{own:?}"); + // Linked to the source: the series' format (General here). + let linked = shown(&line_with(&dl(r#""#), "")); + assert!(linked.contains(&"3.2".to_string()) && linked.contains(&"5".to_string()), "{linked:?}"); + // A format code beyond the cap is ignored, not applied. + let long = "0".repeat(craft_wordmodel::chart::MAX_FMT + 1); + let capped = shown(&line_with(&dl(&format!(r#""#)), "")); + assert!(capped.contains(&"3.2".to_string()), "{capped:?}"); +} + +#[test] +fn value_axis_title_turns_counter_clockwise_like_word() { + let title = |body: &str| format!(r#"{body}Revenue"#); + let angle = |items: &[GraphicItem]| { + items.iter().find_map(|it| if let GraphicItem::Text { text, angle, .. } = it { (text == "Revenue").then_some(*angle) } else { None }) + }; + // The file's rotation (`-5400000` is 90° counter-clockwise), or Word's default when it says none. + assert_eq!(angle(&line_with("", &title(r#""#))), Some(-90.0)); + assert_eq!(angle(&line_with("", &title(""))), Some(-90.0)); + assert_eq!(angle(&line_with("", &title(r#""#))), Some(90.0)); + // A horizontal title (`rot="0"`) stays horizontal: the plot makes room for it (see below). + assert_eq!(angle(&line_with("", &title(r#""#))), Some(0.0)); + // Hostile values stay finite. + let odd = angle(&line_with("", &title(r#""#))); + assert!(odd.is_some_and(f32::is_finite)); +} + +fn first_ellipse(items: &[GraphicItem]) -> Option<([f32; 4], Rgb)> { + items.iter().find_map(|it| match it { + GraphicItem::Shape { kind: ShapeKind::Ellipse, rect, fill: Some(c), .. } if rect[2] > 3.0 && rect[2] < 20.0 => Some((*rect, *c)), + _ => None, + }) +} + +#[test] +fn markers_use_the_automatic_series_colour_not_the_line_colour() { + // The series has an explicit red line/fill but no marker fill: Word draws the markers in the + // palette colour of the series index (accent 1 here). + let items = line_with(&marker_xml("circle", 9), ""); + assert_eq!(first_ellipse(&items).map(|e| e.1), Some(theme()[4])); + // A marker's own fill wins. + let own = + r#""#; + assert_eq!(first_ellipse(&line_with(own, "")).map(|e| e.1), Some(Rgb(0, 0xAA, 0))); +} + +#[test] +fn line_data_labels_default_to_the_right_and_honour_dlblpos() { + let dl = |pos: &str| format!(r#"{}{pos}"#, marker_xml("circle", 9)); + let text_rect = |items: &[GraphicItem]| { + items.iter().find_map(|it| if let GraphicItem::Text { text, rect, .. } = it { (text == "3.2").then_some(*rect) } else { None }) + }; + let at = |pos: &str| { + let items = line_with(&dl(pos), ""); + let (m, _) = first_ellipse(&items).unwrap(); + let t = text_rect(&items).unwrap(); + // Centre of the marker and of the label box. + ((t[0] + t[2] / 2.0) - (m[0] + m[2] / 2.0), (t[1] + t[3] / 2.0) - (m[1] + m[3] / 2.0), t, m) + }; + let (dx, dy, t, m) = at(""); + assert!(t[0] >= m[0] + m[2] && dy.abs() < 1.0, "default is right of the marker: {dx} {dy}"); + let (dx, dy, t, m) = at(r#""#); + assert!(t[0] >= m[0] + m[2] && dy.abs() < 1.0, "{dx} {dy}"); + let (dx, dy, _, _) = at(r#""#); + assert!(dx.abs() < 1.0 && dy < -4.0, "{dx} {dy}"); + let (dx, dy, _, _) = at(r#""#); + assert!(dx.abs() < 1.0 && dy > 4.0, "{dx} {dy}"); + let (dx, dy, t, m) = at(r#""#); + assert!(t[0] + t[2] <= m[0] && dy.abs() < 1.0, "{dx} {dy}"); + let (dx, dy, _, _) = at(r#""#); + assert!(dx.abs() < 1.0 && dy.abs() < 1.0, "{dx} {dy}"); + // Unknown positions fall back to the default. + let (_, _, t, m) = at(r#""#); + assert!(t[0] >= m[0] + m[2]); +} + +#[test] +fn horizontal_value_axis_title_gets_its_own_strip() { + let title = |rot: &str, name: &str| { + format!(r#"{name}"#) + }; + let plot_left = |items: &[GraphicItem]| { + // The first gridline's start is the plot's left edge. + items.iter().find_map(|it| match it { + GraphicItem::Path { segs, stroke: Some(_), fill: None, .. } if segs.len() == 2 => match segs.first() { + Some(PathSeg::Move(x, _)) => Some(*x), + _ => None, + }, + _ => None, + }) + }; + let rect_of = |items: &[GraphicItem], name: &str| { + items.iter().find_map(|it| if let GraphicItem::Text { text, rect, .. } = it { text.starts_with(name).then_some(*rect) } else { None }) + }; + let turned = line_with("", &title(r#" rot="-5400000" vert="horz""#, "Revenue")); + let flat = line_with("", &title(r#" rot="0" vert="horz""#, "Revenue")); + let (a, b) = (plot_left(&turned).unwrap(), plot_left(&flat).unwrap()); + assert!(b > a + 10.0, "a horizontal title takes more room than a turned one: {a} {b}"); + // The title box lies left of the plot and inside the chart. + let r = rect_of(&flat, "Revenue").unwrap(); + assert!(r[0] >= 0.0 && r[0] + r[2] <= b + 0.5, "{r:?} {b}"); + // A hostile title never takes more than a quarter of the width and stays finite. + let long = "Wibble ".repeat(500); + let items = line_with("", &title(r#" rot="0""#, &long)); + let p = plot_left(&items).unwrap(); + assert!(p.is_finite() && p <= 300.0 * 0.3, "{p}"); +} + +/// A one-series chart whose series carries `sp` as its `c:spPr`. +fn chart_with_sp(kind: &str, sp: &str) -> Vec { + let s = ser(0, "S", None, &["Q1", "Q2"], &["4", "8"]).replace("", &format!("{sp}")); + let plot = match kind { + "area" => format!( + r#"{s}"# + ), + dir => format!( + r#"{s}"# + ), + }; + chart_items(&space(&plot), &theme(), 300.0, 200.0) +} + +/// The series' closed shapes: (filled, stroked with the given width). +fn series_shapes(items: &[GraphicItem], width: f32) -> (usize, usize) { + let closed = |segs: &[PathSeg]| segs.iter().any(|s| matches!(s, PathSeg::Close)); + let fills = items.iter().filter(|it| matches!(it, GraphicItem::Path { segs, fill: Some(_), .. } if closed(segs))).count(); + let strokes = items + .iter() + .filter(|it| matches!(it, GraphicItem::Path { segs, fill: None, stroke: Some(_), stroke_width, .. } if closed(segs) && (*stroke_width - width).abs() < 0.01)) + .count(); + (fills, strokes) +} + +#[test] +fn unfilled_bars_keep_their_explicit_outline() { + let outlined = r#""#; + let no_line = r#""#; + for dir in ["col", "bar"] { + // Two points, two outlines, no fill. + assert_eq!(series_shapes(&chart_with_sp(dir, outlined), 2.0), (0, 2), "{dir}"); + // Control: no fill and no line draws nothing. + assert_eq!(series_shapes(&chart_with_sp(dir, no_line), 2.0), (0, 0), "{dir}"); + } +} + +#[test] +fn no_fill_area_series_is_not_painted() { + let filled = r#""#; + let none = r#""#; + assert_eq!(series_shapes(&chart_with_sp("area", filled), 2.0).0, 1); + assert_eq!(series_shapes(&chart_with_sp("area", none), 2.0).0, 0); +} + +#[test] +fn source_linked_series_labels_do_not_inherit_the_chart_format() { + let dl = |fmt: &str| format!(r#""#); + let shown = |items: &[GraphicItem]| -> Vec { + items.iter().filter_map(|it| if let GraphicItem::Text { text, .. } = it { Some(text.clone()) } else { None }).collect() + }; + // Chart-level 0%; the series' own labels are `linked` (or have no labels element at all). + let run = |series_dl: &str| { + let s = ser(0, "S", Some("FF0000"), &["Q1", "Q2", "Q3"], &["3.2", "4.1", "5"]).replace("", &format!("{series_dl}")); + let chart_dl = dl("0%").replace("{}", "0"); + let plot = format!( + r#"{s}{chart_dl}"# + ); + shown(&chart_items(&space(&plot), &theme(), 300.0, 200.0)) + }; + for linked in ["1", "true"] { + let own = run(&dl("General").replace("{}", linked)); + assert!(own.contains(&"3.2".to_string()) && !own.contains(&"320%".to_string()), "sourceLinked={linked}: {own:?}"); + } + // Control: a series without a label format still takes the chart's. + let inherited = run(r#""#); + assert!(inherited.contains(&"320%".to_string()), "{inherited:?}"); +} diff --git a/crates/docx/src/read/diagram.rs b/crates/docx/src/read/diagram.rs index 7f03e8b51..7b7bbeebf 100644 --- a/crates/docx/src/read/diagram.rs +++ b/crates/docx/src/read/diagram.rs @@ -463,7 +463,16 @@ fn text(sp: &El, xf: Xf, b: [f64; 4], theme: &[Rgb], filled: bool, fit: f32, out let mut y = y0 + (th - total) / 2.0; for l in lines { let lh = l.size * LINE; - out.push(GraphicItem::Text { rect: [x0, y, tw, lh], text: l.text, size: l.size, color: l.color, bold: l.bold, align: l.align, font: l.font }); + out.push(GraphicItem::Text { + rect: [x0, y, tw, lh], + text: l.text, + size: l.size, + color: l.color, + bold: l.bold, + align: l.align, + font: l.font, + angle: 0.0, + }); y += lh; } } diff --git a/crates/docx/src/read/props.rs b/crates/docx/src/read/props.rs index 2c8d025b1..dc61b3a28 100644 --- a/crates/docx/src/read/props.rs +++ b/crates/docx/src/read/props.rs @@ -389,15 +389,22 @@ impl PropCtx { } "w:numPr" => { let num = k.child_val("w:numId").and_then(u32_of); - let ilvl = k.child_val("w:ilvl").and_then(u32_of); - let level = ilvl.unwrap_or(0).min(8) as u8; - match (num, ilvl) { - (Some(num), _) => p.numbering = Some(NumRef { num, level }), - // A level without a list: the list comes from the based-on style, as in - // a Heading 3 that only says "level 3" of Heading 2's list. Resolved once - // the styles are read (`resolve_list_levels`). - (None, Some(_)) => p.numbering = Some(NumRef { num: LEVEL_ONLY, level }), - (None, None) => {} + let ilvl_node = k.child("w:ilvl"); + let ilvl = + ilvl_node.and_then(|node| node.attr("w:val")).and_then(int).filter(|level| (0..=8).contains(level)).map(|level| level as u32); + // A missing level means level zero. A present but malformed/out-of-range + // level invalidates this numbering reference instead of silently becoming + // level zero; the latter can turn hostile input into a real list. + if ilvl_node.is_none() || ilvl.is_some() { + let level = ilvl.unwrap_or(0) as u8; + match (num, ilvl) { + (Some(num), _) => p.numbering = Some(NumRef { num, level }), + // A level without a list: the list comes from the based-on style, as in + // a Heading 3 that only says "level 3" of Heading 2's list. Resolved once + // the styles are read (`resolve_list_levels`). + (None, Some(_)) => p.numbering = Some(NumRef { num: LEVEL_ONLY, level }), + (None, None) => {} + } } } "w:pBdr" => p.borders = Some(self.borders(k)), @@ -817,20 +824,21 @@ pub fn sectpr(pc: &PropCtx, e: &El) -> (SectionProps, Vec) { s.footnote_columns = k.attr("w:val").and_then(u32_of).filter(|v| (1..=45).contains(v)); } "w:lnNumType" => { - let mut l = LineNumbering::default(); - if let Some(v) = k.attr("w:countBy").and_then(u32_of) { - l.count_by = v.clamp(1, 100); + // Word leaves line numbering off unless `w:countBy` gives an interval above 0, + // whatever `w:start`, `w:distance` and `w:restart` say (#6209, observed in Word). + if let Some(count_by) = k.attr("w:countBy").and_then(u32_of).filter(|&v| v > 0) { + let mut l = LineNumbering { count_by: count_by.clamp(1, 100), ..LineNumbering::default() }; + if let Some(v) = k.attr("w:start").and_then(int) { + l.start = (v.clamp(0, 32_767) + 1) as u32; + } + l.distance = tw(k, "w:distance").unwrap_or(0.0).max(0.0); + l.restart = match k.attr("w:restart") { + Some("newSection") => LineNumberRestart::Section, + Some("continuous") => LineNumberRestart::Continuous, + _ => LineNumberRestart::Page, + }; + s.line_numbers = Some(l); } - if let Some(v) = k.attr("w:start").and_then(int) { - l.start = (v.clamp(0, 32_767) + 1) as u32; - } - l.distance = tw(k, "w:distance").unwrap_or(0.0).max(0.0); - l.restart = match k.attr("w:restart") { - Some("newSection") => LineNumberRestart::Section, - Some("continuous") => LineNumberRestart::Continuous, - _ => LineNumberRestart::Page, - }; - s.line_numbers = Some(l); } "w:vAlign" => { s.valign = match k.attr("w:val") { diff --git a/crates/docx/src/read/story.rs b/crates/docx/src/read/story.rs index 0550557f1..770375365 100644 --- a/crates/docx/src/read/story.rs +++ b/crates/docx/src/read/story.rs @@ -113,6 +113,8 @@ pub struct StoryCtx { pub story_depth: usize, /// Reading a header or footer (not a text box inside one): where Word keeps its watermark. pub in_hf: bool, + /// Reading a table cell's paragraphs (not a text box inside one). + in_cell: bool, } /// Paragraph content under construction. diff --git a/crates/docx/src/read/story/blocks.rs b/crates/docx/src/read/story/blocks.rs index e72f16bf6..878429454 100644 --- a/crates/docx/src/read/story/blocks.rs +++ b/crates/docx/src/read/story/blocks.rs @@ -249,10 +249,11 @@ impl Reader<'_> { .unwrap_or_default(); para.props.fmt_change = PropChange::boxed(self.revision(RevisionKind::Format, ch), old); } - if para.props.numbering.is_some() - && let Some(ch) = ppr.child("w:numPr").and_then(|n| n.child("w:numberingChange")) - { - let original = ch.attr("w:original").unwrap_or("").chars().take(256).collect(); + if let Some(ch) = ppr.child("w:numPr").and_then(|n| n.child("w:numberingChange")) { + // Multi-level numbering caches (`w:original`) routinely exceed 256 chars while staying + // bounded against hostile attributes. + const MAX_CHANGE_ORIGINAL: usize = 4096; + let original = ch.attr("w:original").unwrap_or("").chars().take(MAX_CHANGE_ORIGINAL).collect(); para.props.num_change = Some(Box::new(NumChange { rev: self.revision(RevisionKind::Format, ch), original })); } if let Some(s) = ppr.child("w:sectPr") { @@ -276,6 +277,10 @@ impl Reader<'_> { if sc.fields.iter().any(|f| f.phase == Phase::Instr) && para.plain_text().is_empty() { para.mark.hidden = Some(true); } + // A cell's paragraphs are marked once their table is read: its style counts. + if !sc.in_cell { + self.doc.styles.mark_runs_as_word_draws_them(&mut para, None); + } para.normalize(); para } @@ -314,15 +319,32 @@ impl Reader<'_> { // ---- tables ---- + /// Word gives a table no cell margins when the file has no default table style and neither the + /// table nor its style sets them; saving may write a default Normal Table, so the zeros are + /// kept as the table's own. shortcut: they stay direct formatting (a table style applied later + /// keeps them), until the model knows the file's default table style. A `w:tblCellMar` that sets + /// only some sides keeps them: only the sides it omits become 0. + fn no_default_margins(&self, props: &mut craft_wordmodel::props::TableProps) { + if self.no_default_table_style + && props.cell_margins.is_none() + && props.style.as_deref().and_then(|id| self.doc.styles.table_style(id)).is_none_or(|s| s.parts.cell_margins.is_none()) + { + props.cell_margins = Some([0.0; 4]); + } + } + fn read_table(&mut self, sc: &mut StoryCtx, t: &El, rels: &Rels, depth: usize) -> Option
{ let mut table = Table::default(); if let Some(p) = t.child("w:tblPr") { table.props = self.pc.tblpr(p); if let Some(ch) = p.child("w:tblPrChange") { - let old = ch.child("w:tblPr").map(|o| self.pc.tblpr(o)).unwrap_or_default(); + let mut old = ch.child("w:tblPr").map(|o| self.pc.tblpr(o)).unwrap_or_default(); + // Rejecting the change brings back these properties: with the same margins. + self.no_default_margins(&mut old); table.props.fmt_change = PropChange::boxed(self.revision(RevisionKind::Format, ch), old); } } + self.no_default_margins(&mut table.props); if let Some(g) = t.child("w:tblGrid") { table.grid = g .children("w:gridCol") @@ -355,8 +377,10 @@ impl Reader<'_> { continue; } let mut blocks = Blocks::new(); + let outer = std::mem::replace(&mut sc.in_cell, true); self.read_blocks(sc, tc, rels, &mut blocks, depth + 4); self.flush_pending(sc, &mut blocks); + sc.in_cell = outer; row.cells.push(Cell { props, blocks, controls: cell_controls }); } if !row.cells.is_empty() { @@ -370,6 +394,20 @@ impl Reader<'_> { let cols = table.cols().clamp(1, MAX_COLS); table.grid = vec![72.0; cols]; } + // The cells' paragraphs, with the run formatting the table style gives each cell (a + // table inside has marked its own). + let chrs = self.doc.styles.table_cells_chr(&table); + for (ri, row) in table.rows.iter_mut().enumerate() { + for (ci, cell) in row.cells.iter_mut().enumerate() { + let chr = chrs.as_ref().and_then(|c| c.get(ri)?.get(ci)); + for b in &mut cell.blocks { + if let Block::Para(p) = Arc::make_mut(b) { + self.doc.styles.mark_runs_as_word_draws_them(p, chr); + p.normalize(); + } + } + } + } Some(table) } } diff --git a/crates/docx/src/read/styles.rs b/crates/docx/src/read/styles.rs index edc17f21f..e12749a87 100644 --- a/crates/docx/src/read/styles.rs +++ b/crates/docx/src/read/styles.rs @@ -1,10 +1,22 @@ use super::{ - AbstractNum, DocxError, El, Level, LevelSuffix, Num, NumFormat, PartFormat, Reader, Style, StyleKind, StyleSheet, TablePart, TableStyleParts, - VAlign, flag, int, props, tw, u32_of, + AbstractNum, DocxError, El, Level, LevelSuffix, Num, NumFormat, PartFormat, Reader, RevisionKind, Style, StyleKind, StyleSheet, TablePart, + TableStyleParts, VAlign, flag, int, props, tw, u32_of, }; use crate::units::tw_i16; +use craft_wordmodel::props::PropChange; use std::sync::Arc; +fn numbering_level_index(value: Option<&str>) -> Option { + match value { + None => Some(0), + Some(value) => int(value).filter(|level| (0..=8).contains(level)).map(|level| level as usize), + } +} + +fn style_is_default(style: &El) -> bool { + style.attr("w:default").is_some_and(|value| !matches!(value.trim(), "0" | "false" | "off" | "False" | "FALSE")) +} + /// Built-in style names Word stores in lower case, and how the UI shows them. pub(crate) const LOWER_NAMES: &[(&str, &str)] = &[ ("caption", "Caption"), @@ -52,6 +64,20 @@ fn builtin_heading_level(name: &str) -> Option { } impl Reader<'_> { + /// Whether a style's kept (last) definition is a default table style. + fn has_default_table_style(&self, root: &El) -> bool { + let mut defaults: Vec = Vec::new(); + for s in root.children("w:style").take(10_000) { + let Some(raw_id) = s.attr("w:styleId").filter(|i| !i.is_empty()) else { continue }; + let id = self.pc.style_id(raw_id); + defaults.retain(|d| *d != id); + if s.attr("w:type") == Some("table") && style_is_default(s) { + defaults.push(id); + } + } + !defaults.is_empty() + } + pub(super) fn read_styles(&mut self, path: &str) -> Result<(), DocxError> { /// Largest `w:latentStyles` kept (Word's has ~380 exceptions, ~40 KB). const MAX_LATENT_STYLES_XML: usize = 1 << 20; @@ -59,9 +85,7 @@ impl Reader<'_> { // The default paragraph style must be called `Normal` in our model. let has_normal = root.children("w:style").any(|s| s.attr("w:styleId") == Some("Normal")); if !has_normal - && let Some(def) = root - .children("w:style") - .find(|s| s.attr("w:type") == Some("paragraph") && s.attr("w:default").is_some_and(|v| v == "1" || v == "true")) + && let Some(def) = root.children("w:style").find(|s| s.attr("w:type") == Some("paragraph") && style_is_default(s)) && let Some(id) = def.attr("w:styleId") { self.pc.alias.insert(id.to_string(), "Normal".into()); @@ -85,6 +109,7 @@ impl Reader<'_> { } } let mut sheet = StyleSheet::empty(); + let mut normal_is_default = None; sheet.default_para_present = root.child("w:docDefaults").and_then(|dd| dd.child("w:pPrDefault")).is_some(); if let Some(dd) = root.child("w:docDefaults") { if let Some(r) = dd.child("w:rPrDefault").and_then(|d| d.child("w:rPr")) { @@ -112,6 +137,13 @@ impl Reader<'_> { "numbering" => StyleKind::Numbering, _ => StyleKind::Paragraph, }; + if kind == StyleKind::Paragraph && id == "Normal" { + // Word treats the built-in Normal style as the paragraph default when the file + // doesn't nominate another style. Keep that fact for an explicitly declared + // Normal; the gallery later synthesizes a visually identical Normal when the + // file did not declare one, and that synthetic style is not source provenance. + normal_is_default = Some(s.attr("w:default").is_none()); + } let name = s.child_val("w:name").map(ui_style_name).unwrap_or_else(|| id.clone()); let r = |n: &str| s.child_val(n).filter(|v| !v.is_empty()).map(|v| self.pc.style_id(v)); let mut st = Style { @@ -127,7 +159,7 @@ impl Reader<'_> { builtin: !s.attr("w:customStyle").is_some_and(|v| matches!(v.trim(), "1" | "true" | "on")), locked: flag(s, "w:locked").unwrap_or(false), // Only where the writer couldn't tell from the id (it flags the English ids). - default: s.attr("w:default").is_some_and(|v| v == "1" || v == "true") + default: style_is_default(s) && !matches!( (kind, id.as_str()), (StyleKind::Paragraph, "Normal") @@ -137,26 +169,50 @@ impl Reader<'_> { ), ..Default::default() }; + let mut old_para = None; + let mut old_chr = None; if let Some(p) = s.child("w:pPr") { st.para = self.pc.ppr(p).0; + if let Some(ch) = p.child("w:pPrChange") { + let old = ch.child("w:pPr").map(|o| self.pc.ppr(o).0.formatting()).unwrap_or_default(); + old_para = Some((self.revision(RevisionKind::Format, ch), old)); + } } // Word gives its built-in heading styles their outline level whether or not the file // writes `w:outlineLvl` (some producers leave it out): "heading 2" is level 2 in the // Navigation pane and the table of contents, whatever its id (#4696). if kind == StyleKind::Paragraph && st.builtin - && st.para.outline_level.is_none() && let Some(level) = s.child_val("w:name").and_then(builtin_heading_level) { - st.para.outline_level = Some(level); + if st.para.outline_level.is_none() { + st.para.outline_level = Some(level); + } + if let Some((_, old)) = &mut old_para + && old.outline_level.is_none() + { + old.outline_level = Some(level); + } } if let Some(rp) = s.child("w:rPr") { st.chr = self.pc.rpr(rp); + if let Some(ch) = rp.child("w:rPrChange") { + let old = ch.child("w:rPr").map(|o| self.pc.rpr(o).formatting()).unwrap_or_default(); + old_chr = Some((self.revision(RevisionKind::Format, ch), old)); + } } if kind == StyleKind::Table { let parts = self.table_style_parts(s, &mut st); st.table = (parts != TableStyleParts::default()).then_some(parts); } + // CT_Style has no `stylePrChange`. Keep its two valid revision carriers separate: a + // foreign document may attribute paragraph and run changes to different authors. + if let Some((rev, old)) = old_para { + st.para_change = PropChange::boxed(rev, old); + } + if let Some((rev, old)) = old_chr { + st.chr_change = PropChange::boxed(rev, old); + } // A style defined twice (some generators write their own Heading 1 after a // placeholder one) is the last definition, whole: Word drops the earlier one rather // than merging the two (observed with synthetic documents). @@ -165,6 +221,7 @@ impl Reader<'_> { None => sheet.styles.push(st), } } + self.no_default_table_style = !self.has_default_table_style(&root); // Styles that give only a list level take the list from the style they are based on // (Heading 3: level 3 of Heading 2's outline list). let fixed: Vec> = sheet @@ -181,6 +238,30 @@ impl Reader<'_> { for (s, n) in sheet.styles.iter_mut().zip(fixed) { s.para.numbering = n; } + if !sheet.styles.iter().any(|style| style.kind == StyleKind::Paragraph && style.default) + && normal_is_default == Some(true) + && let Some(normal) = sheet.styles.iter_mut().find(|style| style.kind == StyleKind::Paragraph && style.id == "Normal") + { + normal.default = true; + } + let old_fixed: Vec>> = sheet + .styles + .iter() + .map(|s| { + s.para_change.as_ref().map(|ch| match ch.old.numbering { + Some(n) if n.num == props::LEVEL_ONLY => { + let base = s.based_on.as_deref().and_then(|b| props::chain_list(&sheet, b)); + base.map(|num| craft_wordmodel::props::NumRef { num, level: n.level }) + } + other => other, + }) + }) + .collect(); + for (s, n) in sheet.styles.iter_mut().zip(old_fixed) { + if let (Some(ch), Some(n)) = (&mut s.para_change, n) { + ch.old.numbering = n; + } + } if sheet.default_para.numbering.is_some_and(|n| n.num == props::LEVEL_ONLY) { sheet.default_para.numbering = None; } @@ -250,11 +331,11 @@ impl Reader<'_> { for a in root.children("w:abstractNum").take(10_000) { let Some(id) = a.attr("w:abstractNumId").and_then(u32_of) else { continue }; let mut levels: Vec = (0..9).map(default_level).collect(); + let mut explicit_levels = [false; 9]; for l in a.children("w:lvl") { - let i = l.attr("w:ilvl").and_then(u32_of).unwrap_or(0) as usize; - if let Some(slot) = levels.get_mut(i) { - *slot = self.level(l, i); - } + let Some(i) = numbering_level_index(l.attr("w:ilvl")) else { continue }; + levels[i] = self.level(l, i); + explicit_levels[i] = true; } if let Some(link) = a.child_val("w:numStyleLink") { links.push((id, link.to_string())); @@ -264,6 +345,7 @@ impl Reader<'_> { id, name: a.child_val("w:name").map(str::to_string), levels, + explicit_levels: Some(explicit_levels), nsid: a.child_val("w:nsid").filter(|v| hex(v)).map(str::to_ascii_uppercase), tmpl: a.child_val("w:tmpl").filter(|v| hex(v)).map(str::to_ascii_uppercase), multi_level_type: a @@ -278,7 +360,9 @@ impl Reader<'_> { let (Some(id), Some(abs)) = (n.attr("w:numId").and_then(u32_of), n.child_val("w:abstractNumId").and_then(u32_of)) else { continue }; let (mut start_overrides, mut level_overrides) = (Vec::new(), Vec::new()); for o in n.children("w:lvlOverride").take(9) { - let lvl = o.attr("w:ilvl").and_then(u32_of).unwrap_or(0).min(8) as u8; + let Some(raw_level) = o.attr("w:ilvl") else { continue }; + let Some(lvl) = numbering_level_index(Some(raw_level)) else { continue }; + let lvl = lvl as u8; if let Some(s) = o.child_val("w:startOverride").and_then(u32_of) { start_overrides.push((lvl, s)); } else if let Some(s) = o.child("w:lvl").and_then(|l| l.child_val("w:start")).and_then(u32_of) { @@ -295,9 +379,13 @@ impl Reader<'_> { for (aid, style) in links { let num = self.doc.styles.get(&self.pc.style_id(&style)).and_then(|s| s.para.numbering).map(|n| n.num); let src = num.and_then(|n| nums.iter().find(|x: &&Num| x.id == n)).map(|x| x.abstract_id); - let levels = src.filter(|s| *s != aid).and_then(|s| abstracts.iter().find(|a: &&AbstractNum| a.id == s)).map(|a| a.levels.clone()); - if let (Some(levels), Some(a)) = (levels, abstracts.iter_mut().find(|a| a.id == aid)) { + let source = src + .filter(|s| *s != aid) + .and_then(|s| abstracts.iter().find(|a: &&AbstractNum| a.id == s)) + .map(|a| (a.levels.clone(), a.explicit_levels)); + if let (Some((levels, explicit_levels)), Some(a)) = (source, abstracts.iter_mut().find(|a| a.id == aid)) { a.levels = levels; + a.explicit_levels = explicit_levels; } } Arc::make_mut(&mut self.doc.numbering).abstracts = abstracts; diff --git a/crates/docx/src/units.rs b/crates/docx/src/units.rs index 74edfb13c..295fe7621 100644 --- a/crates/docx/src/units.rs +++ b/crates/docx/src/units.rs @@ -56,8 +56,12 @@ pub fn measure(s: &str, per_pt: f32) -> Option { "emu" => v / 12_700.0, _ => return None, }; - let pt = pt as f32; - pt.is_finite().then_some(pt.clamp(-MAX_LEN_PT, MAX_LEN_PT)) + // Clamp while still in f64: a large but finite OOXML measurement can overflow f32 before + // the output limit is applied, even though its bounded point value is representable. + if pt.is_nan() { + return None; + } + Some(pt.clamp(-f64::from(MAX_LEN_PT), f64::from(MAX_LEN_PT)) as f32) } /// `w:line` under `lineRule="auto"`: the multiple of single line spacing, in 240ths (240 = @@ -196,6 +200,8 @@ mod tests { assert_eq!(measure("nan", 20.0), None); assert_eq!(measure("1e400", 20.0), None); assert_eq!(measure("99999999999999", 20.0), Some(MAX_LEN_PT)); + assert_eq!(measure("1e300pt", 20.0), Some(MAX_LEN_PT)); + assert_eq!(measure("-1e300pt", 20.0), Some(-MAX_LEN_PT)); assert_eq!(measure("x", 20.0), None); // `w:line` auto: the multiple in 240ths without a unit, through twips with one (#2329). assert_eq!(line_multiple("240"), Some(1.0)); diff --git a/crates/docx/src/write/embed.rs b/crates/docx/src/write/embed.rs index 9a8051e09..7673a6f26 100644 --- a/crates/docx/src/write/embed.rs +++ b/crates/docx/src/write/embed.rs @@ -884,10 +884,11 @@ mod tests { let spec = ChartSpec { kind: ChartType::Column, categories: vec!["a".into(), "b".into()], - series: (0..series).map(|i| ChartSeries { name: format!("S{i}"), values: vec![Some(1.0); 2] }).collect(), + series: (0..series).map(|i| ChartSeries { name: format!("S{i}"), values: vec![Some(1.0); 2], ..Default::default() }).collect(), title: None, legend: None, data_labels: false, + ..Default::default() }; let graphic = Graphic { kind: GraphicKind::Chart, w: 100.0, h: 100.0, chart: Some(Arc::new(spec)), ..Default::default() }; InlineObject::Graphic { w: 100.0, h: 100.0, alt: String::new(), float: Default::default(), graphic: Arc::new(graphic) } diff --git a/crates/docx/src/write/glossary.rs b/crates/docx/src/write/glossary.rs index 35a2289bc..7d5a01791 100644 --- a/crates/docx/src/write/glossary.rs +++ b/crates/docx/src/write/glossary.rs @@ -87,7 +87,7 @@ impl Writer<'_> { } rels.add(rt::GLOSSARY, &file(part), false); push_part(entries, overrides, part, w.into_bytes(), GLOSSARY_CONTENT_TYPE, prels); - push_part(entries, overrides, styles, styles_xml(doc), &format!("{CT_WML}styles+xml"), PartRels::default()); + push_part(entries, overrides, styles, styles_xml(self), &format!("{CT_WML}styles+xml"), PartRels::default()); if lists { push_part(entries, overrides, numbering, numbering_xml(doc), &format!("{CT_WML}numbering+xml"), PartRels::default()); } diff --git a/crates/docx/src/write/parts.rs b/crates/docx/src/write/parts.rs index f1b11e8ff..4cb847276 100644 --- a/crates/docx/src/write/parts.rs +++ b/crates/docx/src/write/parts.rs @@ -126,7 +126,7 @@ pub(crate) fn write_parts(doc: &Document, flavor: Flavor) -> Result = a.iter().map(|(k, v)| (*k, v.as_str())).collect(); w.empty("w:numberingChange", &refs); diff --git a/crates/docx/src/write/story/para.rs b/crates/docx/src/write/story/para.rs index ede79170c..e0242d8eb 100644 --- a/crates/docx/src/write/story/para.rs +++ b/crates/docx/src/write/story/para.rs @@ -294,7 +294,7 @@ impl Writer<'_> { } /// `w:id`, `w:author` and `w:date` of a tracked change by revision `idx`. - pub(super) fn change_attrs(&mut self, idx: u32) -> ChangeAttrs { + pub(crate) fn change_attrs(&mut self, idx: u32) -> ChangeAttrs { let (id, author, date) = self.rev_attrs(idx); let mut a = vec![("w:id", id), ("w:author", author)]; if !date.is_empty() { diff --git a/crates/docx/src/write/styles.rs b/crates/docx/src/write/styles.rs index c4013ebff..3bcf9ab97 100644 --- a/crates/docx/src/write/styles.rs +++ b/crates/docx/src/write/styles.rs @@ -1,4 +1,6 @@ -use super::{Document, Level, LevelSuffix, ParaProps, Style, StyleKind, TablePart, VAlign, W, props, xml}; +use super::{Level, LevelSuffix, ParaProps, Style, StyleKind, TablePart, VAlign, W, Writer, props, xml}; +use crate::write::props::open_change; +use craft_wordmodel::Document; /// Built-in names Word keeps in lower case in styles.xml. fn file_style_name(name: &str) -> String { @@ -16,7 +18,8 @@ fn file_style_name(name: &str) -> String { name.to_string() } -pub(super) fn styles_xml(doc: &Document) -> Vec { +pub(super) fn styles_xml(wr: &mut Writer<'_>) -> Vec { + let doc = wr.doc; let s = &doc.styles; let mut w = W::new(); let ns = xml::body_ns(); @@ -50,13 +53,15 @@ pub(super) fn styles_xml(doc: &Document) -> Vec { // Kinds whose default style the file named (`w:default`); the English ids stand in otherwise. let flagged: Vec = s.styles.iter().filter(|st| st.default).map(|st| st.kind).collect(); for st in &s.styles { - style_xml(&mut w, st, &flagged); + style_xml(wr, &mut w, st, &flagged); } w.close("w:styles"); w.into_bytes() } -fn style_xml(w: &mut W, st: &Style, flagged: &[StyleKind]) { +/// One `w:style` element. A tracked change of its paragraph/run properties is nested in the +/// corresponding property element as `w:pPrChange` / `w:rPrChange`. +fn style_xml(wr: &mut Writer<'_>, w: &mut W, st: &Style, flagged: &[StyleKind]) { let ty = match st.kind { StyleKind::Paragraph => "paragraph", StyleKind::Character => "character", @@ -70,9 +75,14 @@ fn style_xml(w: &mut W, st: &Style, flagged: &[StyleKind]) { | (StyleKind::Table, "TableNormal") | (StyleKind::Numbering, "NoList") ); - let is_default = st.default || (english_default && !flagged.contains(&st.kind)); + let normal_is_explicitly_off = st.kind == StyleKind::Paragraph && st.id == "Normal" && !st.default && flagged.contains(&StyleKind::Paragraph); + let is_default = st.default || (english_default && !flagged.contains(&st.kind) && !normal_is_explicitly_off); let mut a = vec![("w:type", ty), ("w:styleId", st.id.as_str())]; - if is_default { + if normal_is_explicitly_off { + // Preserve a model where Normal is explicitly not default. Without this, the fallback + // `english_default` marker makes the reader infer a change on round trip. + a.push(("w:default", "0")); + } else if is_default { a.push(("w:default", "1")); } if !st.builtin { @@ -102,16 +112,7 @@ fn style_xml(w: &mut W, st: &Style, flagged: &[StyleKind]) { if st.locked { w.empty("w:locked", &[]); } - if !st.para.is_empty() && st.kind != StyleKind::Character { - w.open("w:pPr", &[]); - props::ppr_inner(w, &st.para, false, None); - w.close("w:pPr"); - } - if props::has_rpr(&st.chr) && st.kind != StyleKind::Numbering { - w.open("w:rPr", &[]); - props::rpr_inner(w, &st.chr); - w.close("w:rPr"); - } + style_property_xml(wr, w, st); if let Some(t) = &st.table { if t.borders.is_some() || t.cell_margins.is_some() || t.band_size.is_some() || t.col_band_size.is_some() { w.open("w:tblPr", &[]); @@ -179,6 +180,37 @@ fn style_xml(w: &mut W, st: &Style, flagged: &[StyleKind]) { w.close("w:style"); } +fn style_property_xml(wr: &mut Writer<'_>, w: &mut W, st: &Style) { + let para_change = st.para_change.as_deref().filter(|ch| ch.old.formatting() != st.para.formatting()); + if (!st.para.is_empty() || para_change.is_some()) && st.kind != StyleKind::Character { + w.open("w:pPr", &[]); + props::ppr_inner(w, &st.para, false, None); + if let Some(ch) = para_change { + let a = wr.change_attrs(ch.rev); + open_change(w, "w:pPrChange", &a); + w.open("w:pPr", &[]); + props::ppr_inner(w, &ch.old, false, None); + w.close("w:pPr"); + w.close("w:pPrChange"); + } + w.close("w:pPr"); + } + let run_change = st.chr_change.as_deref().filter(|ch| ch.old.formatting() != st.chr.formatting()); + if (props::has_rpr(&st.chr) || run_change.is_some()) && st.kind != StyleKind::Numbering { + w.open("w:rPr", &[]); + props::rpr_inner(w, &st.chr); + if let Some(ch) = run_change { + let a = wr.change_attrs(ch.rev); + open_change(w, "w:rPrChange", &a); + w.open("w:rPr", &[]); + props::rpr_inner(w, &ch.old); + w.close("w:rPr"); + w.close("w:rPrChange"); + } + w.close("w:rPr"); + } +} + pub(super) fn numbering_xml(doc: &Document) -> Vec { let mut w = W::new(); let ns = xml::body_ns(); diff --git a/crates/docx/src/xml.rs b/crates/docx/src/xml.rs index ff953f43d..bb86755d9 100644 --- a/crates/docx/src/xml.rs +++ b/crates/docx/src/xml.rs @@ -373,6 +373,7 @@ fn parse_with(bytes: &[u8], extra: &[(&'static str, &str)], strict: bool) -> Res max_depth: MAX_DEPTH + 1, // Text stretches count as nodes too: two for each element. max_nodes: 2 * MAX_ELEMENTS, + max_elements: MAX_ELEMENTS, max_attrs: MAX_ATTRS, max_attrs_total: 2 * MAX_ELEMENTS, }; diff --git a/crates/docx/tests/it/bidi.rs b/crates/docx/tests/it/bidi.rs index 3df1c04c9..bea7e2131 100644 --- a/crates/docx/tests/it/bidi.rs +++ b/crates/docx/tests/it/bidi.rs @@ -175,3 +175,155 @@ fn rtl_tables_and_sections_round_trip() { assert_eq!(again, vec![true, false]); assert!(r.last_section.rtl && r.last_section.columns.count == 2); } + +/// A package whose body is `body` and whose styles part holds `styles`. +fn docx_with_styles(body: &str, styles: &str) -> Vec { + let ns = r#"xmlns:w="http://schemas.openxmlformats.org/wordprocessingml/2006/main""#; + let root = r#""#; + let rels = r#""#; + let doc = format!(r#"{body}"#); + let st = format!(r#"{styles}"#); + zip(&[ + ("_rels/.rels", root.as_bytes()), + ("word/_rels/document.xml.rels", rels.as_bytes()), + ("word/document.xml", doc.as_bytes()), + ("word/styles.xml", st.as_bytes()), + ]) +} + +/// Thai text draws in the complex-script formatting only where Word draws it so (measured in Word +/// for Windows; run defaults Angsana New 22 pt, complex script Tahoma 11 pt bold). Without +/// `w:cs` or `w:rtl` on the run, its paragraph style or the run defaults, Thai text measures as +/// Angsana New 22 pt, whatever its language, `w:hint` or character style. +#[test] +fn thai_text_takes_the_complex_script_formatting_only_in_a_complex_script_run() { + let thai = "ทดสอà¸� ภาษาไทย"; + let defaults = |extra: &str| { + format!( + r#"{extra} + + + + + + +"# + ) + }; + let latin = ("Angsana New", 22.0, false); + let complex = ("Tahoma", 11.0, true); + let run = |rpr: &str| format!(r#"{rpr}{thai}"#); + let styled = |p: &str, rpr: &str| { + format!(r#"{rpr}{thai}"#) + }; + let in_table = |style: &str| { + format!( + r#"{}"#, + run("") + ) + }; + let cs_off = r#""#; + let cases: Vec<(&str, String, String, (&str, f32, bool))> = vec![ + ("plain", defaults(""), run(""), latin), + ("bidi language", defaults(""), run(r#""#), latin), + ("hint", defaults(""), run(r#""#), latin), + ("character style", defaults(""), run(r#""#), latin), + ("character style rtl", defaults(""), run(r#""#), latin), + ("run cs", defaults(""), run(""), complex), + ("run rtl", defaults(""), run(""), complex), + ("paragraph style", defaults(""), styled("CsPara", ""), complex), + ("paragraph style, run cs off", defaults(""), styled("CsPara", r#""#), latin), + ("run defaults", defaults(""), run(""), complex), + // A character style's `w:cs w:val="0"` doesn't turn the paragraph style's or the run + // defaults' off either. + ("paragraph style, character style cs off", defaults(""), styled("CsPara", cs_off), complex), + ("run defaults, character style cs off", defaults(""), run(cs_off), complex), + // A table style's `w:cs` counts, and its sizes are the run's. + ("table style", defaults(""), in_table("CsTable"), complex), + ("table style sizes", defaults(""), in_table("SizeTable"), ("Angsana New", 15.0, false)), + ]; + let check = |d: &Document, name: &str, want: (&str, f32, bool), when: &str| { + let (p, table) = match &*d.body[0] { + craft_wordmodel::Block::Table(t) => { + (t.rows[0].cells[0].blocks[0].as_para().unwrap(), d.styles.table_cells_chr(t).map(|c| c[0][0].clone())) + } + b => (b.as_para().unwrap(), None), + }; + let rc = d.styles.resolve_char_in(p.props.style.as_deref(), table.as_ref(), &p.runs[0].props); + for ch in p.text.chars() { + let drawn = if rc.uses_complex(ch) { rc.complex() } else { rc.clone() }; + assert_eq!((drawn.font.as_str(), drawn.size, drawn.bold), want, "{name} {when}: {ch:?}"); + } + let c = &p.runs[0].props; + assert_eq!((c.font_cs.as_deref(), c.size_cs, c.bold_cs), (None, None, None), "{name} {when}: nothing is pinned"); + }; + for (name, styles, body, want) in &cases { + let d = craft_docx::read(&docx_with_styles(body, styles)).unwrap(); + check(&d, name, *want, "read"); + let again = craft_docx::read(&craft_docx::write(&d).unwrap()).unwrap(); + check(&again, name, *want, "after a save"); + } + // Latin text and right-to-left text are left as they were. + let body = r#"Latin onlyسلام"#; + let d = craft_docx::read(&docx_with_styles(body, &defaults(""))).unwrap(); + for p in paras(&d) { + assert_eq!(p.runs[0].props.cs, None, "{}", p.text); + } + // Latin text follows the same rule (measured): a character style's `w:cs` or `w:rtl` leaves it in the + // run's own formatting, a character style's `w:cs w:val="0"` under a paragraph style's + // `w:cs` in the complex-script formatting. + let latin_in = + |p: &str, rpr: &str| format!(r#"{rpr}Latin"#); + let body = + latin_in("Normal", r#""#) + &latin_in("CsPara", cs_off) + &latin_in("Normal", r#""#); + let d = craft_docx::read(&docx_with_styles(&body, &defaults(""))).unwrap(); + let again = craft_docx::read(&craft_docx::write(&d).unwrap()).unwrap(); + for (d, when) in [(&d, "read"), (&again, "after a save")] { + let drawn: Vec<_> = paras(d) + .iter() + .map(|p| { + let rc = d.styles.resolve_char(p.props.style.as_deref(), &p.runs[0].props); + let c = if rc.uses_complex('L') { rc.complex() } else { rc }; + (c.font, c.size) + }) + .collect(); + assert_eq!(drawn, [("Angsana New".to_string(), 22.0), ("Tahoma".to_string(), 11.0), ("Angsana New".to_string(), 22.0)], "{when}"); + } +} + +/// Measured in Word for Windows: Thai text as a field's result follows the same rule as ordinary +/// text, also when the reader keeps the field as one object with its result. +#[test] +fn a_fields_thai_result_takes_the_complex_script_formatting_only_in_a_complex_script_run() { + let styles = r#" +"#; + let field = |rpr: &str| { + let r = |inner: &str| format!("{rpr}{inner}"); + format!( + "{}{}{}{}{}", + r(r#""#), + r(r#" QUOTE "x" "#), + r(r#""#), + r("ทดสอà¸�"), + r(r#""#) + ) + }; + let drawn = |d: &Document| { + let p = paras(d)[0]; + let at = p.text.find(craft_wordmodel::para::OBJ).expect("the field is one object"); + let run = p.runs.iter().scan(0, |end, r| { + let start = *end; + *end += r.len; + Some((start, r)) + }); + let (_, run) = run.into_iter().find(|(start, r)| *start <= at && at < start + r.len).unwrap(); + let rc = d.styles.resolve_char(p.props.style.as_deref(), &run.props); + if rc.uses_complex('ท') { (rc.font_cs.clone(), rc.size_cs) } else { (rc.font.clone(), rc.size) } + }; + for (what, rpr, want) in [("nothing", "", ("Angsana New", 10.0)), ("w:cs", "", ("Tahoma", 20.0))] { + let d = craft_docx::read(&docx_with_styles(&field(rpr), styles)).unwrap(); + assert_eq!(drawn(&d), (want.0.to_string(), want.1), "a run with {what}: read"); + let again = craft_docx::read(&craft_docx::write(&d).unwrap()).unwrap(); + assert_eq!(drawn(&again), (want.0.to_string(), want.1), "a run with {what}: after a save"); + } +} diff --git a/crates/docx/tests/it/cell_margin_inheritance.rs b/crates/docx/tests/it/cell_margin_inheritance.rs index f4d8de5a5..83aafc7d0 100644 --- a/crates/docx/tests/it/cell_margin_inheritance.rs +++ b/crates/docx/tests/it/cell_margin_inheritance.rs @@ -28,3 +28,73 @@ fn partial_margin_stays_partial() { assert!(!margins.contains(") -> Vec { + const W: &str = "http://schemas.openxmlformats.org/wordprocessingml/2006/main"; + let mut zip = zip::ZipWriter::new(Cursor::new(Vec::new())); + let mut add = |name: &str, xml: &str| { + zip.start_file(name, zip::write::SimpleFileOptions::default()).unwrap(); + zip.write_all(xml.as_bytes()).unwrap(); + }; + add("word/document.xml", &format!(r#"{body}"#)); + if let Some(styles) = styles { + add("word/styles.xml", &format!(r#"{styles}"#)); + add( + "word/_rels/document.xml.rels", + r#""#, + ); + } + zip.finish().unwrap().into_inner() +} + +/// Without a default table style (no styles part, or none marked `w:default`) Word gives a table +/// no cell margins: a cell that sets only its right margin has none on its other sides. +#[test] +fn no_default_table_style_means_no_cell_margins() { + let body = r#""#; + let normal_table = r#""#; + let margins = |styles: Option<&str>| { + let doc = craft_docx::read(&package(body, styles)).unwrap(); + let table = doc.body[0].as_table().unwrap(); + (table.props.cell_margins, table.rows[0].cells[0].props.resolved_margins(table.props.cell_margins.unwrap_or([0.0, 5.4, 0.0, 5.4]))) + }; + assert_eq!(margins(None), (Some([0.0; 4]), [0.0, 0.0, 0.0, 5.4])); + assert_eq!(margins(Some(r#""#)), (Some([0.0; 4]), [0.0, 0.0, 0.0, 5.4])); + // With Word's usual Normal Table the table keeps no margins of its own (the defaults apply). + assert_eq!(margins(Some(normal_table)).0, None); + // A style defined twice is its last definition: a default one redefined as not default leaves none. + let redefined = format!(r#"{normal_table}"#); + assert_eq!(margins(Some(&redefined)).0, Some([0.0; 4])); + // A custom default table style marked `on` is a default too, also after saving and reopening. + let plain = r#""#; + let doc = craft_docx::read(&package(body, Some(plain))).unwrap(); + assert_eq!(doc.body[0].as_table().unwrap().props.cell_margins, None); + let reopened = craft_docx::read(&craft_docx::write(&doc).unwrap()).unwrap(); + assert_eq!(reopened.body[0].as_table().unwrap().props.cell_margins, None); + // A table that sets only some sides keeps them; the sides it omits are 0. + let partial = body.replace("", r#""#); + let doc = craft_docx::read(&package(&partial, None)).unwrap(); + assert_eq!(doc.body[0].as_table().unwrap().props.cell_margins, Some([0.0, 7.2, 0.0, 0.0])); + // Rejecting a tracked change to the table's properties brings back the same zero margins. + let tracked = body.replace( + "", + r#""#, + ); + let doc = craft_docx::read(&package(&tracked, None)).unwrap(); + let change = doc.body[0].as_table().unwrap().props.fmt_change.as_deref().expect("a tracked table change"); + assert_eq!(change.old.cell_margins, Some([0.0; 4])); + // A table whose own style gives margins keeps them. + let styled = body.replace("", r#""#); + let boxed = r#""#; + let doc = craft_docx::read(&package(&styled, Some(boxed))).unwrap(); + assert_eq!(doc.body[0].as_table().unwrap().props.cell_margins, None); + // The zeros survive saving, which writes a default Normal Table. + let doc = craft_docx::read(&package(body, None)).unwrap(); + let bytes = craft_docx::write(&doc).unwrap(); + let mut xml = String::new(); + zip::ZipArchive::new(Cursor::new(bytes.clone())).unwrap().by_name("word/document.xml").unwrap().read_to_string(&mut xml).unwrap(); + assert!(xml.contains(""), "{xml}"); + let reopened = craft_docx::read(&bytes).unwrap(); + assert_eq!(reopened.body[0].as_table().unwrap().props.cell_margins, Some([0.0; 4])); +} diff --git a/crates/docx/tests/it/fixtures.rs b/crates/docx/tests/it/fixtures.rs index 868e8cbb0..348d6239e 100644 --- a/crates/docx/tests/it/fixtures.rs +++ b/crates/docx/tests/it/fixtures.rs @@ -2197,6 +2197,50 @@ fn formatting_revisions_round_trip() { assert_eq!(again.revisions, d.revisions); } +#[test] +fn numbering_change_without_active_list_roundtrips() { + let body = r#"Removed"#; + let d = read_body(body); + let p = paras(&d); + assert_eq!(p.len(), 1); + assert!(p[0].props.numbering.is_none()); + let nc = p[0].props.num_change.as_deref().expect("numberingChange read back"); + assert_eq!(nc.original, "%1:1:0:."); + + let bytes = craft_docx::write(&d).unwrap(); + let mut zip = zip::ZipArchive::new(std::io::Cursor::new(&bytes)).unwrap(); + let mut xml = String::new(); + std::io::Read::read_to_string(&mut zip.by_name("word/document.xml").unwrap(), &mut xml).unwrap(); + assert!(xml.contains(r#""#), "{xml}"); + + let again = craft_docx::read(&bytes).unwrap(); + let ap = paras(&again); + assert!(ap[0].props.numbering.is_none()); + let anc = ap[0].props.num_change.as_deref().expect("numberingChange preserved after save"); + assert_eq!(anc.original, "%1:1:0:."); +} + +#[test] +fn numbering_change_original_preserves_long_values_within_bound() { + let long_orig = "%1:1:0:.".repeat(64); // 512 chars, well past the old 256-char cap + assert!(long_orig.len() > 256 && long_orig.len() < 4096); + let over_cap = "x".repeat(5000); + let body = format!( + r#"Long +Bounded"# + ); + let d = read_body(&body); + let p = paras(&d); + assert_eq!(p[0].props.num_change.as_deref().map(|n| n.original.as_str()), Some(long_orig.as_str())); + assert_eq!(p[1].props.num_change.as_deref().map(|n| n.original.chars().count()), Some(4096)); + + let bytes = craft_docx::write(&d).unwrap(); + let again = craft_docx::read(&bytes).unwrap(); + let ap = paras(&again); + assert_eq!(ap[0].props.num_change.as_deref().map(|n| n.original.as_str()), Some(long_orig.as_str())); +} + /// A package with `styles`, a theme (accent1 4472C4, accent2 ED7D31, major font "Major Face") and, /// when given, `settings`. fn themed_docx(body: &str, styles: &str, settings: Option<&str>) -> Document { diff --git a/crates/docx/tests/it/note_comments.rs b/crates/docx/tests/it/note_comments.rs index d333c6adb..749f65897 100644 --- a/crates/docx/tests/it/note_comments.rs +++ b/crates/docx/tests/it/note_comments.rs @@ -17,6 +17,10 @@ fn note(tag: &str, mark: &str, comment: u32, text: &str) -> String { } fn read_doc() -> Document { + read_doc_with_note_relationships(true) +} + +fn read_doc_with_note_relationships(with_note_relationships: bool) -> Document { let doc = format!( r#"Body"# ); @@ -25,10 +29,15 @@ fn read_doc() -> Document { let comments = format!( r#"On the footnoteOn the endnote"# ); + let doc_rels = if with_note_relationships { + DOC_RELS + } else { + r#""# + }; let bytes = zip(&[ ("_rels/.rels", ROOT_RELS.as_bytes()), ("word/document.xml", doc.as_bytes()), - ("word/_rels/document.xml.rels", DOC_RELS.as_bytes()), + ("word/_rels/document.xml.rels", doc_rels.as_bytes()), ("word/footnotes.xml", footnotes.as_bytes()), ("word/endnotes.xml", endnotes.as_bytes()), ("word/comments.xml", comments.as_bytes()), @@ -36,6 +45,26 @@ fn read_doc() -> Document { craft_docx::read(&bytes).unwrap() } +#[test] +fn conventional_note_parts_are_read_without_relationships() { + let d = read_doc_with_note_relationships(false); + let body = d.body[0].as_para().unwrap(); + let refs: Vec<(NoteKind, u32)> = body + .objects + .iter() + .filter_map(|o| match o { + InlineObject::NoteRef { kind, id, .. } => Some((*kind, *id)), + _ => None, + }) + .collect(); + assert_eq!(refs.iter().map(|(kind, _)| *kind).collect::>(), [NoteKind::Footnote, NoteKind::Endnote]); + for ((kind, id), wanted) in refs.into_iter().zip(["Foot.", "End."]) { + let part = d.parts.get(&id).unwrap_or_else(|| panic!("missing {kind:?} part {id}")); + let text = part.blocks[0].as_para().unwrap().plain_text(); + assert!(text.contains(wanted), "{kind:?}: {text:?}"); + } +} + /// comments.xml has to be read before the notes: a note's `w:commentRangeStart` resolves its id /// against the comments already read, and was dropped when the notes came first. #[test] diff --git a/crates/docx/tests/it/passthrough.rs b/crates/docx/tests/it/passthrough.rs index f59131e17..9d3176a76 100644 --- a/crates/docx/tests/it/passthrough.rs +++ b/crates/docx/tests/it/passthrough.rs @@ -320,6 +320,34 @@ fn list_identities_survive_and_a_copied_definition_gets_none() { assert_eq!(n.matches("1F2E3D4C").count(), 1, "{n}"); } +#[test] +fn malformed_numbering_levels_are_skipped_instead_of_becoming_level_zero() { + let numbering = r#""#; + let body = r#"default levelvalid levelnegative levelout of rangeoverflow"#; + let bytes = package(body, &[("rId1", &format!("{REL}/numbering"), "numbering.xml")], &[], &[("word/numbering.xml", numbering)]); + let doc = craft_docx::read(&bytes).unwrap(); + let paragraphs: Vec<_> = doc + .body + .iter() + .filter_map(|block| match &**block { + craft_wordmodel::Block::Para(p) => Some(p), + _ => None, + }) + .collect(); + assert_eq!(paragraphs.len(), 5); + assert!(doc.numbering.level_is_defined(1, 0)); + assert!(!doc.numbering.level_is_defined(1, 2), "effective defaults remain distinguishable from source definitions"); + assert!(doc.numbering.level(1, 2).is_some(), "Word's effective default level remains available"); + assert_eq!(paragraphs[0].props.numbering.map(|n| (n.num, n.level)), Some((1, 0)), "an omitted level defaults to zero"); + assert_eq!(paragraphs[1].props.numbering.map(|n| (n.num, n.level)), Some((1, 2)), "a valid level is preserved"); + for p in ¶graphs[2..] { + assert_eq!(p.props.numbering, None, "malformed explicit level is not treated as a real list: {}", p.text); + } + let level_zero = doc.numbering.level(1, 0).expect("the valid level zero"); + assert_eq!(level_zero.format, craft_wordmodel::section::NumFormat::Decimal); + assert_eq!(level_zero.start, 1, "invalid level overrides do not affect level zero"); +} + #[test] fn extended_properties_are_kept_without_words_counts_and_personal_ones_follow_the_privacy_setting() { let app = r#"4210740000250000Microsoft Office Word020001700falseMFirmafalse290000falsehttps://example.com/false16.0000"#; diff --git a/crates/docx/tests/it/roundtrip.rs b/crates/docx/tests/it/roundtrip.rs index 0ee245180..8f3bbdd0b 100644 --- a/crates/docx/tests/it/roundtrip.rs +++ b/crates/docx/tests/it/roundtrip.rs @@ -269,6 +269,7 @@ fn styles_and_lists_round_trip() { locked: true, default: false, table: None, + ..Default::default() }); Arc::make_mut(&mut d.styles).upsert(Style { id: "MyStyleChar".into(), diff --git a/crates/docx/tests/it/undefined_style.rs b/crates/docx/tests/it/undefined_style.rs index c50394a2a..a00ac4323 100644 --- a/crates/docx/tests/it/undefined_style.rs +++ b/crates/docx/tests/it/undefined_style.rs @@ -3,12 +3,20 @@ //! `` without defining Footer. use craft_wordmodel::props::LineSpacing; use craft_wordmodel::{Block, Document, PartKind}; -use std::io::Write; +use std::io::{Read, Write}; const W: &str = "http://schemas.openxmlformats.org/wordprocessingml/2006/main"; const R: &str = "http://schemas.openxmlformats.org/officeDocument/2006/relationships"; fn package(normal_id: &str) -> Vec { + package_with_default(normal_id, true) +} + +fn package_with_default(normal_id: &str, is_default: bool) -> Vec { + package_with_body_default(normal_id, is_default, false) +} + +fn package_with_body_default(normal_id: &str, is_default: bool, has_body_default: bool) -> Vec { let root = r#""#; let doc = format!( r#"bodynormal"# @@ -16,8 +24,11 @@ fn package(normal_id: &str) -> Vec { let footer = format!( r#"1"# ); + let default_attr = if is_default { r#" w:default="1""# } else { "" }; + let body_style = + if has_body_default { r#""# } else { "" }; let styles = format!( - r#""# + r#"{body_style}"# ); let rels = r#""#; let mut zw = zip::ZipWriter::new(std::io::Cursor::new(Vec::new())); @@ -34,6 +45,27 @@ fn package(normal_id: &str) -> Vec { zw.finish().unwrap().into_inner() } +#[test] +fn an_explicit_normal_style_is_the_default_even_without_w_default() { + let doc = craft_docx::read(&package_with_default("Normal", false)).unwrap(); + let normal = doc.styles.get("Normal").expect("the explicit Normal style"); + assert!(normal.default, "Word treats the explicitly declared built-in Normal style as the default"); +} + +#[test] +fn an_explicit_body_default_on_prevents_normal_inference() { + let doc = craft_docx::read(&package_with_body_default("Normal", false, true)).unwrap(); + assert!(doc.styles.styles.iter().any(|style| style.id == "Body" && style.default)); + assert!(!doc.styles.styles.iter().any(|style| style.id == "Normal" && style.default)); + let saved = craft_docx::write(&doc).unwrap(); + let mut archive = zip::ZipArchive::new(std::io::Cursor::new(saved)).unwrap(); + let mut file = archive.by_name("word/styles.xml").unwrap(); + let mut styles = String::new(); + file.read_to_string(&mut styles).unwrap(); + assert!(styles.contains(r#"w:styleId="Normal" w:default="0""#) || styles.contains(r#"w:default="0" w:styleId="Normal""#)); + assert!(styles.contains(r#"w:styleId="Body" w:default="1""#) || styles.contains(r#"w:default="1" w:styleId="Body""#)); +} + fn footer_para(d: &Document) -> &craft_wordmodel::Paragraph { let part = d.parts.values().find(|p| p.kind == PartKind::Footer).expect("a footer"); match part.blocks.first().map(|b| &**b) { @@ -69,3 +101,37 @@ fn an_undefined_style_is_written_back_as_it_was_named() { assert_eq!(footer_para(&again).props.style.as_deref(), Some("Footer")); assert_eq!(again.styles.resolve_para(&footer_para(&again).props).space_after, 2.0); } + +#[test] +fn conventional_styles_and_numbering_parts_without_relationships_are_read() { + // The old DesignCraft reader opened these conventional paths directly. Keep that tolerated + // input behavior while preferring OPC relationships whenever they resolve successfully. + let doc = format!( + r#"text"# + ); + let styles = format!(r#""#); + let numbering = format!( + r#""# + ); + for main_path in ["word/document.xml", "WORD/DOCUMENT.XML"] { + let root = format!( + r#""# + ); + let mut zw = zip::ZipWriter::new(std::io::Cursor::new(Vec::new())); + for (name, body) in [ + ("_rels/.rels", root.as_str()), + (main_path, doc.as_str()), + ("word/styles.xml", styles.as_str()), + ("word/numbering.xml", numbering.as_str()), + ] { + zw.start_file(name, zip::write::SimpleFileOptions::default()).unwrap(); + zw.write_all(body.as_bytes()).unwrap(); + } + let d = craft_docx::read(&zw.finish().unwrap().into_inner()).unwrap(); + assert!(d.styles.styles.iter().any(|style| style.id == "Body")); + let Some(Block::Para(p)) = d.body.first().map(|b| &**b) else { panic!("a body paragraph") }; + assert_eq!(p.props.style.as_deref(), Some("Body")); + assert_eq!(p.props.numbering.map(|n| (n.num, n.level)), Some((7, 0))); + assert!(d.numbering.level(7, 0).is_some()); + } +} diff --git a/crates/dpx/src/lib.rs b/crates/dpx/src/lib.rs index a762566f8..26d9e8656 100644 --- a/crates/dpx/src/lib.rs +++ b/crates/dpx/src/lib.rs @@ -34,6 +34,9 @@ //! Input is treated as hostile: every header read is bounds-checked, the picture size is capped //! by [`Limits`] and the image data must be present in full before anything is allocated. +/// RGBA8-to-10-bit DPX encoding. +pub mod rgba8_dpx10; + #[cfg(test)] mod tests; diff --git a/crates/dpx/src/rgba8_dpx10.rs b/crates/dpx/src/rgba8_dpx10.rs new file mode 100644 index 000000000..8dec304ec --- /dev/null +++ b/crates/dpx/src/rgba8_dpx10.rs @@ -0,0 +1,122 @@ +//! Uncompressed RGB(A) output from top-down RGBA8 pixels. +/// Why the supplied frame cannot be encoded. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum Rgba8EncodeError { + /// Empty, overflowing, or mismatched dimensions. + FrameSize, + /// A dimension cannot fit the output format. + TooLarge, + /// The output buffer could not be allocated. + OutOfMemory, +} +/// Reserves the DPX output up front; a failed reservation becomes [`Rgba8EncodeError::OutOfMemory`]. +fn output_buffer(size: usize) -> std::result::Result, Rgba8EncodeError> { + let mut buf = Vec::new(); + match buf.try_reserve_exact(size) { + Ok(()) => Ok(buf), + Err(_) => Err(Rgba8EncodeError::OutOfMemory), + } +} +fn checked_len(rgba: &[u8], w: u32, h: u32) -> std::result::Result { + let n = (w as usize).checked_mul(h as usize).and_then(|n| n.checked_mul(4)).ok_or(Rgba8EncodeError::FrameSize)?; + if w == 0 || h == 0 || rgba.len() != n { + return Err(Rgba8EncodeError::FrameSize); + } + Ok(n) +} +/// Header size of the DPX files we write (file + image + orientation + film + TV headers). +const DPX_HEADER: usize = 2048; + +/// A DPX file: 10-bit RGB (8-bit values widened by bit replication, so 0 → 0 and 255 → 1023). +pub fn encode_rgba8_dpx10(rgba: &[u8], w: u32, h: u32, creator: &str) -> std::result::Result, Rgba8EncodeError> { + let n = checked_len(rgba, w, h)?; + // one 32-bit word per pixel; `n` is already `w * h * 4` + let size = n.checked_add(DPX_HEADER).ok_or(Rgba8EncodeError::FrameSize)?; + let file_size = u32::try_from(size).map_err(|_| Rgba8EncodeError::TooLarge)?; + // SMPTE ST 268: undefined numeric fields are all ones, undefined text fields are NUL + let mut hd = vec![0xFF_u8; DPX_HEADER]; + for (a, b) in [ + (36, 660), // file name, creation time, creator, project, copyright + (664, 768), // reserved + (1356, 1408), // reserved + (1432, 1620), // source file name, time, input device, serial number + (1644, 1664), // reserved + (1664, 1712), // film manufacturer, type, offset, prefix, count, format + (1732, 1920), // frame id, slate, reserved + (1972, 2048), // reserved + ] { + hd[a..b].fill(0); + } + for e in 0..8 { + let d = 780 + e * 72 + 40; + hd[d..d + 32].fill(0); // element descriptions + } + let put32 = |hd: &mut Vec, at: usize, v: u32| hd[at..at + 4].copy_from_slice(&v.to_be_bytes()); + let put16 = |hd: &mut Vec, at: usize, v: u16| hd[at..at + 2].copy_from_slice(&v.to_be_bytes()); + // file information header + hd[0..4].copy_from_slice(b"SDPX"); + put32(&mut hd, 4, DPX_HEADER as u32); // offset to the image data + hd[8..16].copy_from_slice(b"V2.0\0\0\0\0"); + put32(&mut hd, 16, file_size); + put32(&mut hd, 20, 1); // ditto key: new frame + put32(&mut hd, 24, 1664); // generic header length + put32(&mut hd, 28, 384); // industry header length + put32(&mut hd, 32, 0); // user data length + let creator = creator.as_bytes(); + let n = creator.len().min(100); + hd[160..160 + n].copy_from_slice(&creator[..n]); + // image information header + put16(&mut hd, 768, 0); // orientation: left to right, top to bottom + put16(&mut hd, 770, 1); // one image element + put32(&mut hd, 772, w); + put32(&mut hd, 776, h); + let el = 780; + put32(&mut hd, el, 0); // unsigned + put32(&mut hd, el + 4, 0); // reference low data code + put32(&mut hd, el + 12, 1023); // reference high data code + hd[el + 20] = 50; // descriptor: RGB + hd[el + 21] = 6; // transfer: ITU-R BT.709 + hd[el + 22] = 6; // colorimetric: ITU-R BT.709 + hd[el + 23] = 10; // bit depth + put16(&mut hd, el + 24, 1); // packing: filled to 32-bit words, method A + put16(&mut hd, el + 26, 0); // no encoding (uncompressed) + put32(&mut hd, el + 28, DPX_HEADER as u32); // offset to the element's data + put32(&mut hd, el + 32, 0); // end-of-line padding + put32(&mut hd, el + 36, 0); // end-of-image padding + // image orientation header: no offset, square pixels + put32(&mut hd, 1408, 0); + put32(&mut hd, 1412, 0); + put32(&mut hd, 1424, w); + put32(&mut hd, 1428, h); + put32(&mut hd, 1628, 1); + put32(&mut hd, 1632, 1); + // television header: progressive + hd[1928] = 0; + // The caller already holds the frame; reserve the output fallibly so a large export reports + // an error instead of aborting the process. + let mut out = output_buffer(size)?; + out.extend_from_slice(&hd); + let ten = |v: u8| (u32::from(v) << 2) | (u32::from(v) >> 6); + for &[r, g, b, _] in rgba.as_chunks::<4>().0 { + out.extend_from_slice(&((ten(r) << 22) | (ten(g) << 12) | (ten(b) << 2)).to_be_bytes()); + } + Ok(out) +} + +#[cfg(test)] +mod tests { + use super::*; + #[test] + fn packed_rgb_and_creator() { + let bytes = encode_rgba8_dpx10(&[255, 0, 128, 255], 1, 1, "FilmCraft").unwrap(); + assert_eq!(&bytes[..4], b"SDPX"); + assert_eq!(&bytes[160..169], b"FilmCraft"); + assert_eq!(&bytes[2048..], &((1023u32 << 22) | (514u32 << 2)).to_be_bytes()); + assert_eq!(encode_rgba8_dpx10(&[], 0, 1, ""), Err(Rgba8EncodeError::FrameSize)); + } + #[test] + fn output_reservation_failure_maps_to_out_of_memory() { + assert!(matches!(output_buffer(usize::MAX), Err(Rgba8EncodeError::OutOfMemory))); + assert_eq!(output_buffer(DPX_HEADER).map(|b| b.capacity() >= DPX_HEADER), Ok(true)); + } +} diff --git a/crates/drawingml/src/chart/read.rs b/crates/drawingml/src/chart/read.rs index 9919433e9..38ce9f605 100644 --- a/crates/drawingml/src/chart/read.rs +++ b/crates/drawingml/src/chart/read.rs @@ -443,7 +443,7 @@ fn read_axes<'a>(c: &mut Chart, pa: &'a El, primary_val: &Option) -> (Op // A gridline whose `c:spPr/a:ln` has `a:noFill` is not drawn, though the element is there. let grid = |a: &El, k: &str| { a.child(Local(k)) - .is_some_and(|g| !g.child(Local("spPr")).and_then(|p| p.child(Local("ln"))).is_some_and(|l| l.child(Local("noFill")).is_some())) + .is_some_and(|g| g.child(Local("spPr")).and_then(|p| p.child(Local("ln"))).is_none_or(|l| l.child(Local("noFill")).is_none())) }; if c.kind == ChartType::Combo && c.series.iter().any(|s| s.secondary_axis) { let primary_id = val_ax.and_then(|a| a.child(Local("axId")).and_then(|c| c.attr_local("val"))); diff --git a/crates/drawingml/src/read.rs b/crates/drawingml/src/read.rs index 496870774..9e9f8ff3e 100644 --- a/crates/drawingml/src/read.rs +++ b/crates/drawingml/src/read.rs @@ -166,6 +166,12 @@ pub fn blip_fill(src: &mut dyn Source, bf: &El) -> Option { { pf.artistic = Some(crate::style::Artistic { effect, original }); } + pf.svg = blip.and_then(|b| b.child(Local("extLst"))).and_then(|extensions| { + extensions + .elements() + .filter(|extension| extension.local() == "ext" && extension.attr_local("uri") == Some("{96DAC541-7B7A-43D3-8B79-37D633B846F1}")) + .find_map(|extension| extension.child(Local("svgBlip")).and_then(|svg| media_ref(src, svg))) + }); if let Some(b) = blip { for e in b.elements() { match e.local() { diff --git a/crates/drawingml/src/style.rs b/crates/drawingml/src/style.rs index 291ee4e76..36b5a6922 100644 --- a/crates/drawingml/src/style.rs +++ b/crates/drawingml/src/style.rs @@ -184,6 +184,9 @@ impl Default for PictureMode { #[serde(default, rename_all = "camelCase")] pub struct PictureFill { pub media: MediaId, + /// Optional vector SVG version of `media`; the regular media remains the bitmap fallback. + #[serde(skip_serializing_if = "Option::is_none")] + pub svg: Option, /// Source crop as fractions of the image (left, top, right, bottom); negative values pad. pub crop: [f64; 4], pub mode: PictureMode, diff --git a/crates/drawingml/src/write.rs b/crates/drawingml/src/write.rs index 0d9b83b1a..25bfb89db 100644 --- a/crates/drawingml/src/write.rs +++ b/crates/drawingml/src/write.rs @@ -107,6 +107,7 @@ pub fn blip_fill(w: &mut Writer, out: &mut dyn Sink, tag: &str, pf: &PictureFill w.tag(tag).attr("rotWithShape", if pf.upright { "0" } else { "1" }).open(); let adj = &pf.adjust; let has_kids = pf.alpha.is_some() + || pf.svg.is_some() || adj.brightness != 0.0 || adj.contrast != 0.0 || adj.grayscale @@ -168,6 +169,13 @@ pub fn blip_fill(w: &mut Writer, out: &mut dyn Sink, tag: &str, pf: &PictureFill w.close(); w.close(); } + if let Some(svg) = pf.svg.and_then(|id| out.media_rel(id, "image")) { + w.open("a:extLst", &[]); + w.tag("a:ext").attr("uri", "{96DAC541-7B7A-43D3-8B79-37D633B846F1}").open(); + w.tag("asvg:svgBlip").attr("xmlns:asvg", "http://schemas.microsoft.com/office/drawing/2016/SVG/main").attr("r:embed", &svg).empty(); + w.close(); + w.close(); + } w.close(); } else { w.tag("a:blip").list(a).empty(); diff --git a/crates/dwg/Cargo.toml b/crates/dwg/Cargo.toml new file mode 100644 index 000000000..d2bc8ffff --- /dev/null +++ b/crates/dwg/Cargo.toml @@ -0,0 +1,16 @@ +[package] +name = "craft-dwg" +description = "Bounded DWG↔DXF byte bridge with an optional child-process worker" +version.workspace = true +edition.workspace = true +license.workspace = true +rust-version.workspace = true +repository.workspace = true + +[lints] +workspace = true + +# acadrust is used unmodified and only on targets supported by its 0.6.6 platform APIs. +[target.'cfg(not(any(target_arch = "wasm32", all(target_os = "freebsd", target_arch = "aarch64"))))'.dependencies] +acadrust = { workspace = true } +log = { workspace = true } diff --git a/crates/dwg/README.md b/crates/dwg/README.md new file mode 100644 index 000000000..80088a89e --- /dev/null +++ b/crates/dwg/README.md @@ -0,0 +1,20 @@ +# craft-dwg + +A byte-level bridge between DWG and DXF. It does not expose `acadrust` types: callers pass bytes, +choose limits, and receive converted bytes plus notes about data the conversion could not preserve. +Application document mapping remains with the caller. + +The default limits reject DWG inputs over 512 MiB and stop DXF output over 1 GiB. The bridge +also catches panics from the dependency and validates its conversion diagnostics. Because a small +hostile file can declare very large internal sections, size checks alone cannot prevent an +allocation failure inside the decoder. On native applications, call `use_own_worker()` during +startup and call `serve_if_asked()` before normal argument handling. Subsequent `decode` and +`encode_*` calls then run in a child process with a 30-minute timeout; a crash or stack overflow +is contained to that child. The worker has no operating-system memory limit yet. Without a +configured worker, conversions are refused rather than run in the current process: call +`allow_in_process()` only for input that is known to be trusted (tests, or bytes this process +produced itself). Coordinates, once a worker is configured but cannot start, also fail closed. + +The bridge is not built for `wasm32` or FreeBSD/aarch64. Those targets retain the public API and +return a clear unsupported-build error. The `acadrust` 0.6.6 dependency is used unmodified under +MPL-2.0; its license and third-party notices are in `licenses/dwg/` at the workspace root. diff --git a/crates/dwg/build.rs b/crates/dwg/build.rs new file mode 100644 index 000000000..e97782ab4 --- /dev/null +++ b/crates/dwg/build.rs @@ -0,0 +1,12 @@ +//! Sets `no_dwg_bridge` on the targets built without acadrust (see Cargo.toml): the DWG functions +//! there return an error that says to use DXF. + +fn main() { + println!("cargo:rerun-if-changed=build.rs"); + println!("cargo::rustc-check-cfg=cfg(no_dwg_bridge)"); + let arch = std::env::var("CARGO_CFG_TARGET_ARCH").unwrap_or_default(); + let os = std::env::var("CARGO_CFG_TARGET_OS").unwrap_or_default(); + if arch == "wasm32" || (os == "freebsd" && arch == "aarch64") { + println!("cargo::rustc-cfg=no_dwg_bridge"); + } +} diff --git a/crates/dwg/src/lib.rs b/crates/dwg/src/lib.rs new file mode 100644 index 000000000..c77fdbf4c --- /dev/null +++ b/crates/dwg/src/lib.rs @@ -0,0 +1,2137 @@ +//! DWG support through a DXF bridge. +//! +//! DWG files are read with the `acadrust` crate (MPL-2.0, used unmodified as a dependency) and +//! converted to DXF bytes for the caller's DXF reader; saving goes the other way. Keeping the +//! dependency behind this byte-level API prevents its document types from leaking to consumers. +//! Native targets only (the dependency memory-maps files). +//! +//! # Guardrails around the third-party decoder +//! +//! The decoder allocates tables and section buffers from sizes a (possibly damaged or hostile) +//! file declares, and once it starts it cannot be interrupted (an allocation failure aborts the +//! process, which `catch_unwind` cannot recover). Because the dependency is used unmodified, this +//! crate guards it from the outside: +//! +//! - Both directions reject oversized input *before* the decoder runs ([`Limits::max_dwg_bytes`] +//! for DWG→DXF, [`Limits::max_dxf_bytes`] for DXF→DWG), and the DWG→DXF output is capped while +//! it is written (`CappedWriter`). +//! - The reader and writer run inside `catch_unwind`, so a decode failure is reported instead of +//! crashing. +//! +//! These bound the input- and output-size variants. A small file that declares a huge internal +//! table or section can still make the dependency allocate before any size is known, so the app +//! runs both directions in a child process ([`worker`]): what that kills is the child. +//! +//! Isolation is not optional for untrusted input. With no worker configured the conversion is +//! refused rather than run here, and a worker that cannot start is an error, so a missing worker +//! never turns into decoding a hostile file in the caller's process. [`worker::allow_in_process`] +//! is the explicit exception, for input the caller knows is trusted. +#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] +#![forbid(unsafe_code)] + +/// DWG version codes we can write. +pub const VERSIONS: &[(&str, &str)] = + &[("AC1015", "2000"), ("AC1018", "2004"), ("AC1021", "2007"), ("AC1024", "2010"), ("AC1027", "2013"), ("AC1032", "2018")]; + +/// True when the bytes look like a DWG file (`AC10xx` magic). +pub fn is_dwg(bytes: &[u8]) -> bool { + bytes.len() > 6 && bytes.starts_with(b"AC10") && bytes.get(4..6).is_some_and(|v| v.iter().all(u8::is_ascii_digit)) +} + +/// The DWG version code of a file (e.g. "AC1032"). +pub fn version(bytes: &[u8]) -> Option { + is_dwg(bytes).then(|| String::from_utf8_lossy(bytes.get(0..6).unwrap_or_default()).to_string()) +} + +/// Size limits for reading a DWG file. DWG is compressed and the DXF text it converts to is +/// several times larger (about 6x in our tests: a 10 MB DWG becomes about 60 MB of DXF), so +/// the limits are generous for real drawings. They exist because a damaged file, or one the reader +/// misreads, can make the DWG reader produce a drawing thousands of times larger than the file. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct Limits { + /// Largest DWG file read at all (checked before reading starts). + pub max_dwg_bytes: usize, + /// Largest DXF rendition of a DWG (checked while it is written). + pub max_dxf_bytes: usize, +} + +impl Limits { + /// 512 MB of DWG, 1 GB of DXF (more than the DXF reader's group-code limit fills). + pub const DEFAULT: Limits = Limits { max_dwg_bytes: 512 << 20, max_dxf_bytes: 1 << 30 }; +} + +/// A byte count in megabytes for messages ("9.9 MB", "1024 MB"). +#[cfg_attr(no_dwg_bridge, allow(dead_code))] +#[expect( + clippy::cast_precision_loss, + reason = "Megabyte formatting is rounded for display; exact integer precision beyond 2^53 bytes is not meaningful." +)] +fn mb(n: u64) -> String { + let m = n as f64 / f64::from(1u32 << 20); + if m >= 100.0 { format!("{m:.0} MB") } else { format!("{m:.1} MB") } +} + +#[cfg(not(no_dwg_bridge))] +pub mod worker; +#[cfg(not(no_dwg_bridge))] +pub use worker::{ + WORKER_ARG, allow_in_process, crash_worker_for_test, decode, encode_as, encode_kept_noted, encode_noted, serve_if_asked, use_own_worker, + use_worker, +}; +#[cfg(not(no_dwg_bridge))] +mod proxy_transport; + +#[cfg(not(no_dwg_bridge))] +mod native { + use std::fmt::Write as _; + use std::io::{Cursor, Write}; + use std::panic::AssertUnwindSafe; + + use super::{Limits, mb}; + + /// Read DWG bytes and return an ASCII DXF rendition, within [`Limits::DEFAULT`]. + /// + /// # Errors + /// + /// Returns an error when the DWG cannot be read or post-processed, or its DXF rendition exceeds the default limits. + pub fn dwg_to_dxf(bytes: &[u8]) -> Result, String> { + dwg_to_dxf_with(bytes, Limits::DEFAULT) + } + + /// [`dwg_to_dxf`] with explicit limits. Both are checked before the expensive step they + /// guard: the input size before the DWG reader runs, the DXF size while it is written, so a + /// runaway conversion stops at the limit instead of filling memory first. + /// + /// # Errors + /// + /// Returns an error when the input exceeds `limits`, decoding fails, or the DXF output exceeds `limits`. + pub fn dwg_to_dxf_with(bytes: &[u8], limits: Limits) -> Result, String> { + dwg_to_dxf_noted(bytes, limits).map(|(dxf, _)| dxf) + } + + /// [`dwg_to_dxf_with`], also returning notes on what the read left out. The reader runs in + /// failsafe mode: an object record it can't decode is skipped instead of failing the whole + /// file, and the notes say how many were skipped and of which types, so a partial read never + /// passes for a complete one. + /// + /// # Errors + /// + /// Returns an error when the input exceeds `limits`, decoding fails, or the DXF output exceeds `limits`. + pub fn dwg_to_dxf_noted(bytes: &[u8], limits: Limits) -> Result<(Vec, Vec), String> { + if bytes.len() > limits.max_dwg_bytes { + return Err(format!( + "DWG: the file is {}, larger than the {} accepted by this bridge", + mb(bytes.len() as u64), + mb(limits.max_dwg_bytes as u64) + )); + } + let outcome = std::panic::catch_unwind(|| { + acadrust::DwgReader::from_stream_with_options(Cursor::new(bytes), acadrust::DwgReadOptions::failsafe()).read_with_stats() + }) + .map_err(|_| "the DWG reader failed on this file".to_string())? + .map_err(|e| format!("DWG: {e}"))?; + // Failsafe skips records, not whole files: when the objects stream itself couldn't be + // read (a damaged header or section map, or not a DWG at all), nothing was read. + if !outcome.stats.record_stream_read { + let why = outcome.stats.diagnostics.first().map(|d| d.message.trim_start_matches("Failsafe: ").to_string()); + return Err(format!("DWG: {}", why.unwrap_or_else(|| "the file's objects couldn't be read".to_string()))); + } + let mut notes = read_notes(&outcome.stats); + let mut doc = outcome.document; + notes.extend( + guarded(AssertUnwindSafe(|| application_objects(&mut doc))).map_err(|()| "the DWG application object conversion failed".to_string())?, + ); + // After application_objects, so the proxy objects it makes are numbered as the DXF's. + notes.extend( + guarded(AssertUnwindSafe(|| super::proxy_transport::prepare(&mut doc))).map_err(|()| "the DWG proxy conversion failed".to_string())?, + ); + let orphans = guarded(AssertUnwindSafe(|| post_process(&mut doc))).map_err(|()| "the DWG post-processing failed on this file".to_string())?; + if orphans > 0 { + let (s, were) = if orphans == 1 { ("", "was") } else { ("s", "were") }; + notes.push(format!("DWG: {orphans} DGN line style component{s} that no line style definition uses {were} left out")); + } + let mut out = CappedWriter { buf: Vec::new(), limit: limits.max_dxf_bytes, exceeded: false }; + let written = std::panic::catch_unwind(AssertUnwindSafe(|| { + acadrust::DxfWriter::new(&doc).write_to_writer(&mut out)?; + mtext_contents(&doc, &mut out) + })) + .map_err(|_| "the DWG→DXF conversion failed on this file".to_string())?; + if out.exceeded { + return Err(format!( + "DWG: converting this {} file produced more than {} of drawing data (the limit); \ + the file is probably damaged, or uses objects the DWG reader misreads", + mb(bytes.len() as u64), + mb(limits.max_dxf_bytes as u64) + )); + } + written.map_err(|e| format!("DWG→DXF: {e}"))?; + Ok((add_header_vars(out.buf, &doc.header), notes)) + } + + /// acadrust 0.6.6's ASCII writer treats a caret followed by a literal space as already + /// escaped, and turns line feeds into paragraph codes. Serialize affected MTEXT contents + /// from the native values instead, using its lossless string writer. All other groups, + /// including embedded-object data, remain byte-for-byte as the DXF writer emitted them. + fn mtext_contents(doc: &acadrust::CadDocument, out: &mut CappedWriter) -> acadrust::Result<()> { + use acadrust::io::dxf::{DxfStreamWriter, DxfTextWriter}; + let contents: std::collections::HashMap<_, _> = doc + .entities() + .filter_map(|e| match e { + acadrust::EntityType::MText(t) if t.value.contains(['^', '\t', '\n']) => Some((t.common.handle.value(), t.value.as_str())), + _ => None, + }) + .collect(); + if contents.is_empty() { + return Ok(()); + } + let dxf = std::mem::take(&mut out.buf); + let mut lines = dxf.split_inclusive(|b| *b == b'\n'); + let (mut mtext, mut text, mut written) = (false, None, false); + while let Some(code) = lines.next() { + let value = lines.next().ok_or_else(|| std::io::Error::other("incomplete DWG bridge DXF pair"))?; + match code.trim_ascii() { + b"0" => { + mtext = value.trim_ascii() == b"MTEXT"; + text = None; + written = false; + } + b"5" if mtext => { + text = std::str::from_utf8(value.trim_ascii()).ok().and_then(|s| u64::from_str_radix(s, 16).ok()).and_then(|h| contents.get(&h)); + } + b"101" => mtext = false, + b"1" | b"3" if mtext && text.is_some() => { + if let Some(&text) = text + && !written + { + let mut writer = DxfTextWriter::new(&mut *out); + if doc.version < acadrust::DxfVersion::AC1021 + && let Some(encoding) = acadrust::io::dxf::code_page::encoding_from_code_page(&doc.header.code_page) + { + writer.set_encoding(encoding); + } + // Each character can expand to two caret-escape characters. Keep + // those pairs together and each group within 250 characters. + let mut rest = text; + while let Some((end, _)) = rest.char_indices().nth(125) { + let (chunk, tail) = rest.split_at(end); + writer.write_xrecord_string(3, chunk)?; + rest = tail; + } + writer.write_xrecord_string(1, rest)?; + written = true; + } + continue; + } + _ => {} + } + out.write_all(code)?; + out.write_all(value)?; + } + Ok(()) + } + + #[cfg(test)] + #[test] + fn mtext_contents_keeps_other_groups_and_the_output_limit() -> acadrust::Result<()> { + for version in [acadrust::DxfVersion::AC1015, acadrust::DxfVersion::AC1032] { + let mut doc = acadrust::CadDocument::new(); + doc.version = version; + doc.header.code_page = "ANSI_1252".into(); + let mut t = acadrust::MText::with_value("Ă© ^ C", acadrust::Vector3::ZERO); + t.common.handle = acadrust::Handle::new(0x10); + doc.add_entity(acadrust::EntityType::MText(Box::new(t)))?; + // Only the MTEXT's own 1/3 groups change. The embedded object's groups and the + // following TEXT record must survive byte-for-byte, including their spaces. + let prefix = b" 0\r\nMTEXT\r\n 5\r\n10\r\n 8\r\n0\r\n"; + let suffix = b"101\r\nEmbedded Object\r\n 1\r\n^ embedded\r\n 3\r\n^ tail\r\n 0\r\nTEXT\r\n 1\r\n^ plain\r\n"; + let raw = [prefix.as_slice(), b" 3\r\nold\r\n 1\r\ntext\r\n", suffix.as_slice()].concat(); + let mut out = CappedWriter { buf: raw.clone(), limit: 4096, exceeded: false }; + mtext_contents(&doc, &mut out)?; + let value: &[u8] = if version == acadrust::DxfVersion::AC1015 { b"\xe9 ^ C" } else { "Ă© ^ C".as_bytes() }; + assert_eq!(out.buf, [prefix.as_slice(), b" 1\r\n", value, b"\r\n", suffix.as_slice()].concat()); + let mut limited = CappedWriter { buf: raw, limit: 1, exceeded: false }; + assert!(mtext_contents(&doc, &mut limited).is_err()); + assert!(limited.exceeded); + } + Ok(()) + } + + /// Most record types named in the skipped-records note. + const NOTE_TYPES: usize = 6; + + /// The note on object records a failsafe read skipped: how many, and how many of each type + /// (most first), or nothing when the read was complete. + pub(crate) fn read_notes(stats: &acadrust::ReadStats) -> Vec { + let skipped = stats.skipped_source_records; + if skipped == 0 { + return Vec::new(); + } + let label = |t: Option<&str>| match t.map(str::parse::) { + Some(Ok(code)) => type_name(code).map_or_else(|| format!("type {code}"), str::to_string), + Some(Err(_)) => t.unwrap_or_default().to_string(), + None => "unknown".to_string(), + }; + let mut by_type: std::collections::BTreeMap = std::collections::BTreeMap::new(); + for d in stats.diagnostics.iter().filter(|d| d.record_handle.is_some()) { + *by_type.entry(label(d.record_type.as_deref())).or_default() += 1; + } + let mut types: Vec<(String, usize)> = by_type.into_iter().collect(); + types.sort_by(|a, b| b.1.cmp(&a.1).then(a.0.cmp(&b.0))); + let mut list: Vec = types.iter().take(NOTE_TYPES).map(|(t, n)| format!("{t} Ă— {n}")).collect(); + if types.len() > NOTE_TYPES { + list.push(format!("{} more types", types.len() - NOTE_TYPES)); + } + let of = if stats.source_records > 0 { format!(" of {}", stats.source_records) } else { String::new() }; + let records = if skipped == 1 && of.is_empty() { "object record" } else { "object records" }; + let (were, them) = if skipped == 1 { ("was", "it") } else { ("were", "them") }; + let which = if list.is_empty() { String::new() } else { format!(" ({})", list.join(", ")) }; + vec![format!( + "DWG: {skipped}{of} {records} couldn't be read and {were} left out{which}; \ + saving over the original file would lose {them}" + )] + } + + /// The DXF name of a DWG object type code with a fixed number (classes, 500 and up, are + /// numbered per file). + fn type_name(code: i16) -> Option<&'static str> { + Some(match code { + 1 => "TEXT", + 2 => "ATTRIB", + 3 => "ATTDEF", + 4 => "BLOCK", + 5 => "ENDBLK", + 6 => "SEQEND", + 7 => "INSERT", + 8 => "MINSERT", + 10..=14 => "VERTEX", + 15 | 16 | 29 | 30 => "POLYLINE", + 17 => "ARC", + 18 => "CIRCLE", + 19 => "LINE", + 20..=26 => "DIMENSION", + 27 => "POINT", + 28 => "3DFACE", + 31 => "SOLID", + 32 => "TRACE", + 33 => "SHAPE", + 34 => "VIEWPORT", + 35 => "ELLIPSE", + 36 => "SPLINE", + 37 => "REGION", + 38 => "3DSOLID", + 39 => "BODY", + 40 => "RAY", + 41 => "XLINE", + 42 => "DICTIONARY", + 43 => "OLEFRAME", + 44 => "MTEXT", + 45 => "LEADER", + 46 => "TOLERANCE", + 47 => "MLINE", + 48 => "BLOCK_CONTROL", + 49 => "BLOCK_RECORD", + 50 => "LAYER_CONTROL", + 51 => "LAYER", + 52 => "STYLE_CONTROL", + 53 => "STYLE", + 56 => "LTYPE_CONTROL", + 57 => "LTYPE", + 60 => "VIEW_CONTROL", + 61 => "VIEW", + 62 => "UCS_CONTROL", + 63 => "UCS", + 64 => "VPORT_CONTROL", + 65 => "VPORT", + 66 => "APPID_CONTROL", + 67 => "APPID", + 68 => "DIMSTYLE_CONTROL", + 69 => "DIMSTYLE", + 70 => "VX_CONTROL", + 71 => "VX_TABLE_RECORD", + 72 => "GROUP", + 73 => "MLINESTYLE", + 74 => "OLE2FRAME", + 77 => "LWPOLYLINE", + 78 => "HATCH", + 79 => "XRECORD", + 80 => "ACDBPLACEHOLDER", + 82 => "LAYOUT", + _ => return None, + }) + } + + /// Merged table cells onto the cells. acadrust 0.6.3 reads a 2010+ (AC1024+) DWG table's + /// merged cells into the table's `merged_ranges` only, while its DXF writer writes each cell's + /// own merge groups (173, 175, 176), so without this the merges would be lost. Tables whose + /// cells already carry merges are left alone. Each range's first cell gets the span; the + /// cells it covers are marked merged with no span of their own, as CADCraft's DXF writer does. + fn table_merges(doc: &mut acadrust::CadDocument) { + table_merges_within(doc, MAX_MERGE_CELLS); + } + + /// The most table cells [`table_merges`] visits in one document: each range is capped, but + /// can cover rows Ă— cols cells, so the total is bounded too. + const MAX_MERGE_CELLS: usize = 1 << 24; + + /// A table cell's own text height only when the file says it overrides the table style's. + /// acadrust's DXF writer writes a cell style's height (group 140) whenever the cell has a + /// style record, even a default 0.18 with no override flag, and CADCraft reads a 140 as an + /// override (#2501). The override is the cell style's TEXT_HEIGHT property flag or the + /// content's format override bit 128; without either the height is cleared (0 is no + /// override), so the table style's height applies. + fn table_cell_heights(doc: &mut acadrust::CadDocument) { + use acadrust::entities::table::CellStylePropertyFlags; + const CONTENT_TEXT_HEIGHT: i32 = 128; + for e in doc.entities_mut() { + let acadrust::entities::EntityType::Table(t) = e else { continue }; + for cell in t.rows.iter_mut().flat_map(|r| r.cells.iter_mut()) { + // The first content's own height (group 144) when its format override sets bit 128. + // CADCraft keeps one height for the cell's text, so a later content's override + // must not stand in for the first content's. + let content_height = cell + .contents + .first() + .filter(|c| c.format_override_flags & CONTENT_TEXT_HEIGHT != 0) + .map(|c| c.text_height) + .filter(|h| h.is_finite() && *h > 0.0); + if let Some(h) = content_height { + // CADCraft reads only the style's height (group 140), so carry it there. + cell.style.get_or_insert_with(acadrust::entities::table::CellStyle::new).text_height = h; + continue; + } + // A style overrides through its property flags or, in newer files, its override + // flags; anything else is the writer's default and is cleared. + if let Some(style) = cell.style.as_mut() + && !style.property_flags.contains(CellStylePropertyFlags::TEXT_HEIGHT) + && style.override_flags & CellStylePropertyFlags::TEXT_HEIGHT.bits() as i32 == 0 + { + style.text_height = 0.0; + } + } + } + } + + /// The fix-ups made to a document after the DWG read, returning how many orphaned DGN + /// components were dropped. The document comes from a third-party reader and is hostile. + fn post_process(doc: &mut acadrust::CadDocument) -> usize { + super::fail_in_test("post_process"); + current_style_names(doc); + table_merges(doc); + table_cell_heights(doc); + table_style_rows(doc, false); + keep_sort_handles(doc); + drop_orphaned_dgn_components(doc) + } + + /// Runs `f`, turning a panic into an error (the reader and writer are guarded the same way). + fn guarded(f: impl FnOnce() -> R + std::panic::UnwindSafe) -> Result { + std::panic::catch_unwind(f).map_err(|_| ()) + } + + /// The version a proxy record of a DWG of `version` names its format by (Open Design + /// Specification for .dwg, proxy objects): 19 for R13, 21 for R14, 23 for 2000, 25 for 2004, + /// 27 for 2007, 29 for 2010, 31 for 2013 and 33 for 2018; `None` for any other version. + pub(super) fn proxy_dwg_version(version: acadrust::DxfVersion) -> Option { + Some(match version.version_code() { + 1012 => 19, + 1014 => 21, + 1015 => 23, + 1018 => 25, + 1021 => 27, + 1024 => 29, + 1027 => 31, + 1032 => 33, + _ => return None, + }) + } + + /// Objects of another application's class that acadrust keeps only as their DWG record + /// (`ObjectType::Unknown`, which has no DXF form): its DXF writer left them out, and the + /// references to them went with them (a dimension's extension dictionary entry naming its + /// application data, say), with no note. Each is carried as a proxy object of its class instead + /// (DXF Reference, ACAD_PROXY_OBJECT), so the DXF keeps it and a DWG written from that DXF has it + /// again: its common object data (owner, reactors, extension dictionary) and the streams after + /// it (data bits, R2007+ strings, object references), as the Open Design Specification for .dwg + /// splits an object record. Objects of the drawing database's own classes are not turned into + /// another application's proxies. What is still without a DXF form is named in the notes, so a + /// save over the original is not mistaken for a complete one. + fn application_objects(doc: &mut acadrust::CadDocument) -> Vec { + use acadrust::objects::{ObjectType, ProxyObject}; + let classes: Vec = doc.classes.iter().cloned().collect(); + let source = doc.dwg_source_version.unwrap_or(doc.version); + if let Some(dwg_version) = proxy_dwg_version(source) { + let maintenance = i32::from(doc.maintenance_version); + for object in doc.objects.values_mut() { + let ObjectType::Unknown { handle, raw_dwg_data: Some(data), raw_dwg_handle_bits, raw_dwg_version: Some(version), type_name, .. } = + &*object + else { + continue; + }; + if !application_class(&classes, type_name) { + continue; + } + let raw = acadrust::entities::RawRecord { data: data.clone(), handle_bits: *raw_dwg_handle_bits, version: *version }; + let handle = *handle; + // A record that doesn't split stays as it was, and is named below. + let Ok(Ok(record)) = std::panic::catch_unwind(AssertUnwindSafe(|| object_record(&raw, handle.value()))) else { continue }; + let Some(class) = classes.iter().find(|c| c.class_number == record.kind && !c.is_an_entity) else { continue }; + *object = ObjectType::ProxyObject(ProxyObject { + handle, + owner: record.owner, + reactors: record.reactors, + xdictionary_handle: record.xdictionary, + proxy_id: 499, + class_id: i32::from(class.class_number), + dwg_version, + maintenance_version: maintenance, + version: (maintenance << 16) | dwg_version, + payload: record.payload, + text_payload: record.text_payload, + object_ids: record.object_ids, + ..Default::default() + }); + } + } + let mut left: std::collections::BTreeMap<&str, usize> = std::collections::BTreeMap::new(); + for object in doc.objects.values() { + if let ObjectType::Unknown { type_name, raw_dxf_codes: None, .. } = object { + let n = left.entry(type_name.as_str()).or_insert(0); + *n = n.saturating_add(1); + } + } + if left.is_empty() { + return Vec::new(); + } + let total = left.values().fold(0usize, |a, n| a.saturating_add(*n)); + let mut list: Vec = left.iter().take(NOTE_TYPES).map(|(t, n)| format!("{t:.60} Ă— {n}")).collect(); + if left.len() > NOTE_TYPES { + list.push(format!("{} more types", left.len() - NOTE_TYPES)); + } + vec![format!( + "DWG: {total} objects of types this bridge can't preserve were left out ({}); saving over the original file would lose them", + list.join(", ") + )] + } + + /// Whether `name` is an object class of another application, which only that application + /// rebuilds: not one of the drawing database's own classes (`AcDb…`/`ACDB…`/`ACAD_…`), whose + /// leftovers the database regenerates itself. + fn application_class(classes: &[acadrust::classes::DxfClass], name: &str) -> bool { + classes.iter().any(|c| { + !c.is_an_entity + && c.dxf_name.eq_ignore_ascii_case(name) + && !c.cpp_class_name.starts_with("AcDb") + && !["ACDB", "ACAD_", "ACAD"].iter().any(|p| c.dxf_name.to_ascii_uppercase().starts_with(p)) + }) + } + + /// An application object acadrust kept only as its record: the parts a proxy object carries. + struct ObjectRecord { + kind: i16, + owner: acadrust::types::Handle, + reactors: Vec, + xdictionary: Option, + payload: acadrust::objects::ProxyPayload, + text_payload: acadrust::objects::ProxyPayload, + object_ids: Vec, + } + + /// Split an object record into its common object data and the uninterpreted streams after it + /// (data bits, R2007+ strings and object references), as an application object's proxy holds + /// them (Open Design Specification for .dwg, common object data and proxy objects). The record + /// is hostile: sizes are checked against it, and acadrust's checked stream reader does the rest. + fn object_record(raw: &acadrust::entities::RawRecord, handle: u64) -> Result { + use acadrust::io::dwg::{ + DwgVersion, dwg_reference_type::DwgReferenceType, dwg_stream_readers::object_reader::DwgObjectReader, + dwg_stream_writers::bit_writer::DwgBitWriter, + }; + use acadrust::objects::{ProxyObjectReference, ProxyPayload, ProxyReferenceKind}; + use acadrust::types::{DxfVersion, Handle}; + if raw.data.len() > 32 << 20 { + return Err("object record exceeds the transport limit".into()); + } + let handle_bits = usize::try_from(raw.handle_bits).map_err(|_| "object has a negative handle stream size")?; + let total_bits = raw.data.len().checked_mul(8).ok_or("object record is too large")?; + if handle_bits > total_bits { + return Err("object handle stream exceeds its record".into()); + } + let version = DwgVersion::from_dxf_version(raw.version).map_err(|e| e.to_string())?; + let mut frame = DwgBitWriter::new(version, raw.version); + frame.write_modular_short(raw.data.len()); + if raw.version >= DxfVersion::AC1024 { + frame.write_modular_char(handle_bits); + } + frame.write_bytes(&raw.data); + let reader = DwgObjectReader::new(frame.into_bytes(), raw.version, std::collections::HashMap::new()).map_err(|e| e.to_string())?; + let (kind, mut streams) = reader.read_record_at(0).map_err(|e| e.to_string())?; + if kind < 500 { + return Err("object has no application class".into()); + } + let common = reader.read_common_non_entity_data(&mut streams, kind); + streams.check_error().map_err(|e| e.to_string())?; + if common.common.handle != handle { + return Err("object record names another handle".into()); + } + let read_bits = |count: usize, bit: &mut dyn FnMut() -> bool| -> Result { + if count > total_bits { + return Err("object stream exceeds its record".into()); + } + let n = u32::try_from(count).map_err(|_| "object stream is too large")?; + let mut bytes = vec![0u8; count.div_ceil(8)]; + for index in 0..count { + if bit() + && let Some(b) = bytes.get_mut(index / 8) + { + *b |= 0x80 >> (index % 8); + } + } + Ok(ProxyPayload::from_bits(&bytes, n)) + }; + let main = usize::try_from(streams.main_remaining_bits()).map_err(|_| "object body has an invalid size")?; + let payload = read_bits(main, &mut || streams.read_bit())?; + let text_payload = if raw.version >= DxfVersion::AC1021 { + let bits = usize::try_from(streams.text_remaining_bits()).map_err(|_| "object has a negative string stream size")?; + read_bits(bits, &mut || streams.read_text_bit())? + } else { + ProxyPayload::default() + }; + // Every reference to the end of the handle stream, null ones too: an object keeps a null pointer in its + // field's place, and the references after it are the next fields' (as acadrust reads a proxy's own). R2010+ + // records give the stream's exact length; the bits of its last byte past it are padding. + let mut object_ids = Vec::new(); + let mut left = if raw.version >= DxfVersion::AC1024 { + streams.get_handle_bits().max(0).min(streams.handle_remaining_bits()) + } else { + streams.handle_remaining_bits() + }; + while left >= 8 { + if object_ids.len() >= 100_000 { + return Err("object has too many references".into()); + } + let before = streams.handle_remaining_bits(); + let (target, reference) = streams.read_handle_reference(handle); + let used = before.saturating_sub(streams.handle_remaining_bits()); + if used <= 0 || used > left { + break; + } + left -= used; + let kind = match reference { + DwgReferenceType::SoftOwnership => ProxyReferenceKind::SoftOwnership, + DwgReferenceType::HardOwnership => ProxyReferenceKind::HardOwnership, + DwgReferenceType::SoftPointer => ProxyReferenceKind::SoftPointer, + DwgReferenceType::HardPointer => ProxyReferenceKind::HardPointer, + DwgReferenceType::Undefined => ProxyReferenceKind::Undefined, + }; + object_ids.push(ProxyObjectReference { handle: Handle::from(target), kind }); + } + streams.check_error().map_err(|e| e.to_string())?; + Ok(ObjectRecord { + kind, + owner: Handle::from(common.owner_handle), + reactors: common.reactors.into_iter().map(Handle::from).collect(), + xdictionary: common.xdictionary_handle.map(Handle::from), + payload, + text_payload, + object_ids, + }) + } + + /// [`table_merges`] visiting at most `budget` cells in all; a range that would pass it is + /// left unmerged. Returns the cells visited. + fn table_merges_within(doc: &mut acadrust::CadDocument, budget: usize) -> usize { + let mut left = budget; + for e in doc.entities_mut() { + let acadrust::entities::EntityType::Table(t) = e else { continue }; + if t.merged_ranges.is_empty() || t.rows.iter().flat_map(|r| &r.cells).any(|c| c.merge_width > 1 || c.merge_height > 1) { + continue; + } + let (rows, cols) = (t.rows.len(), t.columns.len()); + // A real table has at most one merge per cell; more ranges than that come from a + // damaged file and would only make this rows Ă— cols Ă— ranges. + let max_ranges = rows.saturating_mul(cols).min(1 << 16); + for r in t.merged_ranges.iter().take(max_ranges).copied().collect::>() { + // Clamp to the table: the ranges come from the file. + let (bottom, right) = (r.bottom_row.min(rows.saturating_sub(1)), r.right_col.min(cols.saturating_sub(1))); + if r.top_row > bottom || r.left_col > right || rows == 0 || cols == 0 || (r.top_row, r.left_col) == (bottom, right) { + continue; + } + let area = (bottom - r.top_row + 1).saturating_mul(right - r.left_col + 1); + if area > left { + continue; + } + left -= area; + let span = |n: usize| i32::try_from(n + 1).unwrap_or(i32::MAX); + for (ri, row) in t.rows.iter_mut().enumerate().take(bottom + 1).skip(r.top_row) { + for (ci, cell) in row.cells.iter_mut().enumerate().take(right + 1).skip(r.left_col) { + cell.merged = 1; + let first = (ri, ci) == (r.top_row, r.left_col); + cell.merge_width = if first { span(right - r.left_col) } else { 0 }; + cell.merge_height = if first { span(bottom - r.top_row) } else { 0 }; + } + } + } + } + budget - left + } + + /// Draw-order tables (SORTENTSTABLE) keep the sort handles the file gave them. acadrust's DXF + /// writer renumbers them past every entity handle, but an entity left out of a table sorts by + /// its own handle, so after renumbering it would draw below every listed entity (#2410). Giving + /// the table its codes verbatim makes the writer emit them unchanged. + fn keep_sort_handles(doc: &mut acadrust::CadDocument) { + use acadrust::objects::ObjectType; + let version = doc.version; + for o in doc.objects.values_mut() { + let ObjectType::SortEntitiesTable(t) = o else { continue }; + let hex = |h: acadrust::types::Handle| format!("{:X}", h.value()); + let mut codes = + vec![(5, hex(t.handle)), (330, hex(t.owner_handle)), (100, "AcDbSortentsTable".to_string()), (330, hex(t.block_owner_handle))]; + for e in t.entries() { + codes.push((331, hex(e.entity_handle))); + codes.push((5, hex(e.sort_handle))); + } + t.raw_dxf_codes = Some(codes); + t.raw_dxf_version = Some(version); + } + } + + /// Table styles' title and header suppression between the two places acadrust 0.6.3 keeps + /// it. 2010+ (AC1024+) DWG stores it only in the style's flags (title 1, header 2), while + /// acadrust's DXF reader and writer use the `title_suppressed`/`header_suppressed` fields + /// (groups 280, 281). `to_flags` copies the fields into the flags before a DWG is written; + /// otherwise the flags are copied into the fields after one is read. Older versions store + /// both and are left alone. + fn table_style_rows(doc: &mut acadrust::CadDocument, to_flags: bool) { + use acadrust::objects::{ObjectType, TableStyleFlags}; + if doc.version < acadrust::DxfVersion::AC1024 { + return; + } + for o in doc.objects.values_mut() { + let ObjectType::TableStyle(s) = o else { continue }; + for (bit, suppressed) in + [(TableStyleFlags::TITLE_SUPPRESSED, &mut s.title_suppressed), (TableStyleFlags::HEADER_SUPPRESSED, &mut s.header_suppressed)] + { + if to_flags { + s.flags.set(bit, *suppressed); + } else { + *suppressed = s.flags.contains(bit); + } + } + } + } + + /// The DWG dimension flags byte. Its bit 1 is DXF group 70's bit 32 (ODA DWG specification, + /// common dimension data), which acadrust keeps no copy of from DXF: it is set here as + /// CADCraft's DXF writer always sets it (#1094). Bit 0, the opposite of group 70's bit 128, + /// acadrust 0.6.6 reads and writes itself. + fn dimension_flags(doc: &mut acadrust::CadDocument) { + for e in doc.entities_mut() { + let acadrust::entities::EntityType::Dimension(d) = e else { continue }; + d.base_mut().dwg_flags_byte |= 0x02; + } + } + + /// Leave out DGN line style components (`LS*COMPONENT`) that no line style definition uses + /// (#1482). A drawing that once came from `MicroStation` can carry hundreds of thousands of + /// them, orphaned copies piled up by earlier DGN imports and copy-pastes (`AutoCAD`'s PURGE + /// removes them as orphaned data): in a real 83 MB site plan they were 888,483 of 889,053 + /// components and made its DXF rendition 1.2 GB, over [`Limits::DEFAULT`], so the file + /// didn't open at all. No definition that was read refers to them (a definition the reader + /// skipped can't use them either) and CADCraft doesn't draw DGN line styles, so nothing + /// visible is lost. + /// Every component a definition reaches (directly, or through compound and point + /// components) is kept, as are the definitions; dictionary entries that named a dropped + /// component go with it. Returns how many were left out. + fn drop_orphaned_dgn_components(doc: &mut acadrust::CadDocument) -> usize { + use acadrust::objects::{DgnLineStyleData, ObjectType}; + use std::collections::{HashMap, HashSet}; + // Each component's references, and the definitions' root components to start from. + let mut refs: HashMap> = HashMap::new(); + let mut stack: Vec = Vec::new(); + for (h, o) in &doc.objects { + let ObjectType::DgnLineStyle(ls) = o else { continue }; + match &ls.data { + DgnLineStyleData::Definition { root_component, .. } => stack.push(*root_component), + DgnLineStyleData::Component { component, .. } => { + refs.insert(*h, component.references()); + } + DgnLineStyleData::Registered { .. } => {} + } + } + if refs.is_empty() { + return 0; + } + // Each component is visited once, so cycles and shared subtrees end. + let mut used: HashSet = HashSet::new(); + while let Some(h) = stack.pop() { + if let Some(r) = refs.get(&h) + && used.insert(h) + { + stack.extend(r.iter().copied()); + } + } + let orphans: HashSet = refs.into_keys().filter(|h| !used.contains(h)).collect(); + if orphans.is_empty() { + return 0; + } + doc.objects.retain(|h, _| !orphans.contains(h)); + doc.dgn_ls_components.retain(|h, _| !orphans.contains(h)); + for o in doc.objects.values_mut() { + let ObjectType::Dictionary(d) = o else { continue }; + if d.entries.iter().any(|(_, e)| orphans.contains(e)) { + d.entries.retain(|(_, e)| !orphans.contains(e)); + let keys: HashSet<&str> = d.entries.iter().map(|(k, _)| k.as_str()).collect(); + d.hard_owner_entries.retain(|k| keys.contains(k.as_str())); + } + } + orphans.len() + } + + /// Leave out what acadrust can't write as this DWG version: other programs' custom entities + /// and objects without raw DWG data of that version. acadrust 0.6.4 refuses the whole file + /// for one of them (0.6.3 left them out); CADCraft already tells the user which entity types + /// a DWG doesn't keep (`not_saved`), so the rest of the drawing is written. Returns how many + /// records of each DXF type were left out, so a save can say so. + fn drop_unwritable(doc: &mut acadrust::CadDocument) -> super::Unwritten { + let version = doc.version; + let mut dropped = super::Unwritten::new(); + let mut count = |name: &str| { + let n = dropped.entry(name.trim().to_ascii_uppercase()).or_insert(0); + *n = n.saturating_add(1); + }; + let unwritable: Vec = doc + .entities() + .filter_map(|e| match e { + acadrust::entities::EntityType::Unknown(u) + if u.raw_dwg_data.is_none() || u.dwg_source_version.is_some_and(|v| v != version) || version == acadrust::DxfVersion::AC1009 => + { + count(&u.dxf_name); + Some(u.common.handle) + } + _ => None, + }) + .collect(); + for h in &unwritable { + doc.remove_entity(*h); + } + let objects: Vec = doc + .objects + .iter() + .filter_map(|(h, o)| match o { + acadrust::objects::ObjectType::Unknown { type_name, raw_dwg_data, raw_dwg_version, .. } + if raw_dwg_data.is_none() || raw_dwg_version.is_some_and(|v| v != version) || version == acadrust::DxfVersion::AC1009 => + { + count(type_name); + Some(*h) + } + _ => None, + }) + .collect(); + for h in &objects { + doc.objects.remove(h); + } + dropped + } + + /// Header variables the DWG reader fills in but acadrust's DXF writer (0.6.3) leaves out of + /// the HEADER section: (name, group code, value). + fn unwritten_header_vars(h: &acadrust::document::HeaderVariables) -> [(&'static str, i32, String); 7] { + [ + // DWG holds radians; `$ANGBASE` is in degrees. + ("$ANGBASE", 50, h.angle_base.to_degrees().to_string()), + ("$ANGDIR", 70, h.angle_direction.to_string()), + ("$FILLETRAD", 40, h.fillet_radius.to_string()), + ("$CHAMFERA", 40, h.chamfer_distance_a.to_string()), + ("$CHAMFERB", 40, h.chamfer_distance_b.to_string()), + ("$ELEVATION", 40, h.elevation.to_string()), + ("$THICKNESS", 40, h.thickness.to_string()), + ] + } + + /// Add the [`unwritten_header_vars`] to the HEADER section of acadrust's DXF text, before its + /// ENDSEC; a variable the writer already wrote is left alone. Text of another layout than + /// acadrust's (` 0\r\nSECTION\r\n 2\r\nHEADER\r\n` … ` 0\r\nENDSEC\r\n`) is returned unchanged. + fn add_header_vars(mut dxf: Vec, h: &acadrust::document::HeaderVariables) -> Vec { + const START: &[u8] = b" 0\r\nSECTION\r\n 2\r\nHEADER\r\n"; + const END: &[u8] = b"\r\n 0\r\nENDSEC\r\n"; + if !dxf.starts_with(START) { + return dxf; + } + let Some(end) = dxf.windows(END.len()).position(|w| w == END).map(|p| p + 2) else { return dxf }; + let section = dxf.get(..end).unwrap_or_default(); + let mut add = String::new(); + for (name, code, value) in unwritten_header_vars(h) { + let written = section.windows(name.len() + 2).any(|w| w.starts_with(name.as_bytes()) && w.ends_with(b"\r\n")); + if !written && value.parse::().is_ok_and(f64::is_finite) { + let _ = write!(add, " 9\r\n{name}\r\n{code:>3}\r\n{value}\r\n"); + } + } + dxf.splice(end..end, add.into_bytes()); + dxf + } + + /// Collects the DXF text and fails the write once it would grow past `limit`. + struct CappedWriter { + buf: Vec, + limit: usize, + exceeded: bool, + } + + impl Write for CappedWriter { + fn write(&mut self, b: &[u8]) -> std::io::Result { + // Sticky: once over, every later write fails too. + if self.exceeded || self.buf.len().saturating_add(b.len()) > self.limit { + self.exceeded = true; + return Err(std::io::Error::other("DXF size limit reached")); + } + self.buf.extend_from_slice(b); + Ok(b.len()) + } + fn flush(&mut self) -> std::io::Result<()> { + Ok(()) + } + } + + /// Oldest DWG version written; older DXF input is written as this version. + const MIN_DWG_VERSION: acadrust::DxfVersion = acadrust::DxfVersion::AC1018; + + /// Convert DXF bytes into a DWG file of the DXF's version, and at least `AutoCAD` 2004 + /// (AC1018). + /// + /// # Errors + /// + /// Returns an error when the DXF cannot be decoded or the DWG cannot be written, including oversized input. + pub fn dxf_to_dwg(dxf: &[u8]) -> Result, String> { + convert(dxf, None, None).map(|(dwg, _)| dwg) + } + + /// Convert DXF bytes into a DWG file of `version`, a code from [`super::VERSIONS`] ("AC1032"), + /// whatever version the DXF declares. + /// + /// # Errors + /// + /// Returns an error for an unsupported version, invalid DXF, oversized input, or a DWG write failure. + pub fn dxf_to_dwg_as(dxf: &[u8], version: &str) -> Result, String> { + dxf_to_dwg_noted(dxf, version).map(|(dwg, _)| dwg) + } + + /// [`dxf_to_dwg_as`], also returning what the DWG writer left out: how many records of each + /// DXF type it can't write (other programs' custom entities and objects that the DWG reader + /// didn't read from a DWG of this version), empty when the DWG keeps everything. + pub fn dxf_to_dwg_noted(dxf: &[u8], version: &str) -> Result<(Vec, super::Unwritten), String> { + convert(dxf, Some(writable(version)?), None) + } + + /// Like [`dxf_to_dwg_as`], for a DXF that declares an older version than the groups it carries + /// over from another file. CADCraft's R2000 DWG bridge holds the objects it keeps without + /// modelling them (render settings, fields, associative and dynamic block data) as they were + /// read from a file of version `kept` ("AC1032"), and acadrust reads several of those + /// differently by version: read as R2000, an AC1032 file's mental ray render settings came out + /// a class version too high, for which the ODA rejects the whole DWG, and a field lost its + /// cached value. So the DXF is read as `kept` declares. A `kept` version that is unknown, or + /// not newer than the DXF's own, changes nothing. + pub fn dxf_to_dwg_kept_as(dxf: &[u8], version: &str, kept: &str) -> Result, String> { + dxf_to_dwg_kept_noted(dxf, version, kept).map(|(dwg, _)| dwg) + } + + /// [`dxf_to_dwg_kept_as`], also returning what the DWG writer left out, as + /// [`dxf_to_dwg_noted`] does. + pub fn dxf_to_dwg_kept_noted(dxf: &[u8], version: &str, kept: &str) -> Result<(Vec, super::Unwritten), String> { + let kept = acadrust::DxfVersion::from_version_string(kept.trim().to_ascii_uppercase().as_str()); + let kept = (kept != acadrust::DxfVersion::Unknown).then_some(kept); + convert(dxf, Some(writable(version)?), kept) + } + + /// The DWG version for a code from [`super::VERSIONS`]. + fn writable(version: &str) -> Result { + Ok(match version.trim().to_ascii_uppercase().as_str() { + "AC1015" => acadrust::DxfVersion::AC1015, + "AC1018" => acadrust::DxfVersion::AC1018, + "AC1021" => acadrust::DxfVersion::AC1021, + "AC1024" => acadrust::DxfVersion::AC1024, + "AC1027" => acadrust::DxfVersion::AC1027, + "AC1032" => acadrust::DxfVersion::AC1032, + _ => return Err(super::unsupported(version)), + }) + } + + /// The ASCII DXF with its `$ACADVER` set to `kept` when that is newer than the version it + /// declares, or `None` to read it as it is (binary, no `$ACADVER`, or non-ASCII text without a + /// `$DWGCODEPAGE` to decode it by). From AutoCAD 2007 (AC1021) a DXF is UTF-8, so other bytes + /// are first decoded with the code page the DXF declares; `\U+XXXX` escapes read the same in + /// every version, and CADCraft's R2000 DXF is ASCII with such escapes. + pub(super) fn declare_version(dxf: &[u8], kept: acadrust::DxfVersion) -> Option> { + if dxf.starts_with(b"AutoCAD Binary DXF") { + return None; + } + let (_, declared) = header_value(dxf, b"$ACADVER")?; + let declared = acadrust::DxfVersion::from_version_string(std::str::from_utf8(declared).ok()?); + if declared >= kept { + return None; + } + let text: std::borrow::Cow<'_, [u8]> = if dxf.is_ascii() || kept < acadrust::DxfVersion::AC1021 { + std::borrow::Cow::Borrowed(dxf) + } else { + // As acadrust decodes an older DXF: by its code page, Windows-1252 for one it doesn't + // know; a code page that names UTF-8 needs no decoding. + let (_, code_page) = header_value(dxf, b"$DWGCODEPAGE")?; + match acadrust::io::dxf::code_page::encoding_from_code_page(std::str::from_utf8(code_page).ok()?) { + Some(encoding) => { + let (text, errors) = encoding.decode_without_bom_handling(dxf); + if errors { + return None; + } + std::borrow::Cow::Owned(text.into_owned().into_bytes()) + } + None => std::borrow::Cow::Borrowed(std::str::from_utf8(dxf).ok()?.as_bytes()), + } + }; + let (at, value) = header_value(&text, b"$ACADVER")?; + let code = kept.to_string(); + let mut out = Vec::with_capacity(text.len().saturating_add(code.len())); + out.extend_from_slice(text.get(..at)?); + out.extend_from_slice(code.as_bytes()); + out.extend_from_slice(text.get(at.checked_add(value.len())?..)?); + Some(out) + } + + /// The byte offset and the trimmed value of header variable `name` in an ASCII DXF: the value + /// of the group that follows the `9` group naming it. The file is read as (group code, + /// value) line pairs, in one pass with no buffering, up to the first `0`/`ENDSEC`, the end + /// of the HEADER section: a large input without one costs time, not memory (a line index of + /// a 32 MiB file used 0.8 GB). Lines may end in `\r\n`. + pub(super) fn header_value<'a>(dxf: &'a [u8], name: &[u8]) -> Option<(usize, &'a [u8])> { + let mut offset = 0usize; + let mut lines = dxf.split(|b| *b == b'\n').map(|line| { + let at = offset.saturating_add(line.len() - line.trim_ascii_start().len()); + offset = offset.saturating_add(line.len()).saturating_add(1); + (at, line.trim_ascii()) + }); + while let (Some((_, code)), Some((_, value))) = (lines.next(), lines.next()) { + if code == b"0" && value == b"ENDSEC" { + return None; + } + if code == b"9" && value == name { + let (_, code) = lines.next()?; + let (at, value) = lines.next()?; + return (code != b"0" && code != b"9").then_some((at, value)); + } + } + None + } + + /// The current dimension and text styles' names from their handles: a DWG stores the + /// handles, and the reader leaves the names at "Standard", so a drawing reopened from DWG + /// lost its current dimension style (#1051). + fn current_style_names(doc: &mut acadrust::CadDocument) { + let h = doc.header.current_dimstyle_handle; + if let Some(n) = doc.dim_styles.iter().find(|ds| ds.handle == h && !h.is_null()).map(|ds| ds.name.clone()) { + doc.header.current_dimstyle_name = n; + } + let h = doc.header.current_text_style_handle; + if let Some(n) = doc.text_styles.iter().find(|ts| ts.handle == h && !h.is_null()).map(|ts| ts.name.clone()) { + doc.header.current_text_style_name = n; + } + } + + /// The current dimension style (`$DIMSTYLE`) by its name: the DXF gives the name, and the + /// handle read with it needn't be the style's in the document, in which case the DWG writer + /// fell back to Standard and `AutoCAD` opened the drawing with Standard current (#1051). + fn current_dimstyle(doc: &mut acadrust::CadDocument) { + let name = doc.header.current_dimstyle_name.clone(); + if let Some(h) = doc.dim_styles.iter().find(|ds| ds.name.eq_ignore_ascii_case(&name)).map(|ds| ds.handle) { + doc.header.current_dimstyle_handle = h; + } + } + + /// The current text style (`$TEXTSTYLE`) by its name, the same way as [`current_dimstyle`]: + /// the handle read with the DXF needn't be the style's in the document, in which case the DWG + /// writer fell back to Standard and `AutoCAD` opened the drawing with Standard current (#1650). + fn current_text_style(doc: &mut acadrust::CadDocument) { + let name = doc.header.current_text_style_name.clone(); + if let Some(h) = doc.text_styles.iter().find(|ts| ts.name.eq_ignore_ascii_case(&name)).map(|ts| ts.handle) { + doc.header.current_text_style_handle = h; + } + } + + /// The DXF names of an ASCII DXF's CLASS records, in CLASSES order: a proxy's class number + /// (group 91) is 500 plus its class's place here (DXF Reference, ACAD_PROXY_OBJECT). Empty for + /// one without a CLASSES section. `None` for a binary DXF, whose classes this doesn't read: + /// CADCraft's own bridge DXF is always ASCII, and only a binary DXF given to the public + /// conversion functions comes here. + pub(super) fn dxf_class_names(dxf: &[u8]) -> Option> { + let pairs = super::proxy_transport::text_pairs(dxf)?; + let mut names = Vec::new(); + let (mut section, mut in_classes, mut in_class) = (false, false, false); + for pair in pairs { + let (code, value) = pair?; + match (code, value.as_str()) { + (0, "ENDSEC") if in_classes => break, + (0, "SECTION") => section = true, + (2, name) if section => { + in_classes = name == "CLASSES"; + section = false; + } + (0, kind) if in_classes => { + in_class = kind == "CLASS"; + // Even an unnamed record keeps its place in the source class table. + if in_class { + names.push(String::new()); + } + } + (1, value) if in_class => { + if let Some(last) = names.last_mut() { + *last = value.to_string(); + } + in_class = false; + } + _ => section = false, + } + } + Some(names) + } + + /// Proxy objects' and entities' class numbers as the DWG numbers its classes. A DXF proxy + /// names its class by its place in the DXF's CLASSES (`classes`, from 500), but acadrust's + /// reader puts its own default classes first and numbers the DXF's after them, so the DXF + /// number named another class in the DWG: a 2018 drawing's piping proxies became + /// LARGE_RADIAL_DIMENSION and SECTIONOBJECT, and the ODA File Converter refused the file. A + /// proxy whose number names no class of the DXF is left out: kept as it is, its number would + /// name some other class of the DWG, for the same reason. + /// + /// Before 2013 (AC1027) acadrust writes only the classes it knows a pre-2013 DWG needs, and a + /// proxy's class is not one of them: the proxy would name a missing class (the ODA refuses + /// the file: "No ClassId") or, by its old number, some other class. Such a DWG leaves the + /// proxies out instead, as it does other programs' data it can't write (`not_saved` reports + /// the entities). So does a DXF whose classes are unknown (`classes` is `None`, a binary + /// DXF): its proxies' numbers can't be mapped, and kept as they are they would name the + /// wrong classes and the ODA would refuse the file. Returns the proxies left out, by DXF type, for + /// what the DWG writer left out ([`super::Unwritten`]). + fn proxy_classes(doc: &mut acadrust::CadDocument, classes: Option<&[String]>) -> super::Unwritten { + let classes = match classes { + Some(c) if doc.version >= acadrust::DxfVersion::AC1027 => c, + _ => { + let entities: Vec = doc + .entities() + .filter_map(|e| match e { + acadrust::entities::EntityType::Extended(x) if matches!(x.data, acadrust::entities::ExtendedEntityData::Proxy(_)) => { + Some(x.common.handle) + } + _ => None, + }) + .collect(); + let objects: Vec = + doc.objects.iter().filter_map(|(h, o)| matches!(o, acadrust::objects::ObjectType::ProxyObject(_)).then_some(*h)).collect(); + for h in &entities { + doc.remove_entity(*h); + } + for h in &objects { + doc.objects.remove(h); + } + return left_out(objects.len(), entities.len()); + } + }; + // Name matching cannot choose between different source classes with the same name. + // Count those proxies as unwritten rather than silently changing their class identity. + let mut unique = std::collections::HashMap::new(); + for name in classes { + unique.entry(name.to_uppercase()).and_modify(|value| *value = false).or_insert(true); + } + let number = |id: i32| -> Option { + let name = classes.get(usize::try_from(id.checked_sub(500)?).ok()?)?; + if unique.get(&name.to_uppercase()) != Some(&true) { + return None; + } + doc.classes.get_by_name(name).map(|c| i32::from(c.class_number)) + }; + let objects: Vec<(acadrust::Handle, Option)> = doc + .objects + .iter() + .filter_map(|(h, o)| match o { + acadrust::objects::ObjectType::ProxyObject(p) => Some((*h, number(p.class_id))), + _ => None, + }) + .collect(); + let entities: Vec<(acadrust::Handle, Option)> = doc + .entities() + .filter_map(|e| match e { + acadrust::entities::EntityType::Extended(x) => match &x.data { + acadrust::entities::ExtendedEntityData::Proxy(p) => Some((x.common.handle, number(p.class_id))), + _ => None, + }, + _ => None, + }) + .collect(); + let unmapped = |list: &[(acadrust::Handle, Option)]| list.iter().filter(|(_, n)| n.is_none()).map(|(h, _)| *h).collect::>(); + let (gone_objects, gone_entities) = (unmapped(&objects), unmapped(&entities)); + for h in &gone_objects { + doc.objects.remove(h); + } + for h in &gone_entities { + doc.remove_entity(*h); + } + let objects = objects.into_iter().filter_map(|(h, n)| Some((h, n?))); + let entities = entities.into_iter().filter_map(|(h, n)| Some((h, n?))); + for (h, n) in objects { + if let Some(acadrust::objects::ObjectType::ProxyObject(p)) = doc.objects.get_mut(&h) { + p.class_id = n; + } + } + let entities: std::collections::HashMap = entities.into_iter().collect(); + for e in doc.entities_mut() { + if let acadrust::entities::EntityType::Extended(x) = e { + let Some(n) = entities.get(&x.common.handle).copied() else { continue }; + if let acadrust::entities::ExtendedEntityData::Proxy(p) = &mut x.data { + p.class_id = n; + } + } + } + left_out(gone_objects.len(), gone_entities.len()) + } + + /// Proxy objects and entities left out of the DWG, by their DXF types. + fn left_out(objects: usize, entities: usize) -> super::Unwritten { + [("ACAD_PROXY_OBJECT", objects), ("ACAD_PROXY_ENTITY", entities)] + .into_iter() + .filter(|(_, n)| *n > 0) + .map(|(t, n)| (t.to_string(), n)) + .collect() + } + + /// The current layer (`$CLAYER`) by its name, the same way as [`current_dimstyle`]: a DWG + /// stores the handle of the layer, and the one read with the DXF needn't be the layer's in + /// the document, so the drawing reopened with layer "0" current (#2252). + fn current_layer(doc: &mut acadrust::CadDocument) { + let name = doc.header.current_layer_name.clone(); + // The table's own lookup compares Unicode case-insensitively ("Survey_Ă„" = "Survey_ä"). + if let Some(layer) = doc.layers.get(&name) { + doc.header.current_layer_handle = layer.handle; + } + } + + /// `kept`: the version the DXF's carried-over groups were read from ([`dxf_to_dwg_kept_as`]). + fn convert(dxf: &[u8], version: Option, kept: Option) -> Result<(Vec, super::Unwritten), String> { + // Reject a pathologically large DXF before handing it to the third-party decoder, which + // allocates tables and section buffers from the input's declared sizes and cannot be + // interrupted once it starts (CWE-789). This bounds the large-input variant; the decoder + // runs inside `catch_unwind` so a decode failure is reported rather than crashing. + if dxf.len() > Limits::DEFAULT.max_dxf_bytes { + return Err(format!( + "DXF: the input is {}, larger than the {} accepted for DWG conversion", + mb(dxf.len() as u64), + mb(Limits::DEFAULT.max_dxf_bytes as u64) + )); + } + // Decoding a code page to UTF-8 can grow the text, so the limit applies again. + let data = match kept.and_then(|k| declare_version(dxf, k)) { + Some(d) if d.len() > Limits::DEFAULT.max_dxf_bytes => { + return Err(format!( + "DXF: the input is {} as UTF-8, larger than the {} accepted for DWG conversion", + mb(d.len() as u64), + mb(Limits::DEFAULT.max_dxf_bytes as u64) + )); + } + Some(d) => d, + None => dxf.to_vec(), + }; + let (classes, source) = guarded(AssertUnwindSafe(|| (dxf_class_names(&data), super::proxy_transport::source_classes(&data)))) + .map_err(|_| "the DXF proxy class conversion failed".to_string())?; + let mut doc = std::panic::catch_unwind(move || acadrust::DxfReader::from_reader(Cursor::new(data)).and_then(acadrust::DxfReader::read)) + .map_err(|_| "the DXF→DWG conversion failed".to_string())? + .map_err(|e| format!("DXF: {e}"))?; + match version { + Some(v) => doc.version = v, + // Without a version asked for, write at least AutoCAD 2004 (AC1018) DWG: R2000 DWG + // has no true colours, transparency, table styles or gradients. The DXF stays R2000; + // only the DWG is newer. + None if doc.version < MIN_DWG_VERSION => doc.version = MIN_DWG_VERSION, + None => {} + } + // The fix-ups run on the third-party reader's document, so a panic in one is an error, not a crash. + let unwritten = guarded(AssertUnwindSafe(|| { + super::fail_in_test("convert"); + // From 2013 the DWG takes the source classes as they are, duplicates and metadata + // included; otherwise (before 2013, a binary DXF, classes that can't be written back) + // proxies are matched to classes by name or left out. + let restored = match source { + Some(source) if doc.version >= acadrust::DxfVersion::AC1027 => super::proxy_transport::restore_classes(&mut doc, source), + _ => None, + }; + let proxies = match restored { + Some((objects, entities)) => left_out(objects, entities), + None => proxy_classes(&mut doc, classes.as_deref()), + }; + other_layout_owners(&mut doc); + table_style_rows(&mut doc, true); + dimension_flags(&mut doc); + let mut unwritten = drop_unwritable(&mut doc); + for (kind, n) in proxies { + let total = unwritten.entry(kind).or_insert(0); + *total = total.saturating_add(n); + } + current_dimstyle(&mut doc); + add_variable_dictionary(&mut doc); + current_text_style(&mut doc); + current_layer(&mut doc); + unwritten + })) + .map_err(|()| "the DXF to DWG conversion failed on this file".to_string())?; + let dwg = std::panic::catch_unwind(AssertUnwindSafe(|| acadrust::DwgWriter::write_to_vec(&doc))) + .map_err(|_| "the DWG writer failed on this drawing".to_string())? + .map_err(|e| format!("DWG write: {e}"))?; + Ok((dwg, unwritten)) + } + + /// `AutoCAD`'s entry name for the drawing variables dictionary (the DXF Reference calls it the + /// ACDBVARIABLEDICTIONARY; dictionary keys are case-insensitive). + const VARIABLE_DICTIONARY: &str = "AcDbVariableDictionary"; + + /// A style or scale name, or `default` when the drawing has none. + fn name_or_default(name: &str, default: &str) -> String { + let name = name.trim(); + if name.is_empty() { default.to_string() } else { name.to_string() } + } + + /// A drawing variable's value as text, or `default` when the header holds a value the + /// variable doesn't accept (a hostile file can put any integer in it). + fn variable_value(value: i16, valid: &[i16], default: i16) -> String { + if valid.contains(&value) { value.to_string() } else { default.to_string() } + } + + /// Add the drawing variables dictionary `AutoCAD` always writes into the named object + /// dictionary, with the variables it keeps there, when the drawing has none (#1813). + /// Vectric Aspire 10.5 reads a 2004 (AC1018) DWG without it as empty ("No data was imported + /// from the DXF / DWG file"), though it reads the same drawing as DXF — which has no such + /// dictionary either, as CADCraft's and ezdxf's writers don't make one. `AutoCAD`'s AC1032 DWGs + /// read there without it, so newer output is left as it was, and a dictionary the drawing + /// does have keeps its own entries and values. + fn add_variable_dictionary(doc: &mut acadrust::CadDocument) { + use acadrust::objects::{Dictionary, DictionaryVariable, ObjectType}; + if doc.version >= acadrust::DxfVersion::AC1032 { + return; + } + let root_handle = doc.header.named_objects_dict_handle; + match doc.objects.get(&root_handle) { + Some(ObjectType::Dictionary(root)) => { + if root.get(VARIABLE_DICTIONARY).is_some() { + return; + } + } + _ => return, + } + // The variables AutoCAD writes for a drawing with default styles and a 1:1 annotation + // scale. DIMASSOC and HIDETEXT keep the drawing's own header values; LAYEREVAL and + // LAYERNOTIFY have no DXF header variable. + let variables = [ + ("CMLEADERSTYLE", name_or_default(&doc.header.current_mleader_style_name, "Standard")), + ("CTABLESTYLE", name_or_default(&doc.header.current_table_style_name, "Standard")), + ("CANNOSCALE", name_or_default(&doc.header.current_annotation_scale, "1:1")), + ("DIMASSOC", variable_value(doc.header.dimension_associativity, &[0, 1, 2], 2)), + ("HIDETEXT", variable_value(doc.header.hide_text, &[0, 1], 1)), + ("LAYEREVAL", "0".to_string()), + ("LAYERNOTIFY", "0".to_string()), + ]; + let dictionary_handle = doc.allocate_handle(); + let mut dictionary = Dictionary::new(); + dictionary.handle = dictionary_handle; + dictionary.owner = root_handle; + for (name, value) in variables { + let handle = doc.allocate_handle(); + let mut variable = DictionaryVariable::new(name, &value); + variable.handle = handle; + variable.owner_handle = dictionary_handle; + dictionary.add_entry(name, handle); + doc.objects.insert(handle, ObjectType::DictionaryVariable(variable)); + } + doc.objects.insert(dictionary_handle, ObjectType::Dictionary(dictionary)); + if let Some(ObjectType::Dictionary(root)) = doc.objects.get_mut(&root_handle) { + root.add_entry(VARIABLE_DICTIONARY, dictionary_handle); + } + } + + /// Entities of a layout other than the one in `*Paper_Space` are written owned by their block + /// (entity mode 0, with its handle). acadrust writes every paper-space entity with mode 1, + /// "paper space", which `AutoCAD` reads as the `*Paper_Space` block's: everything drawn on a + /// second layout landed on the first one, and the second stayed empty (#1057). + fn other_layout_owners(doc: &mut acadrust::CadDocument) { + let others: std::collections::HashSet = doc + .block_records + .iter() + .filter(|br| { + let n = br.name.to_ascii_uppercase(); + n.len() > 12 && n.starts_with("*PAPER_SPACE") && n[12..].bytes().all(|b| b.is_ascii_digit()) + }) + .map(|br| br.handle) + .collect(); + if others.is_empty() { + return; + } + for e in doc.entities_mut() { + let c = e.common_mut(); + if others.contains(&c.owner_handle) { + c.entity_mode = Some(0); + } + } + } + + #[cfg(test)] + mod post_tests { + use super::*; + use acadrust::entities::{CellRange, EntityType, Table}; + + fn doc_with_table(rows: usize, cols: usize, ranges: usize) -> acadrust::CadDocument { + let mut t = Table::new(acadrust::types::Vector3::new(0.0, 0.0, 0.0), rows, cols); + t.merged_ranges = (0..ranges).map(|_| CellRange::new(0, 0, rows - 1, cols - 1)).collect(); + let mut doc = acadrust::CadDocument::new(); + assert!(doc.add_entity(EntityType::Table(Box::new(t))).is_ok()); + doc + } + + /// Overlapping whole-table ranges can't make the merge pass visit more than its budget (#1934). + #[test] + fn merge_work_is_bounded_by_the_cell_budget() { + let mut doc = doc_with_table(10, 10, 50); + assert_eq!(table_merges_within(&mut doc, 250), 200, "two whole-table ranges fit, the rest are skipped"); + let mut doc = doc_with_table(10, 10, 1); + assert_eq!(table_merges_within(&mut doc, usize::MAX), 100); + } + + /// A height marked only in the style's override flags (as newer files store it) is kept, + /// and the same height with no flag at all is cleared (review on #2588). Tested on the + /// read document because acadrust's DWG writer doesn't write the override flags. + #[test] + fn a_style_override_flag_keeps_the_cell_height() { + use acadrust::entities::table::{CellStyle, CellStylePropertyFlags}; + let height = |override_flags: i32| { + let mut doc = doc_with_table(1, 1, 0); + for e in doc.entities_mut() { + if let EntityType::Table(t) = e { + for cell in t.rows.iter_mut().flat_map(|r| r.cells.iter_mut()) { + let mut s = CellStyle::new(); + s.text_height = 0.5; + s.override_flags = override_flags; + cell.style = Some(s); + } + } + } + table_cell_heights(&mut doc); + doc.entities().find_map(|e| match e { + EntityType::Table(t) => t.rows.first()?.cells.first()?.style.as_ref().map(|s| s.text_height), + _ => None, + }) + }; + assert_eq!(height(CellStylePropertyFlags::TEXT_HEIGHT.bits() as i32), Some(0.5)); + assert_eq!(height(0), Some(0.0)); + } + + /// A panic in the post-read fix-ups is an error, not a crash (#1934). + #[test] + fn a_panic_in_post_processing_is_an_error() { + #[allow(clippy::panic)] + let r = guarded(AssertUnwindSafe(|| -> usize { panic!("boom") })); + assert!(r.is_err()); + assert_eq!(guarded(|| 3), Ok(3)); + } + } + + #[cfg(test)] + mod application_object_tests { + use super::*; + use acadrust::CadDocument; + use acadrust::classes::DxfClass; + use acadrust::objects::{ObjectType, ProxyObjectReference, ProxyPayload, ProxyReferenceKind, RegisteredClassObject}; + use acadrust::types::{DxfVersion, Handle}; + + /// An object of another application's class that acadrust keeps only as its record (a + /// vertical product's dimension data, say) is carried as that class's proxy object, with its + /// data bits, references, owner and extension dictionary, through DXF and back to DWG, twice. + #[test] + fn undecoded_application_object_is_kept_as_its_proxy() { + for version in [DxfVersion::AC1027, DxfVersion::AC1032] { + let mut source = CadDocument::new(); + source.version = version; + source.classes.add_or_update(DxfClass::new("TESTCONTAINER", "xyzDbTestContainer")); + let handle = Handle::from(0x2Bu64); + let target = Handle::from(0x2Cu64); + let payload = ProxyPayload::from_bits(&[0x5A, 0x3C, 0xC0], 18); + // A null pointer between two references: it keeps its place, the one after it is not lost. + let references = vec![ + ProxyObjectReference { handle: target, kind: ProxyReferenceKind::HardPointer }, + ProxyObjectReference { handle: Handle::from(0u64), kind: ProxyReferenceKind::SoftPointer }, + ProxyObjectReference { handle: Handle::from(0x0Cu64), kind: ProxyReferenceKind::HardPointer }, + ]; + source.objects.insert( + handle, + ObjectType::RegisteredClass(RegisteredClassObject { + handle, + owner: Handle::from(0x0Cu64), + dxf_name: "TESTCONTAINER".into(), + cpp_class_name: "xyzDbTestContainer".into(), + payload: payload.clone(), + object_ids: references.clone(), + ..Default::default() + }), + ); + let mut dwg = acadrust::DwgWriter::write_to_vec(&source).unwrap(); + // acadrust has no model of it: read back, it is only its record. + let decoded = acadrust::DwgReader::from_stream(Cursor::new(dwg.clone())).read().unwrap(); + assert!(matches!(decoded.objects.get(&handle), Some(ObjectType::Unknown { .. })), "{:?}", decoded.objects.get(&handle)); + for generation in 0..2 { + let (dxf, notes) = dwg_to_dxf_noted(&dwg, Limits::DEFAULT).unwrap(); + assert!(notes.is_empty(), "{version:?} {generation}: {notes:?}"); + // The DXF has the object, by its handle, as a proxy record. + let text = String::from_utf8_lossy(&dxf); + let lines: Vec<&str> = text.lines().map(str::trim).collect(); + let record = lines.windows(4).any(|w| w == ["0", "ACAD_PROXY_OBJECT", "5", "2B"]); + assert!(record, "{version:?} {generation}: no proxy record for the object"); + dwg = dxf_to_dwg_as(&dxf, "AC1032").unwrap(); + let doc = acadrust::DwgReader::from_stream(Cursor::new(dwg.clone())).read().unwrap(); + let Some(ObjectType::ProxyObject(proxy)) = doc.objects.get(&handle) else { panic!("{version:?} {generation}: proxy object") }; + assert_eq!(proxy.payload.bit_count, payload.bit_count); + assert_eq!(proxy.payload.data(), payload.data()); + assert_eq!(proxy.object_ids, references); + let class = doc.classes.iter().find(|c| i32::from(c.class_number) == proxy.class_id).unwrap(); + assert_eq!(class.dxf_name, "TESTCONTAINER"); + assert!(!class.is_an_entity); + } + } + } + + /// The database's own classes are not turned into another application's proxies; what is + /// left without a DXF form is said, not dropped silently. + #[test] + fn database_objects_without_a_form_are_reported() { + let mut doc = CadDocument::new(); + doc.classes.add_or_update(DxfClass::new("BLOCKPROPERTIESTABLE", "AcDbBlockPropertiesTable")); + let handle = Handle::from(0x40u64); + doc.objects.insert( + handle, + ObjectType::Unknown { + type_name: "BLOCKPROPERTIESTABLE".into(), + handle, + owner: Handle::from(0x0Cu64), + raw_dxf_codes: None, + raw_dwg_data: Some(vec![0; 8]), + raw_dwg_handle_bits: 0, + raw_dwg_version: Some(DxfVersion::AC1032), + }, + ); + let notes = application_objects(&mut doc); + assert!(matches!(doc.objects.get(&handle), Some(ObjectType::Unknown { .. }))); + assert_eq!( + notes, + [ + "DWG: 1 objects of types this bridge can't preserve were left out (BLOCKPROPERTIESTABLE Ă— 1); saving over the original file would lose them" + ] + ); + } + + /// A record that doesn't split stays as it was and is named, never guessed at. + #[test] + fn invalid_object_records_are_rejected() { + for handle_bits in [-1, 1000] { + let raw = acadrust::entities::RawRecord { data: vec![0; 8], handle_bits, version: DxfVersion::AC1032 }; + assert!(object_record(&raw, 42).is_err()); + } + let mut doc = CadDocument::new(); + doc.classes.add_or_update(DxfClass::new("TESTCONTAINER", "xyzDbTestContainer")); + let handle = Handle::from(0x41u64); + doc.objects.insert( + handle, + ObjectType::Unknown { + type_name: "TESTCONTAINER".into(), + handle, + owner: Handle::from(0x0Cu64), + raw_dxf_codes: None, + raw_dwg_data: Some(vec![0xFF; 4]), + raw_dwg_handle_bits: 1000, + raw_dwg_version: Some(DxfVersion::AC1032), + }, + ); + let notes = application_objects(&mut doc); + assert!(matches!(doc.objects.get(&handle), Some(ObjectType::Unknown { .. }))); + assert!(notes.iter().any(|n| n.contains("(TESTCONTAINER Ă— 1)")), "{notes:?}"); + } + } +} + +#[cfg(not(no_dwg_bridge))] +pub use native::{ + dwg_to_dxf, dwg_to_dxf_noted, dwg_to_dxf_with, dxf_to_dwg, dxf_to_dwg_as, dxf_to_dwg_kept_as, dxf_to_dwg_kept_noted, dxf_to_dwg_noted, +}; + +/// What saving as DWG left out: how many records of each DXF type (upper case) the DWG writer +/// can't write. +pub type Unwritten = std::collections::BTreeMap; + +// Test-only: the stage ("post_process" or "convert") whose guard the current test checks. The +// stage then panics on entry, so a test through the public API shows whether it is guarded. +#[cfg(all(test, not(no_dwg_bridge)))] +thread_local! { + static PANIC_IN: std::cell::Cell> = const { std::cell::Cell::new(None) }; +} + +/// Panics when a test asked for `stage` to fail (see [`PANIC_IN`]); does nothing otherwise. +#[cfg(not(no_dwg_bridge))] +#[allow(clippy::panic)] +fn fail_in_test(stage: &str) { + #[cfg(not(test))] + let _ = stage; + #[cfg(test)] + assert!(PANIC_IN.with(std::cell::Cell::get) != Some(stage), "test-forced panic in {stage}"); +} + +/// The error for a DWG version this writer can't create. +#[cfg_attr(no_dwg_bridge, allow(dead_code))] +fn unsupported(version: &str) -> String { + let list: Vec<&str> = VERSIONS.iter().map(|(_, year)| *year).collect(); + format!("DWG version `{version}` can't be written (this bridge writes {})", list.join(", ")) +} + +/// Which build has no DWG bridge, for the stubs' errors. +#[cfg(no_dwg_bridge)] +const NO_BRIDGE_BUILD: &str = if cfg!(target_arch = "wasm32") { "web" } else { "FreeBSD aarch64" }; + +/// There's no worker without a bridge. +#[cfg(no_dwg_bridge)] +pub const WORKER_ARG: &str = "--dwg-worker"; +#[cfg(no_dwg_bridge)] +pub fn serve_if_asked() -> Option { + None +} +#[cfg(no_dwg_bridge)] +pub fn use_own_worker() {} +#[cfg(no_dwg_bridge)] +pub fn allow_in_process() {} +#[cfg(no_dwg_bridge)] +pub fn decode(bytes: &[u8], limits: Limits) -> Result<(Vec, Vec), String> { + dwg_to_dxf_noted(bytes, limits) +} +#[cfg(no_dwg_bridge)] +pub fn encode_as(dxf: &[u8], version: &str) -> Result, String> { + dxf_to_dwg_as(dxf, version) +} +#[cfg(no_dwg_bridge)] +pub fn encode_noted(dxf: &[u8], version: &str) -> Result<(Vec, Unwritten), String> { + dxf_to_dwg_noted(dxf, version) +} +#[cfg(no_dwg_bridge)] +pub fn encode_kept_noted(dxf: &[u8], version: &str, kept: Option<&str>) -> Result<(Vec, Unwritten), String> { + match kept { + Some(kept) => dxf_to_dwg_kept_noted(dxf, version, kept), + None => dxf_to_dwg_noted(dxf, version), + } +} + +#[cfg(no_dwg_bridge)] +pub fn dwg_to_dxf(_bytes: &[u8]) -> Result, String> { + Err(format!("DWG files can't be opened in the {NO_BRIDGE_BUILD} build yet; save as DXF")) +} +#[cfg(no_dwg_bridge)] +pub fn dwg_to_dxf_noted(_bytes: &[u8], _limits: Limits) -> Result<(Vec, Vec), String> { + Err(format!("DWG files can't be opened in the {NO_BRIDGE_BUILD} build yet; save as DXF")) +} +#[cfg(no_dwg_bridge)] +pub fn dwg_to_dxf_with(_bytes: &[u8], _limits: Limits) -> Result, String> { + Err(format!("DWG files can't be opened in the {NO_BRIDGE_BUILD} build yet; save as DXF")) +} +#[cfg(no_dwg_bridge)] +pub fn dxf_to_dwg(_dxf: &[u8]) -> Result, String> { + Err(format!("DWG files can't be written in the {NO_BRIDGE_BUILD} build yet; save as DXF")) +} +#[cfg(no_dwg_bridge)] +pub fn dxf_to_dwg_as(_dxf: &[u8], _version: &str) -> Result, String> { + Err(format!("DWG files can't be written in the {NO_BRIDGE_BUILD} build yet; save as DXF")) +} +#[cfg(no_dwg_bridge)] +pub fn dxf_to_dwg_noted(_dxf: &[u8], _version: &str) -> Result<(Vec, Unwritten), String> { + Err(format!("DWG files can't be written in the {NO_BRIDGE_BUILD} build yet; save as DXF")) +} +#[cfg(no_dwg_bridge)] +pub fn dxf_to_dwg_kept_as(_dxf: &[u8], _version: &str, _kept: &str) -> Result, String> { + Err(format!("DWG files can't be written in the {NO_BRIDGE_BUILD} build yet; save as DXF")) +} +#[cfg(no_dwg_bridge)] +pub fn dxf_to_dwg_kept_noted(_dxf: &[u8], _version: &str, _kept: &str) -> Result<(Vec, Unwritten), String> { + Err(format!("DWG files can't be written in the {NO_BRIDGE_BUILD} build yet; save as DXF")) +} + +#[cfg(test)] +mod tests { + use std::fmt::Write as _; + + use super::*; + + /// The DXF of an empty drawing, enough for the bridge to make a small DWG. + #[cfg(not(no_dwg_bridge))] + const MINIMAL_DXF: &str = " 0\nSECTION\n 2\nHEADER\n 9\n$ACADVER\n 1\nAC1032\n 0\nENDSEC\n 0\nEOF\n"; + + /// Makes `stage` panic on the current thread until dropped. + #[cfg(not(no_dwg_bridge))] + struct PanicIn; + + #[cfg(not(no_dwg_bridge))] + impl PanicIn { + fn at(stage: &'static str) -> Self { + PANIC_IN.with(|p| p.set(Some(stage))); + Self + } + } + + #[cfg(not(no_dwg_bridge))] + impl Drop for PanicIn { + fn drop(&mut self) { + PANIC_IN.with(|p| p.set(None)); + } + } + + /// The public read reports a panic in the post-read fix-ups as an error: this fails if the + /// `guarded` call around `post_process` is taken out (#2402). + #[cfg(not(no_dwg_bridge))] + #[test] + fn reading_a_dwg_survives_a_panic_in_post_processing() { + let dwg = dxf_to_dwg(MINIMAL_DXF.as_bytes()).unwrap(); + assert!(dwg_to_dxf(&dwg).is_ok(), "the DWG reads without the forced panic"); + let _panic = PanicIn::at("post_process"); + let err = dwg_to_dxf(&dwg).unwrap_err(); + assert!(err.contains("the DWG post-processing failed on this file"), "{err}"); + } + + /// The DXF to DWG conversion reports a panic in its fix-ups as an error, not a crash (#2402). + #[cfg(not(no_dwg_bridge))] + #[test] + fn converting_to_dwg_survives_a_panic_in_the_fix_ups() { + let _panic = PanicIn::at("convert"); + let err = dxf_to_dwg(MINIMAL_DXF.as_bytes()).unwrap_err(); + assert!(err.contains("the DXF to DWG conversion failed on this file"), "{err}"); + } + + #[test] + fn detects_magic() { + assert!(is_dwg(b"AC1032\0\0\0\0")); + assert!(!is_dwg(b" 0\r\nSECTION")); + assert_eq!(version(b"AC1018xxxx").as_deref(), Some("AC1018")); + } + + /// A DXF with ROMANS as `$TEXTSTYLE` gives a DWG whose current text style is ROMANS, not + /// Standard: the DXF names the style, and the handle read with it was Standard's (#1650). + #[cfg(not(no_dwg_bridge))] + #[test] + fn current_text_style_reaches_the_dwg() { + let header = " 0\nSECTION\n 2\nHEADER\n 9\n$ACADVER\n 1\nAC1032\n 9\n$TEXTSTYLE\n 7\nROMANS\n 0\nENDSEC\n"; + let style = " 0\nSTYLE\n 5\n1F0\n100\nAcDbSymbolTableRecord\n100\nAcDbTextStyleTableRecord\n 2\nROMANS\n 70\n0\n"; + let style = format!("{style} 40\n0\n 41\n1\n 50\n0\n 71\n0\n 42\n2.5\n 3\nromans.shx\n 4\n\n"); + let dxf = format!("{header} 0\nSECTION\n 2\nTABLES\n 0\nTABLE\n 2\nSTYLE\n 70\n1\n{style} 0\nENDTAB\n 0\nENDSEC\n 0\nEOF\n"); + let dwg = dxf_to_dwg(dxf.as_bytes()).unwrap(); + let doc = acadrust::DwgReader::from_stream(std::io::Cursor::new(&dwg)).read().unwrap(); + let romans = doc.text_styles.iter().find(|ts| ts.name == "ROMANS").map(|ts| ts.handle); + assert!(romans.is_some(), "ROMANS is missing from the DWG"); + assert_eq!(Some(doc.header.current_text_style_handle), romans); + } + + #[test] + fn garbage_is_an_error_not_a_crash() { + assert!(dwg_to_dxf(b"AC1032 this is not a dwg file at all").is_err()); + assert!(dwg_to_dxf(b"").is_err()); + } + + #[cfg(not(no_dwg_bridge))] + #[test] + fn dxf_to_dwg_rejects_garbage_without_crashing() { + // The DXF→DWG path runs the third-party reader inside catch_unwind and reports failures + // rather than crashing (and rejects oversized input before decoding). + assert!(dxf_to_dwg(b"not a dxf file").is_err()); + assert!(dxf_to_dwg(b"").is_err()); + } + + /// A DXF with one DIMENSION per (text, group 70, subclass, points) row. + #[cfg(not(no_dwg_bridge))] + type DimensionSpec<'a> = (&'a str, i64, &'a str, &'a [(i32, [f64; 2])]); + + #[cfg(not(no_dwg_bridge))] + fn dims_dxf(dims: &[DimensionSpec<'_>]) -> String { + let mut s = String::from(" 0\nSECTION\n 2\nHEADER\n 9\n$ACADVER\n 1\nAC1032\n 0\nENDSEC\n 0\nSECTION\n 2\nENTITIES\n"); + for (text, flags, sub, pts) in dims { + s += " 0\nDIMENSION\n 8\n0\n100\nAcDbEntity\n100\nAcDbDimension\n 10\n0\n 20\n0\n 30\n0\n"; + let _ = write!(s, " 11\n50\n 21\n-20\n 31\n0\n 70\n{flags}\n 1\n{text}\n 3\nStandard\n100\n{sub}\n"); + for (code, [x, y]) in *pts { + let _ = write!(s, "{code:>3}\n{x}\n{:>3}\n{y}\n{:>3}\n0\n", code + 10, code + 20); + } + } + s + " 0\nENDSEC\n 0\nEOF\n" + } + + /// DWG stores DXF group 70's bit 128 inverted, in bit 0 of the dimension flags byte (set: + /// the text is at its default position), and bit 32 in bit 1 (ODA DWG specification, common + /// dimension data). The bytes are read back raw, so a reader that inverts bit 0 the same way + /// as the writer can't hide a wrong byte. + #[cfg(not(no_dwg_bridge))] + #[test] + fn dimension_text_position_flags_reach_the_dwg_as_autocad_reads_them() { + let linear: &[(i32, [f64; 2])] = &[(13, [0.0, 0.0]), (14, [100.0, 0.0])]; + let radius: &[(i32, [f64; 2])] = &[(15, [10.0, 0.0])]; + let angular: &[(i32, [f64; 2])] = &[(13, [0.0, 0.0]), (14, [10.0, 0.0]), (15, [0.0, 0.0]), (16, [0.0, 10.0])]; + let kinds = [ + (0, "AcDbAlignedDimension", linear), + (1, "AcDbAlignedDimension", linear), + (4, "AcDbRadialDimension", radius), + (2, "AcDb2LineAngularDimension", angular), + ]; + let mut dims = Vec::new(); + for (ty, sub, pts) in kinds { + dims.push(("HOME", ty | 32, sub, pts)); + dims.push(("MOVED", ty | 32 | 128, sub, pts)); + } + let dwg = dxf_to_dwg(dims_dxf(&dims).as_bytes()).unwrap(); + let doc = acadrust::DwgReader::from_stream(std::io::Cursor::new(&dwg)).read().unwrap(); + let mut seen = Vec::new(); + for e in doc.entities() { + let acadrust::entities::EntityType::Dimension(d) = e else { continue }; + let (text, byte) = (d.base().text_override().unwrap_or_default(), d.base().dwg_flags_byte); + seen.push((text.to_string(), byte & 0x03)); + assert_eq!(byte & 0x02, 0x02, "{text} {:?}: bit 32 of group 70 is lost (flags byte {byte:#04x})", d.base().dimension_type); + let default_pos = text == "HOME"; + assert_eq!(byte & 0x01 != 0, default_pos, "{text} {:?}: flags byte {byte:#04x}", d.base().dimension_type); + } + for text in ["HOME", "MOVED"] { + assert_eq!(seen.iter().filter(|(t, _)| t == text).count(), 4, "{seen:?}"); + } + + // And CADCraft reads its own DWG back as it wrote it: 128 only on the moved texts. + let dxf = String::from_utf8(dwg_to_dxf(&dwg).unwrap()).unwrap(); + let lines: Vec<&str> = dxf.lines().map(str::trim).collect(); + let mut flags = Vec::new(); + for rec in lines.chunks(2).collect::>().split(|p| p.first() == Some(&"0")) { + let value = |code: &str| rec.iter().find(|p| p.first() == Some(&code)).and_then(|p| p.get(1).copied()); + if let (Some(text @ ("HOME" | "MOVED")), Some(v)) = (value("1"), value("70")) { + flags.push((text, v.parse::().unwrap_or(-1) & 128)); + } + } + assert_eq!(flags.len(), 8, "{flags:?}"); + assert!(flags.iter().all(|(l, f)| (*f == 128) == (*l == "MOVED")), "{flags:?}"); + } + + /// The handles a DXF text refers to or gives its records (groups 5, 105 and 330–369). + #[cfg(not(no_dwg_bridge))] + fn dxf_handles(dxf: &str) -> std::collections::HashSet { + let lines: Vec<&str> = dxf.lines().collect(); + lines + .chunks(2) + .filter_map(|p| match p { + [c, v] if matches!(c.trim().parse::(), Ok(5 | 105 | 330..=369)) => u64::from_str_radix(v.trim(), 16).ok(), + _ => None, + }) + .collect() + } + + /// Issue #1482: a DWG that once came from `MicroStation` can carry hundreds of thousands of DGN + /// line style components no definition uses (a real 83 MB site plan: 888,483 of 889,053, a + /// 1.2 GB DXF rendition over the limit, so it didn't open). They're left out with a note, and + /// their dictionary entries with them; the definition and every component it reaches — + /// through compound and point components, around a cycle — are kept. + #[cfg(not(no_dwg_bridge))] + #[test] + #[expect( + clippy::too_many_lines, + reason = "This integration test builds and verifies one complete DGN component graph; splitting the fixture would obscure reachability." + )] + fn orphaned_dgn_line_style_components_are_left_out() { + use acadrust::objects::*; + let mut doc = acadrust::CadDocument::new(); + let container = doc.allocate_handle(); + let nod = doc.header.named_objects_dict_handle; + let Some(ObjectType::Dictionary(root)) = doc.objects.get_mut(&nod) else { panic!("no named object dictionary") }; + root.entries.push(("ACAD_DGNLINESTYLECOMP".into(), container)); + let pattern = || DgnLsStrokePattern { + has_iteration_limit: false, + is_single_segment: false, + iteration_limit: 0, + auto_phase: 0.0, + phase: 0.0, + phase_mode: DgnLsPhaseMode::Distance, + strokes: vec![DgnLsStroke { + is_dash: true, + bypass_corner: false, + can_be_scaled: true, + invert_at_origin: false, + invert_at_end: false, + length: 2.5, + start_width: 0.0, + end_width: 0.0, + width_mode: 0, + cap_mode: 0, + }], + }; + let kind = |c: &DgnLsComponentData| match c { + DgnLsComponentData::Symbol(_) => DgnLsComponentType::Symbol, + DgnLsComponentData::Compound(_) => DgnLsComponentType::Compound, + DgnLsComponentData::Stroke(_) => DgnLsComponentType::Stroke, + DgnLsComponentData::Point(_) => DgnLsComponentType::Point, + DgnLsComponentData::Internal(_) => DgnLsComponentType::Internal, + }; + let compound = |of: &[acadrust::Handle]| { + DgnLsComponentData::Compound(DgnLsCompoundComponent { + entries: of.iter().map(|&component| DgnLsCompoundEntry { component, offset: 0.0 }).collect(), + }) + }; + let mut entries = Vec::new(); + let mut put = |doc: &mut acadrust::CadDocument, h: acadrust::Handle, data: DgnLineStyleData| { + entries.push((format!("{:X}", h.value()), h)); + let reactors = vec![container]; + doc.objects + .insert(h, ObjectType::DgnLineStyle(DgnLineStyleObject { handle: h, owner: container, reactors, xdictionary_handle: None, data })); + }; + let component = |c: DgnLsComponentData| DgnLineStyleData::Component { + kind: kind(&c), + description: "dash".into(), + version: 1, + component_uid: [7; 16], + scale: 1.0, + property_flags: 0, + component: c, + properties: vec![], + }; + let h: Vec = (0..9).map(|_| doc.allocate_handle()).collect(); + let [def, root, stroke, point, back, orphan, orphan_cycle, internal, shared] = h[..] else { panic!("nine handles") }; + put( + &mut doc, + def, + DgnLineStyleData::Definition { + description: "{ dash }".into(), + version: 1, + style_number: 1, + component_uid: [1; 16], + is_continuous: false, + unit_definition: 1.0, + unit_scale: 1.0, + units_type: 0, + is_element: false, + is_physical: false, + is_scale_independent: false, + is_snappable: false, + root_component: root, + properties: vec![], + }, + ); + // Used: def → root → {stroke, point → shared, back → root (a cycle)}. + put(&mut doc, root, component(compound(&[stroke, point, back]))); + put(&mut doc, stroke, component(DgnLsComponentData::Stroke(pattern()))); + put(&mut doc, point, component(DgnLsComponentData::Point(DgnLsPointComponent { stroke_component: shared, symbols: vec![] }))); + put(&mut doc, back, component(compound(&[root]))); + put(&mut doc, shared, component(DgnLsComponentData::Stroke(pattern()))); + // Orphans: an internal component, and two compounds that refer to each other and to a + // used component (which stays). + let pattern_internal = + DgnLsInternalComponent { pattern: pattern(), internal_version: 0, hardware_style: 0, is_hardware_style: false, line_code: 0 }; + put(&mut doc, internal, component(DgnLsComponentData::Internal(pattern_internal))); + put(&mut doc, orphan, component(compound(&[orphan_cycle, shared]))); + put(&mut doc, orphan_cycle, component(compound(&[orphan]))); + let mut dict = Dictionary::new(); + dict.handle = container; + dict.owner = nod; + dict.entries = entries; + doc.objects.insert(container, ObjectType::Dictionary(dict)); + let dwg = acadrust::DwgWriter::write_to_vec(&doc).unwrap(); + + let (dxf, notes) = dwg_to_dxf_noted(&dwg, Limits::DEFAULT).unwrap(); + let dxf = String::from_utf8(dxf).unwrap(); + let handles = dxf_handles(&dxf); + for kept in [def, root, stroke, point, back, shared] { + assert!(handles.contains(&kept.value()), "{:X} was left out", kept.value()); + } + for gone in [internal, orphan, orphan_cycle] { + assert!(!handles.contains(&gone.value()), "{:X} was kept or is still referred to", gone.value()); + } + let lines: Vec<&str> = dxf.lines().collect(); + let ls: Vec<&str> = lines.chunks(2).filter_map(|p| p.get(1).filter(|_| p[0].trim() == "0" && p[1].starts_with("LS")).copied()).collect(); + assert_eq!(ls.iter().filter(|t| t.ends_with("COMPONENT")).count(), 5, "{ls:?}"); + assert!(ls.contains(&"LSDEFINITION") && !ls.contains(&"LSINTERNALCOMPONENT"), "{ls:?}"); + assert_eq!(notes, ["DWG: 3 DGN line style components that no line style definition uses were left out"]); + } + + /// An R2000 DXF header (`\r\n` lines, as CADCraft writes them) with `extra` after `$ACADVER`. + #[cfg(not(no_dwg_bridge))] + fn r2000(extra: &[u8]) -> Vec { + let mut dxf = b" 0\r\nSECTION\r\n 2\r\nHEADER\r\n 9\r\n$ACADVER\r\n 1\r\nAC1015\r\n 9\r\n$DWGCODEPAGE\r\n 3\r\nANSI_1252\r\n".to_vec(); + dxf.extend_from_slice(b" 0\r\nENDSEC\r\n 0\r\nSECTION\r\n 2\r\nENTITIES\r\n 0\r\nTEXT\r\n 1\r\n"); + dxf.extend_from_slice(extra); + dxf.extend_from_slice(b"\r\n 0\r\nENDSEC\r\n 0\r\nEOF\r\n"); + dxf + } + + /// The kept version replaces only the `$ACADVER` value; everything else stays byte for byte. + #[cfg(not(no_dwg_bridge))] + #[test] + fn declare_version_changes_only_acadver() { + let dxf = r2000(b"A\\U+00E7\\U+00E3o"); + let out = native::declare_version(&dxf, acadrust::DxfVersion::AC1032).unwrap(); + assert_eq!(String::from_utf8(out).unwrap(), String::from_utf8(dxf).unwrap().replacen("AC1015", "AC1032", 1)); + } + + /// A kept version not newer than the declared one, a binary DXF and a DXF without `$ACADVER` + /// are read as they are. + #[cfg(not(no_dwg_bridge))] + #[test] + fn declare_version_leaves_what_it_need_not_change() { + let dxf = r2000(b"x"); + assert_eq!(native::declare_version(&dxf, acadrust::DxfVersion::AC1015), None); + assert_eq!(native::declare_version(&dxf, acadrust::DxfVersion::AC1009), None); + assert_eq!(native::declare_version(b"AutoCAD Binary DXF\r\n\x1a\0", acadrust::DxfVersion::AC1032), None); + let no_version = b" 0\r\nSECTION\r\n 2\r\nHEADER\r\n 0\r\nENDSEC\r\n 0\r\nEOF\r\n"; + assert_eq!(native::declare_version(no_version, acadrust::DxfVersion::AC1032), None); + // A `$ACADVER` outside the HEADER section is not the file's version. + let late = b" 0\r\nSECTION\r\n 2\r\nHEADER\r\n 0\r\nENDSEC\r\n 9\r\n$ACADVER\r\n 1\r\nAC1015\r\n"; + assert_eq!(native::declare_version(late, acadrust::DxfVersion::AC1032), None); + } + + /// The header is read as group code and value pairs up to the first `0`/`ENDSEC`: a `$ACADVER` + /// name whose value would come after it, or a file that never ends its HEADER, is no version + /// to replace, and `\r\n` lines keep the value's offset right. + #[cfg(not(no_dwg_bridge))] + #[test] + fn declare_version_reads_the_header_in_pairs_up_to_endsec() { + let cut = b" 0\r\nSECTION\r\n 2\r\nHEADER\r\n 9\r\n$ACADVER\r\n 0\r\nENDSEC\r\n 1\r\nAC1015\r\n"; + assert_eq!(native::declare_version(cut, acadrust::DxfVersion::AC1032), None); + let endless = vec![b'\n'; 1 << 20]; + assert_eq!(native::declare_version(&endless, acadrust::DxfVersion::AC1032), None); + let crlf = b" 0\r\nSECTION\r\n 2\r\nHEADER\r\n 9\r\n$ACADVER\r\n 1\r\n AC1015\r\n 0\r\nENDSEC\r\n"; + let out = native::declare_version(crlf, acadrust::DxfVersion::AC1032).unwrap(); + assert_eq!(out, b" 0\r\nSECTION\r\n 2\r\nHEADER\r\n 9\r\n$ACADVER\r\n 1\r\n AC1032\r\n 0\r\nENDSEC\r\n"); + } + + /// Declared as 2007 or later, a DXF is UTF-8: code page text is decoded first. Declared as an + /// older version, the code page bytes stay as they are. + #[cfg(not(no_dwg_bridge))] + #[test] + fn declare_version_decodes_a_code_page_for_utf8_versions() { + let dxf = r2000(b"A\xe7\xe3o"); + let utf8 = native::declare_version(&dxf, acadrust::DxfVersion::AC1032).unwrap(); + let text = String::from_utf8(utf8).unwrap(); + assert!(text.contains("\r\nAĂ§Ă£o\r\n") && text.contains("$ACADVER\r\n 1\r\nAC1032\r\n"), "{text}"); + let legacy = native::declare_version(&dxf, acadrust::DxfVersion::AC1018).unwrap(); + assert!(legacy.windows(4).any(|w| w == b"A\xe7\xe3o"), "the code page bytes stay"); + assert!(legacy.windows(6).any(|w| w == b"AC1018")); + // A code page it doesn't know is read as Windows-1252, as acadrust reads an older DXF. + let at = dxf.windows(9).position(|w| w == b"ANSI_1252").unwrap(); + let unknown = [&dxf[..at], b"NO_SUCH_CP".as_slice(), &dxf[at + 9..]].concat(); + let text = String::from_utf8(native::declare_version(&unknown, acadrust::DxfVersion::AC1032).unwrap()).unwrap(); + assert!(text.contains("\r\nAĂ§Ă£o\r\n"), "{text}"); + // Without a `$DWGCODEPAGE` there is nothing to decode it by: it is read as it is. + let none = [&dxf[..at - 21], &dxf[at + 11..]].concat(); + assert!(!none.windows(12).any(|w| w == b"$DWGCODEPAGE"), "{}", String::from_utf8_lossy(&none)); + assert_eq!(native::declare_version(&none, acadrust::DxfVersion::AC1032), None); + } + + /// An unknown kept version converts as [`dxf_to_dwg_as`] does. + #[cfg(not(no_dwg_bridge))] + #[test] + fn unknown_kept_version_converts_as_before() { + let dxf = r2000(b"x"); + assert_eq!(dxf_to_dwg_kept_as(&dxf, "AC1032", "AC9999"), dxf_to_dwg_as(&dxf, "AC1032")); + assert!(dxf_to_dwg_kept_as(&dxf, "AC0000", "AC1032").is_err(), "an unwritable version is still an error"); + } + + /// The CLASS names in CLASSES order; nothing for a binary DXF or one without CLASSES. + #[cfg(not(no_dwg_bridge))] + #[test] + fn dxf_class_names_in_classes_order() { + let dxf = b" 0\r\nSECTION\r\n 2\r\nHEADER\r\n 0\r\nENDSEC\r\n 0\r\nSECTION\r\n 2\r\nCLASSES\r\n 0\r\nCLASS\r\n 1\r\nWIDGETA\r\n 2\r\nAcDbWidgetA\r\n 3\r\nApp\r\n 0\r\nCLASS\r\n 1\r\nWIDGETB\r\n 2\r\nAcDbWidgetB\r\n 0\r\nENDSEC\r\n 0\r\nSECTION\r\n 2\r\nENTITIES\r\n 0\r\nTEXT\r\n 1\r\nNOT_A_CLASS\r\n 0\r\nENDSEC\r\n 0\r\nEOF\r\n"; + assert_eq!(native::dxf_class_names(dxf).unwrap(), ["WIDGETA", "WIDGETB"]); + // A CLASS record without a name keeps its place, so the classes after it keep theirs + // (#2837 review). + let unnamed = b" 0\r\nSECTION\r\n 2\r\nCLASSES\r\n 0\r\nCLASS\r\n 2\r\nAcDbNoName\r\n 0\r\nCLASS\r\n 1\r\nWIDGETB\r\n 0\r\nENDSEC\r\n 0\r\nEOF\r\n"; + assert_eq!(native::dxf_class_names(unnamed).unwrap(), ["", "WIDGETB"]); + assert_eq!(native::dxf_class_names(b" 0\nSECTION\n 2\nHEADER\n 0\nENDSEC\n 0\nEOF\n"), Some(Vec::new())); + assert_eq!(native::dxf_class_names(b"AutoCAD Binary DXF\r\n\x1a\0"), None); + } + + /// AutoCAD's DWG scale action (BLOCKSCALEACTION) ends with a byte after its three + /// connections that DXF doesn't carry. acadrust 0.6.4 neither read nor wrote it, so AutoCAD + /// read past the object and refused every DWG CADCraft wrote with a dynamic block that scales + /// ("Error Decrypting Data"); 0.6.5 added it and 0.6.6, which this uses, still keeps it (0.6.6 also + /// fixes multi-layout DWGs, #1463). The byte reaches the DWG, and a scale action + /// from DXF is written whole, with the byte AutoCAD writes (0). + #[cfg(not(no_dwg_bridge))] + #[test] + fn scale_actions_keep_the_byte_autocad_reads_after_their_connections() { + use acadrust::objects::*; + let scale_action = |doc: &acadrust::CadDocument| { + doc.objects.values().find_map(|o| match o { + ObjectType::DynamicBlock(DynamicBlockObject { data: DynamicBlockData::ScaleAction(s), .. }) => Some(s.clone()), + _ => None, + }) + }; + let mut doc = acadrust::CadDocument::new(); + let h = doc.allocate_handle(); + let conn = |code, name: &str| BlockConnection { code, name: name.into() }; + let mut action = DynamicBlockObject::new("BLOCKSCALEACTION", "AcDbBlockScaleAction"); + action.handle = h; + action.owner = doc.header.named_objects_dict_handle; + let mut a = BlockBasePointAction::default(); + a.action.action.element.name = "Scale".into(); + a.action.connections = vec![conn(2, "UpdatedBaseX"), conn(2, "UpdatedBaseY")]; + a.action.dependent = true; + a.connections = vec![conn(2, "Scale"), conn(2, "XScale"), conn(2, "YScale")]; + a.dwg_tail_byte = 7; + action.data = DynamicBlockData::ScaleAction(a.clone()); + doc.objects.insert(h, ObjectType::DynamicBlock(action)); + + let dwg = acadrust::DwgWriter::write_to_vec(&doc).unwrap(); + let back = acadrust::DwgReader::from_stream(std::io::Cursor::new(dwg)).read().unwrap(); + assert_eq!(scale_action(&back).map(|s| s.dwg_tail_byte), Some(7), "the byte isn't written"); + + let mut dxf = Vec::new(); + acadrust::DxfWriter::new(&doc).write_to_writer(&mut dxf).unwrap(); + for version in ["AC1018", "AC1032"] { + let dwg = dxf_to_dwg_as(&dxf, version).unwrap(); + let back = acadrust::DwgReader::from_stream(std::io::Cursor::new(dwg)).read().unwrap(); + let read = scale_action(&back).unwrap_or_else(|| panic!("{version}: the scale action wasn't written")); + assert_eq!(read.connections, a.connections, "{version}"); + assert_eq!(read.action.connections, a.action.connections, "{version}"); + assert_eq!(read.dwg_tail_byte, 0, "{version}"); + } + } + + /// Saving says what the DWG writer left out, per type: entities of types acadrust's DXF + /// reader doesn't model (another program's custom entity) can't be written; a drawing the DWG + /// keeps whole reports nothing. + #[cfg(not(no_dwg_bridge))] + #[test] + fn saving_lists_the_types_the_dwg_writer_leaves_out() { + let line = " 0\nLINE\n 8\n0\n100\nAcDbEntity\n100\nAcDbLine\n 10\n0.0\n 20\n0.0\n 30\n0.0\n 11\n10.0\n 21\n0.0\n 31\n0.0\n"; + let widget = " 0\nACME_WIDGET\n 8\n0\n100\nAcDbEntity\n100\nAcmeWidget\n 10\n1.0\n 20\n2.0\n 30\n0.0\n"; + let dxf = |body: &str| { + format!(" 0\nSECTION\n 2\nHEADER\n 9\n$ACADVER\n 1\nAC1032\n 0\nENDSEC\n 0\nSECTION\n 2\nENTITIES\n{body} 0\nENDSEC\n 0\nEOF\n") + }; + for version in ["AC1015", "AC1018", "AC1032"] { + let (dwg, unwritten) = dxf_to_dwg_noted(dxf(&format!("{line}{widget}{widget}")).as_bytes(), version).unwrap(); + assert!(is_dwg(&dwg), "{version}"); + assert_eq!(unwritten, [("ACME_WIDGET".to_string(), 2)].into_iter().collect::(), "{version}"); + } + let (_, unwritten) = dxf_to_dwg_noted(dxf(line).as_bytes(), "AC1032").unwrap(); + assert!(unwritten.is_empty(), "{unwritten:?}"); + } + + /// An OLE2FRAME, which CADCraft keeps as its DXF groups without modelling it, is written to + /// the DWG with its data: it isn't in the list of what the writer left out. + #[cfg(not(no_dwg_bridge))] + #[test] + fn ole_frames_are_written_with_their_data() { + let ole = " 0\nOLE2FRAME\n 8\n0\n100\nAcDbEntity\n100\nAcDbOle2Frame\n 70\n2\n 10\n1.0\n 20\n9.0\n 30\n0.0\n 11\n8.0\n 21\n2.0\n 31\n0.0\n 71\n2\n 72\n0\n 73\n1\n 90\n4\n310\nDEADBEEF\n 1\nOLE\n"; + let dxf = + format!(" 0\nSECTION\n 2\nHEADER\n 9\n$ACADVER\n 1\nAC1032\n 0\nENDSEC\n 0\nSECTION\n 2\nENTITIES\n{ole} 0\nENDSEC\n 0\nEOF\n"); + for version in ["AC1015", "AC1018", "AC1032"] { + let (dwg, unwritten) = dxf_to_dwg_noted(dxf.as_bytes(), version).unwrap(); + assert!(unwritten.is_empty(), "{version} {unwritten:?}"); + let back = acadrust::DwgReader::from_stream(std::io::Cursor::new(dwg)).read().unwrap(); + let frames: Vec> = back + .entities() + .filter_map(|e| match e { + acadrust::entities::EntityType::Ole2Frame(o) => Some(o.encoded_payload()), + _ => None, + }) + .collect(); + assert_eq!(frames, [vec![0xDE, 0xAD, 0xBE, 0xEF]], "{version}"); + } + } +} diff --git a/crates/dwg/src/proxy_transport.rs b/crates/dwg/src/proxy_transport.rs new file mode 100644 index 000000000..db0b984ec --- /dev/null +++ b/crates/dwg/src/proxy_transport.rs @@ -0,0 +1,632 @@ +//! Another application's entities and their classes through the DWG bridge: on reading a DWG, +//! registered entities acadrust has no model for become proxy entities of their class; on writing +//! one, the DWG gets the source DXF's classes in their places, so every proxy names its own class. + +use std::collections::{HashMap, HashSet}; + +use acadrust::CadDocument; +use acadrust::classes::{DxfClass, ProxyFlags}; +use acadrust::entities::{EntityType, ExtendedEntityData, ProxyEntityData}; +use acadrust::objects::{ObjectType, ProxyPayload}; + +fn class_number(index: usize) -> Result { + index.checked_add(500).and_then(|n| i16::try_from(n).ok()).ok_or_else(|| "DXF: too many application classes".into()) +} + +/// After a DWG read: each registered entity of another application's class that acadrust keeps +/// only as its record and opaque data (no decoded properties) becomes a proxy entity of that class +/// (DXF Reference, ACAD_PROXY_ENTITY), with its graphics, data bits, R2007+ strings and object +/// references, so the DXF keeps it; and every proxy's class number becomes 500 plus its class's +/// place in the classes, as a DXF numbers it. A record that can't be converted (its class is +/// missing or not unique) is left out and named in the notes; one with decoded properties stays as +/// read, also named, as its opaque data has no DXF form beside them. +pub(super) fn prepare(doc: &mut CadDocument) -> Vec { + let classes: Vec<_> = doc.classes.iter().cloned().collect(); + let mut numbers = HashMap::new(); + let mut identities = HashMap::new(); + for (index, class) in classes.iter().enumerate() { + let number = class_number(index).ok(); + numbers.entry(i32::from(class.class_number)).and_modify(|v| *v = None).or_insert(number.map(i32::from)); + let identity = (class.dxf_name.as_str(), class.cpp_class_name.as_str(), class.is_an_entity); + identities.entry(identity).and_modify(|v| *v = None).or_insert(number); + } + let identity = |dxf: &str, cpp: &str, entity: bool| { + identities + .get(&(dxf, cpp, entity)) + .copied() + .flatten() + .map(i32::from) + .ok_or_else(|| format!("missing or ambiguous application CLASS for {dxf:.80} ({cpp:.80})")) + }; + let Some(dwg_version) = super::native::proxy_dwg_version(doc.dwg_source_version.unwrap_or(doc.version)) else { return Vec::new() }; + let maintenance = i32::from(doc.maintenance_version); + let remap = |id: i32| numbers.get(&id).copied().flatten().ok_or_else(|| format!("proxy names missing or ambiguous application class {id}")); + // Only mutate proxies: entities_mut invalidates the decoder's verbatim records for every entity. + let handles: Vec<_> = doc + .entities() + .filter_map(|e| match e { + EntityType::Extended(e) => match &e.data { + ExtendedEntityData::Proxy(_) => Some(e.common.handle), + ExtendedEntityData::RegisteredClass(data) if data.properties.is_empty() || data.payload.bit_count > 0 => Some(e.common.handle), + _ => None, + }, + _ => None, + }) + .collect(); + let mut notes = Vec::new(); + let mut entity_count = 0usize; + let mut object_count = 0usize; + let mut semantic = 0usize; + for handle in handles { + // Ok(None): not an application record; Ok(Some(None)): kept as read, with its properties. + let converted = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| -> Result>, String> { + let Some(EntityType::Extended(entity)) = doc.get_entity(handle) else { return Ok(None) }; + let data = match &entity.data { + ExtendedEntityData::Proxy(data) => { + let mut proxy = data.clone(); + proxy.class_id = remap(proxy.class_id)?; + ExtendedEntityData::Proxy(proxy) + } + ExtendedEntityData::RegisteredClass(data) => { + if !data.properties.is_empty() { + return Ok(Some(None)); + } + let class_id = identity(&data.dxf_name, &data.cpp_class_name, true)?; + let text_payload = entity.common.raw_record.as_deref().map(registered_text).transpose()?.unwrap_or_default(); + ExtendedEntityData::Proxy(ProxyEntityData { + proxy_id: 498, + class_id, + dxf_subclass: String::new(), + version: (maintenance << 16) | dwg_version, + dwg_version, + maintenance_version: maintenance, + from_dxf: false, + graphics: ProxyPayload::from_bytes(entity.common.graphic_data.as_deref().unwrap_or_default()), + payload: data.payload.clone(), + text_payload, + object_ids: data.object_ids.clone(), + }) + } + _ => return Ok(None), + }; + Ok(Some(Some(data))) + })) + .unwrap_or_else(|_| Err("application record conversion failed".into())); + match converted { + Ok(Some(Some(data))) => { + if let Some(EntityType::Extended(entity)) = doc.get_entity_mut(handle) { + entity.data = data; + } + } + Ok(Some(None)) => semantic = semantic.saturating_add(1), + Ok(None) => {} + Err(why) => { + doc.remove_entity(handle); + entity_count = entity_count.saturating_add(1); + skipped_note(&mut notes, "entity", handle, &why); + } + } + } + doc.objects.retain(|handle, object| { + if let ObjectType::ProxyObject(proxy) = object { + match remap(proxy.class_id) { + Ok(id) => proxy.class_id = id, + Err(why) => { + object_count = object_count.saturating_add(1); + skipped_note(&mut notes, "object", *handle, &why); + return false; + } + } + } + true + }); + if entity_count > 0 || object_count > 0 { + notes.insert(0, format!("DWG: partial read; application proxy records left out (entities: {entity_count}, objects: {object_count})")); + } + if semantic > 0 { + notes.push(format!( + "DWG: {semantic} registered entities with decoded properties keep them, without their other data; saving over the original file would lose it" + )); + } + notes +} + +fn skipped_note(notes: &mut Vec, kind: &str, handle: acadrust::types::Handle, why: &str) { + if notes.len() < 6 { + let why: String = why.chars().take(256).collect(); + notes.push(format!("DWG: proxy {kind} {handle} left out: {why}")); + } +} + +/// The decoder's registered-class representation omits the separate R2007+ string stream. +fn registered_text(raw: &acadrust::entities::RawRecord) -> Result { + use acadrust::io::dwg::DwgVersion; + use acadrust::io::dwg::dwg_stream_readers::object_reader::DwgObjectReader; + use acadrust::io::dwg::dwg_stream_writers::bit_writer::DwgBitWriter; + use acadrust::types::DxfVersion; + + if raw.version < DxfVersion::AC1021 { + return Ok(ProxyPayload::default()); + } + let version = DwgVersion::from_dxf_version(raw.version).map_err(|e| format!("DWG proxy text: {e}"))?; + let handle_bits = usize::try_from(raw.handle_bits).map_err(|_| "DWG proxy text: negative handle stream size")?; + let total_bits = raw.data.len().checked_mul(8).ok_or("DWG proxy text: record too large")?; + if handle_bits > total_bits { + return Err("DWG proxy text: handle stream exceeds record".into()); + } + // Frame the verbatim record for the dependency's checked object reader. + let mut frame = DwgBitWriter::new(version, raw.version); + frame.write_modular_short(raw.data.len()); + if raw.version >= DxfVersion::AC1024 { + frame.write_modular_char(handle_bits); + } + frame.write_bytes(&raw.data); + let reader = DwgObjectReader::new(frame.into_bytes(), raw.version, HashMap::new()).map_err(|e| format!("DWG proxy text: {e}"))?; + let (kind, mut streams) = reader.read_record_at(0).map_err(|e| format!("DWG proxy text: {e}"))?; + reader.read_common_entity_data(&mut streams, kind); + if kind == 498 { + // An existing registered-class envelope is a proxy and carries its class name first. + streams.read_variable_text(); + } else if kind < 500 { + return Err("DWG proxy text: registered entity has no application class".into()); + } + streams.check_error().map_err(|e| format!("DWG proxy text: {e}"))?; + let bits = usize::try_from(streams.text_remaining_bits()).map_err(|_| "DWG proxy text: negative string stream size")?; + if bits > total_bits { + return Err("DWG proxy text: string stream exceeds record".into()); + } + let count = u32::try_from(bits).map_err(|_| "DWG proxy text: string stream too large")?; + let mut bytes = vec![0; bits.div_ceil(8)]; + for (index, byte) in bytes.iter_mut().enumerate() { + for bit in 0..8.min(bits.saturating_sub(index * 8)) { + if streams.read_text_bit() { + *byte |= 0x80 >> bit; + } + } + } + streams.check_error().map_err(|e| format!("DWG proxy text: {e}"))?; + Ok(ProxyPayload::from_bits(&bytes, count)) +} + +/// ASCII DXF pairs decoded with the bridge's public code-page and escape codecs. A malformed +/// pair yields `None`, so both class scans fall back rather than guessing its identity. +pub(super) fn text_pairs(dxf: &[u8]) -> Option> + '_> { + use acadrust::io::dxf::code_page::{decode_legacy_escapes, encoding_from_code_page}; + + if dxf.starts_with(b"AutoCAD Binary DXF") { + return None; + } + let version = + super::native::header_value(dxf, b"$ACADVER").and_then(|(_, v)| std::str::from_utf8(v).ok()).map(acadrust::DxfVersion::from_version_string); + let unicode = version.is_some_and(|v| v >= acadrust::DxfVersion::AC1021); + let encoding = if version.is_some_and(|v| v < acadrust::DxfVersion::AC1021) { + super::native::header_value(dxf, b"$DWGCODEPAGE").and_then(|(_, page)| std::str::from_utf8(page).ok()).and_then(encoding_from_code_page) + } else { + None + }; + let mut lines = dxf.split(|b| *b == b'\n').map(|line| line.strip_suffix(b"\r").unwrap_or(line)); + Some(std::iter::from_fn(move || { + let code = lines.next()?; + let Some(value) = lines.next() else { return (!code.trim_ascii().is_empty()).then_some(None) }; + let pair = || { + let code = std::str::from_utf8(code).ok()?.trim().parse().ok()?; + let text = if let Some(encoding) = encoding { + encoding.decode(value).0.into_owned() + } else if unicode { + String::from_utf8_lossy(value).into_owned() + } else { + // The decoder uses UTF-8 with Latin-1 fallback when no encoding was declared. + std::str::from_utf8(value).map_or_else(|_| value.iter().map(|&b| char::from(b)).collect(), str::to_string) + }; + let text = decode_legacy_escapes(&text); + // DXF's escaped control characters are decoded after CIF/MIF, as in the bridge. + let text = text.replace("^J", "\n").replace("^M", "\r").replace("^I", "\t").replace("^ ", "^"); + Some((code, text)) + }; + Some(pair()) + })) +} + +/// The CLASS records of an ASCII DXF, in CLASSES order and numbered by place from 500 (DXF +/// Reference, CLASSES; a proxy's group 91 is that number). `None` for a binary DXF (the bridge's +/// generated DXF is ASCII), and for one with a record that can't be written back as it is: no names, +/// flags or instance count out of range, more classes than the numbers allow. Then +/// `proxy_classes` matches the classes by name instead. +pub(super) fn source_classes(dxf: &[u8]) -> Option> { + let pairs = text_pairs(dxf)?; + let mut records: Vec> = Vec::new(); + let (mut section, mut in_classes) = (false, false); + for pair in pairs { + let (code, value) = pair?; + match (code, value.as_str()) { + (0, "ENDSEC") if in_classes => break, + (0, "SECTION") => section = true, + (2, name) if section => { + in_classes = name == "CLASSES"; + section = false; + } + (0, kind) if in_classes => { + if kind != "CLASS" { + return None; + } + records.push(Vec::new()); + } + (code, value) if in_classes => records.last_mut()?.push((code, value.to_string())), + _ => section = false, + } + } + records.iter().enumerate().map(|(index, groups)| source_class(groups, class_number(index).ok()?)).collect() +} + +/// One CLASS record's groups as the class `number`, or `None` when it can't be written back. +fn source_class(groups: &[(i32, String)], number: i16) -> Option { + let text = |code| groups.iter().find(|(c, _)| *c == code).map(|(_, v)| v.clone()).unwrap_or_default(); + let integer = |code| groups.iter().find(|(c, _)| *c == code).map_or(Some(0), |(_, v)| v.trim().parse::().ok()); + let mut class = DxfClass::new(text(1), text(2)); + if class.dxf_name.is_empty() || class.cpp_class_name.is_empty() { + return None; + } + class.application_name = text(3); + class.proxy_flags = ProxyFlags(u16::try_from(integer(90)?).ok()?); + class.instance_count = i32::try_from(integer(91)?).ok()?; + class.was_zombie = integer(280)? != 0; + class.is_an_entity = integer(281)? != 0; + class.item_class_id = if class.is_an_entity { 498 } else { 499 }; + class.class_number = number; + Some(class) +} + +/// Give a DWG of 2013 or later the source DXF's classes (`classes`, from [`source_classes`]) in +/// their places, with their metadata and duplicate names, so each proxy's DXF class number names +/// its own class; the decoder's default classes the source lacks follow them. A proxy whose number +/// names no source class is left out, as `proxy_classes` leaves it out. Returns how many proxy objects and +/// entities were left out; nothing changes when the drawing has no proxies. `None`, with nothing +/// changed, when the classes would be more than the class numbers allow. +pub(super) fn restore_classes(doc: &mut CadDocument, classes: Vec) -> Option<(usize, usize)> { + let fits = |number: i32| number.checked_sub(500).and_then(|n| usize::try_from(n).ok()).is_some_and(|i| i < classes.len()); + let mut any = false; + let mut entities = Vec::new(); + for entity in doc.entities() { + if let EntityType::Extended(e) = entity + && let ExtendedEntityData::Proxy(p) = &e.data + { + any = true; + if !fits(p.class_id) { + entities.push(e.common.handle); + } + } + } + let mut objects = Vec::new(); + for (handle, object) in &doc.objects { + if let ObjectType::ProxyObject(p) = object { + any = true; + if !fits(p.class_id) { + objects.push(*handle); + } + } + } + if !any { + return Some((0, 0)); + } + let mut names: HashSet<_> = classes.iter().map(|c| c.dxf_name.to_uppercase()).collect(); + let defaults: Vec<_> = doc.classes.iter().filter(|c| names.insert(c.dxf_name.to_uppercase())).cloned().collect(); + if let Some(last) = classes.len().checked_add(defaults.len())?.checked_sub(1) { + class_number(last).ok()?; + } + for handle in &entities { + doc.remove_entity(*handle); + } + for handle in &objects { + doc.objects.remove(handle); + } + doc.classes.clear(); + for (index, mut class) in classes.into_iter().chain(defaults).enumerate() { + class.class_number = class_number(index).ok()?; + doc.classes.push_preserving(class); + } + Some((objects.len(), entities.len())) +} + +#[cfg(test)] +mod tests { + use super::*; + use acadrust::entities::{EntityCommon, ExtendedEntity, RegisteredClassEntityData}; + use acadrust::types::DxfVersion; + + fn proxy_doc() -> CadDocument { + let mut doc = CadDocument::new(); + doc.version = DxfVersion::AC1032; + doc.add_entity(EntityType::Extended(Box::new(ExtendedEntity { + common: EntityCommon::default(), + data: ExtendedEntityData::Proxy(ProxyEntityData { + proxy_id: 498, + class_id: 501, + dxf_subclass: String::new(), + version: 33, + dwg_version: 33, + maintenance_version: 0, + from_dxf: false, + graphics: ProxyPayload::default(), + payload: ProxyPayload::from_bits(&[0x80], 1), + text_payload: ProxyPayload::default(), + object_ids: vec![], + }), + }))) + .unwrap(); + doc + } + + /// An ASCII DXF of `groups`, `(code, value)` per pair of lines. + fn dxf(groups: &[(i32, &str)]) -> Vec { + groups.iter().map(|(c, v)| format!("{c}\n{v}\n")).collect::().into_bytes() + } + + #[test] + fn source_classes_keep_duplicates_and_metadata_before_defaults() { + let bytes = dxf(&[ + (0, "SECTION"), + (2, "HEADER"), + (9, "$DIMSTYLE"), + (2, "CLASSES"), + (0, "ENDSEC"), + (0, "SECTION"), + (2, "CLASSES"), + (0, "CLASS"), + (1, "Đ�бÑ�ĐµĐºÑ‚"), + (2, "First"), + (3, "ĐŸĐµÑ€Đ²Ñ‹Đ¹"), + (90, "1025"), + (91, "7"), + (280, "1"), + (281, "1"), + (0, "CLASS"), + (1, "Đ�бÑ�ĐµĐºÑ‚"), + (2, "Second"), + (3, "Đ’Ñ‚Đ¾Ñ€Đ¾Đ¹"), + (90, "128"), + (281, "1"), + (0, "ENDSEC"), + (0, "EOF"), + ]); + let classes = source_classes(&bytes).unwrap(); + assert_eq!(classes.len(), 2); + assert_eq!(classes[0].application_name, "ĐŸĐµÑ€Đ²Ñ‹Đ¹"); + assert_eq!(classes[0].proxy_flags.0, 1025); + assert_eq!(classes[0].instance_count, 7); + assert!(classes[0].was_zombie); + assert_eq!((classes[0].class_number, classes[1].class_number), (500, 501)); + assert_eq!(classes[1].cpp_class_name, "Second"); + let mut doc = proxy_doc(); + assert_eq!(restore_classes(&mut doc, classes.clone()), Some((0, 0))); + assert_eq!(doc.classes.iter().take(2).cloned().collect::>(), classes); + // A binary DXF, and a CLASS record that can't be written back as it is, give no classes. + assert!(source_classes(b"AutoCAD Binary DXF\r\n\x1a\0").is_none()); + for (code, value) in [(1, ""), (2, ""), (90, "65536"), (91, "9223372036854775807"), (281, "x")] { + let mut class = vec![(1, "A"), (2, "A")]; + class.retain(|(c, _)| *c != code); + class.push((code, value)); + let mut groups = vec![(0, "SECTION"), (2, "CLASSES"), (0, "CLASS")]; + groups.extend(class); + groups.extend([(0, "ENDSEC"), (0, "EOF")]); + assert!(source_classes(&dxf(&groups)).is_none(), "{code} {value}"); + } + } + + #[test] + fn source_classes_decode_escapes_once_and_respect_legacy_code_pages() { + let body = "0\nCLASS\n1\n\\U+041E\\U+0431\\U+044A\\U+0435\\U+043A\\U+0442\n2\nClass\n3\nCaf\\U+00E9\\U+D83D\\U+DE42\\U+005CU+0041\n281\n1\n0\nENDSEC\n0\nEOF\n"; + let text = format!("0\nSECTION\n2\nCLASSES\n{body}"); + let classes = source_classes(text.as_bytes()).unwrap(); + assert_eq!(classes[0].dxf_name, "Đ�бÑ�ĐµĐºÑ‚"); + assert_eq!(classes[0].application_name, "CafĂ©đŸ™‚\\U+0041"); + + let prefix = + b"0\nSECTION\n2\nHEADER\n9\n$ACADVER\n1\nAC1015\n9\n$DWGCODEPAGE\n3\nANSI_1251\n0\nENDSEC\n0\nSECTION\n2\nCLASSES\n0\nCLASS\n1\n"; + let encoded = [prefix.as_slice(), b"\xCF\xF0\xE8\xEC\xE5\xF0", b"\n2\nClass\n3\nApp\n281\n1\n0\nENDSEC\n0\nEOF\n"].concat(); + assert_eq!(source_classes(&encoded).unwrap()[0].dxf_name, "ĐŸÑ€Đ¸Đ¼ĐµÑ€"); + // R2007+ text is UTF-8 even with a legacy code-page declaration. + let modern = String::from_utf8_lossy(prefix).replace("AC1015", "AC1032") + "ĐŸÑ€Đ¸Đ¼ĐµÑ€\n2\nClass\n3\nApp\n281\n1\n0\nENDSEC\n0\nEOF\n"; + assert_eq!(source_classes(modern.as_bytes()).unwrap()[0].dxf_name, "ĐŸÑ€Đ¸Đ¼ĐµÑ€"); + } + + #[test] + fn proxies_of_missing_classes_are_left_out() { + for id in [-1, 499, 501, i32::MAX] { + let mut doc = proxy_doc(); + if let Some(EntityType::Extended(e)) = doc.entities_mut().next() + && let ExtendedEntityData::Proxy(p) = &mut e.data + { + p.class_id = id; + } + let mut class = DxfClass::new_entity("ENTITY", "Entity"); + class.class_number = 500; + assert_eq!(restore_classes(&mut doc, vec![class]), Some((0, 1)), "{id}"); + assert_eq!(doc.entities().count(), 0); + } + // Class 501 is there: the proxy keeps it, in its place. + let mut doc = proxy_doc(); + assert_eq!(restore_classes(&mut doc, vec![DxfClass::new("A", "A"), DxfClass::new_entity("B", "B")]), Some((0, 0))); + assert_eq!(doc.entities().count(), 1); + assert_eq!(doc.classes.iter().nth(1).map(|c| (c.dxf_name.clone(), c.class_number)), Some(("B".to_string(), 501))); + // Without proxies the classes stay the decoder's. + let mut doc = CadDocument::new(); + let before: Vec<_> = doc.classes.iter().cloned().collect(); + assert_eq!(restore_classes(&mut doc, vec![DxfClass::new("A", "A")]), Some((0, 0))); + assert_eq!(doc.classes.iter().cloned().collect::>(), before); + } + + #[test] + fn unrepresentable_proxies_keep_other_geometry_and_bounded_diagnostics() { + use acadrust::entities::{Line, RawRecord}; + use acadrust::objects::ProxyObject; + use acadrust::types::{Handle, Vector3}; + for count in [1, 10] { + let mut doc = proxy_doc(); + if let Some(EntityType::Extended(e)) = doc.entities_mut().next() + && let ExtendedEntityData::Proxy(p) = &mut e.data + { + p.class_id = i32::MAX; + } + let template = doc.entities().next().unwrap().clone(); + for _ in 1..count { + let mut entity = template.clone(); + entity.common_mut().handle = Handle::NULL; + doc.add_entity(entity).unwrap(); + } + let object = Handle::new(u64::MAX - 1); + assert!(!doc.objects.contains_key(&object)); + doc.objects.insert(object, ObjectType::ProxyObject(ProxyObject { handle: object, class_id: i32::MAX, ..Default::default() })); + let line = doc.add_entity(EntityType::Line(Line::from_points(Vector3::new(2.0, 3.0, 0.0), Vector3::new(12.0, 13.0, 0.0)))).unwrap(); + let raw = RawRecord { data: vec![0xAA, 0x55], handle_bits: 0, version: DxfVersion::AC1032 }; + doc.get_entity_mut(line).unwrap().common_mut().raw_record = Some(std::sync::Arc::new(raw.clone())); + let notes = prepare(&mut doc); + assert_eq!(notes.len(), (count + 1).min(6) + 1); + assert!(notes.iter().all(|n| n.starts_with("DWG:"))); + assert!(notes[0].contains(&format!("entities: {count}, objects: 1"))); + assert_eq!(doc.entities().count(), 1); + assert!(!doc.objects.contains_key(&object)); + let Some(EntityType::Line(actual)) = doc.get_entity(line) else { panic!("readable line") }; + assert_eq!(actual.start, Vector3::new(2.0, 3.0, 0.0)); + assert_eq!(actual.end, Vector3::new(12.0, 13.0, 0.0)); + assert_eq!(doc.get_entity(line).unwrap().common().raw_record.as_deref(), Some(&raw)); + } + } + #[test] + fn opaque_registered_class_ambiguity_is_not_guessed() { + let mut doc = CadDocument::new(); + doc.classes.push_preserving(DxfClass::new_entity("CUSTOM", "Custom")); + doc.classes.push_preserving(DxfClass::new_entity("CUSTOM", "Custom")); + doc.add_entity(EntityType::Extended(Box::new(ExtendedEntity { + common: EntityCommon::default(), + data: ExtendedEntityData::RegisteredClass(RegisteredClassEntityData { + dxf_name: "CUSTOM".into(), + cpp_class_name: "Custom".into(), + properties: vec![], + payload: ProxyPayload::from_bits(&[0x80], 1), + object_ids: vec![], + }), + }))) + .unwrap(); + assert!(prepare(&mut doc).iter().any(|n| n.contains("ambiguous"))); + assert_eq!(doc.entities().count(), 0); + assert!(class_number(usize::MAX).is_err()); + } + #[test] + fn semantic_records_keep_verbatim_data_and_mixed_payloads_are_reported() { + use acadrust::entities::RawRecord; + use acadrust::objects::{SemanticProperty, SemanticPropertyValue}; + for payload in [ProxyPayload::default(), ProxyPayload::from_bits(&[0x80], 1)] { + let mut doc = CadDocument::new(); + let raw = RawRecord { data: vec![0xAA], handle_bits: 0, version: DxfVersion::AC1032 }; + let handle = doc + .add_entity(EntityType::Extended(Box::new(ExtendedEntity { + common: EntityCommon::default(), + data: ExtendedEntityData::RegisteredClass(RegisteredClassEntityData { + dxf_name: "SEMANTIC".into(), + cpp_class_name: "Semantic".into(), + properties: vec![SemanticProperty { + subclass: "Semantic".into(), + code: 1, + value: SemanticPropertyValue::Text("value".into()), + }], + payload: payload.clone(), + object_ids: vec![], + }), + }))) + .unwrap(); + doc.get_entity_mut(handle).unwrap().common_mut().raw_record = Some(std::sync::Arc::new(raw.clone())); + assert_eq!(doc.get_entity(handle).unwrap().common().raw_record.as_deref(), Some(&raw)); + if payload.bit_count == 0 { + assert!(prepare(&mut doc).is_empty()); + assert_eq!(doc.get_entity(handle).unwrap().common().raw_record.as_deref(), Some(&raw)); + } else { + // Kept as read, and named: its opaque data has no DXF form beside the properties. + assert!(prepare(&mut doc).iter().any(|n| n.contains("decoded properties"))); + assert_eq!(doc.get_entity(handle).unwrap().common().raw_record.as_deref(), Some(&raw)); + } + } + } +} + +#[cfg(test)] +mod string_stream_tests { + use super::*; + use acadrust::entities::{EntityCommon, ExtendedEntity, RawRecord, RegisteredClassEntityData}; + use acadrust::types::DxfVersion; + use std::io::Cursor; + + #[test] + fn registered_transport_recovers_separate_non_byte_aligned_strings() { + for version in [DxfVersion::AC1027, DxfVersion::AC1032] { + let mut source = CadDocument::new(); + source.version = version; + source.classes.add_or_update(DxfClass::new_entity("CUSTOM_TEXT", "CustomText")); + let class_id = i32::from(source.classes.get_by_name("CUSTOM_TEXT").unwrap().class_number); + let text = ProxyPayload::from_bits(&[0xA5, 0xC7, 0xE0], 19); + let payload = ProxyPayload::from_bits(&[0x80], 1); + source + .add_entity(EntityType::Extended(Box::new(ExtendedEntity { + common: EntityCommon::default(), + data: ExtendedEntityData::Proxy(ProxyEntityData { + proxy_id: 498, + class_id, + dxf_subclass: "CustomText".into(), + version: 33, + dwg_version: 33, + maintenance_version: 0, + from_dxf: false, + graphics: ProxyPayload::default(), + payload: payload.clone(), + text_payload: text.clone(), + object_ids: vec![], + }), + }))) + .unwrap(); + let bytes = acadrust::DwgWriter::write_to_vec(&source).unwrap(); + let mut decoded = acadrust::DwgReader::from_stream(Cursor::new(bytes)).read().unwrap(); + let entity = decoded.entities().next().unwrap(); + let handle = entity.common().handle; + let raw = entity.common().raw_record.clone().unwrap(); + assert_eq!(registered_text(&raw).unwrap().data(), text.data()); + assert_eq!(registered_text(&raw).unwrap().bit_count, 19); + let EntityType::Extended(entity) = decoded.get_entity_mut(handle).unwrap() else { panic!() }; + entity.data = ExtendedEntityData::RegisteredClass(RegisteredClassEntityData { + dxf_name: "CUSTOM_TEXT".into(), + cpp_class_name: "CustomText".into(), + properties: vec![], + payload: payload.clone(), + object_ids: vec![], + }); + entity.common.raw_record = Some(raw); + assert!(prepare(&mut decoded).is_empty()); + for _ in 0..2 { + let mut dxf = Vec::new(); + acadrust::DxfWriter::new(&decoded).write_to_writer(&mut dxf).unwrap(); + let bytes = crate::dxf_to_dwg_as(&dxf, "AC1032").unwrap(); + decoded = acadrust::DwgReader::from_stream(Cursor::new(bytes)).read().unwrap(); + let p = decoded + .entities() + .find_map(|e| match e { + EntityType::Extended(e) => match &e.data { + ExtendedEntityData::Proxy(p) => Some(p), + _ => None, + }, + _ => None, + }) + .unwrap(); + assert_eq!(p.payload.bit_count, 1); + assert_eq!(p.payload.data(), payload.data()); + assert_eq!(p.text_payload.bit_count, 19); + assert_eq!(p.text_payload.data(), text.data()); + } + } + } + + #[test] + fn invalid_string_stream_boundaries_return_errors() { + for handle_bits in [-1, 1000] { + let raw = RawRecord { data: vec![0; 8], handle_bits, version: DxfVersion::AC1032 }; + assert!(registered_text(&raw).is_err()); + } + } +} diff --git a/crates/dwg/src/worker.rs b/crates/dwg/src/worker.rs new file mode 100644 index 000000000..1d16f5701 --- /dev/null +++ b/crates/dwg/src/worker.rs @@ -0,0 +1,425 @@ +//! The DWG reader and writer in a child process. +//! +//! An allocation failure or stack overflow in the third-party code ends the process it runs in, and +//! `catch_unwind` can't stop that. So the app runs them in a copy of itself: [`use_own_worker`] says +//! how to start one, and [`serve_if_asked`], first thing in `main`, turns the copy into a worker +//! that reads one request from standard input and writes one reply to standard output. A file that +//! kills the worker then costs that one open or save, and every other open drawing stays safe. +//! +//! Without a worker [`decode`] and [`encode_as`] refuse rather than run the conversion in the +//! calling process, because the third-party code is the part that can end a process. +//! [`allow_in_process`] is the explicit exception, for input the caller knows is trusted (tests, or +//! bytes this process produced itself). There's no memory limit on the worker yet: that takes +//! per-platform system calls, which `unsafe_code = "forbid"` keeps out of this crate. +//! +//! A reply comes from a process that just handled a possibly hostile file, so it is parsed as +//! untrusted input. + +use std::ffi::OsString; +use std::io::{Read, Write}; +use std::path::PathBuf; +use std::process::{Command, Stdio}; +use std::sync::Mutex; +use std::sync::atomic::{AtomicBool, Ordering}; +use std::sync::mpsc::{RecvTimeoutError, channel}; +use std::time::Duration; + +use super::Limits; + +/// The argument that makes the app a worker. +pub const WORKER_ARG: &str = "--dwg-worker"; +/// Bumped when the messages change: an app updated while running may start a newer copy. +const PROTOCOL: u8 = 2; +const DECODE: u8 = 1; +const ENCODE: u8 = 2; +const CRASH: u8 = 0xff; +/// How long one file may take before the worker is stopped. It can't be interrupted otherwise. +const TIMEOUT: Duration = Duration::from_secs(30 * 60); +/// Most bytes accepted for what a reply carries besides the drawing. +const MAX_NOTES_BYTES: u64 = 64 << 20; +/// Most bytes accepted as a written DWG. +const MAX_DWG_REPLY: u64 = 2 << 30; +/// Most bytes one request holds: the largest input, whichever the direction, and its header. +const MAX_REQUEST: u64 = (2 << 30) + 256; +/// What a conversion says when it has no worker and is not allowed to run here. +const NO_WORKER: &str = "DWG: no worker is configured, so this file was not converted; \ + call use_worker or use_own_worker at startup, or allow_in_process for input that is trusted"; + +struct Worker { + program: PathBuf, + args: Vec, +} + +static WORKER: Mutex> = Mutex::new(None); +/// Whether this process may run the conversion itself, set only by [`allow_in_process`]. +static IN_PROCESS: AtomicBool = AtomicBool::new(false); + +/// Start `program args` as the worker from now on. `program` must call [`serve_if_asked`]. +pub fn use_worker(program: PathBuf, args: Vec) { + if let Ok(mut w) = WORKER.lock() { + *w = Some(Worker { program, args }); + } +} + +/// Use a copy of this program as the worker, started with [`WORKER_ARG`]. If its path can't be +/// found, the reader and writer stay in this process. +pub fn use_own_worker() { + match std::env::current_exe() { + Ok(program) => use_worker(program, vec![WORKER_ARG.into()]), + Err(e) => log::warn!("DWG worker: can't find this program: {e}"), + } +} + +/// Let [`decode`] and [`encode_as`] run the conversion in this process when no worker is set. +/// +/// This opts out of the isolation the crate exists for: the third-party code allocates from sizes +/// a hostile file declares, and a failed allocation ends the process. Call it for input the caller +/// knows is trusted (tests, or bytes this process produced), never for a file that arrived from +/// outside. +pub fn allow_in_process() { + IN_PROCESS.store(true, Ordering::Relaxed); +} + +fn in_process_allowed() -> bool { + IN_PROCESS.load(Ordering::Relaxed) +} + +/// Serve one request and return the exit code if this process was started as a worker. +pub fn serve_if_asked() -> Option { + (std::env::args_os().nth(1)? == WORKER_ARG).then(|| serve(std::io::stdin().lock(), std::io::stdout().lock())) +} + +/// [`super::dwg_to_dxf_noted`] in the worker. With no worker configured this is refused unless +/// [`allow_in_process`] was called; a configured worker that cannot start is an error. +pub fn decode(bytes: &[u8], limits: Limits) -> Result<(Vec, Vec), String> { + // Before starting a process, and so the same limit holds with or without one. + if bytes.len() > limits.max_dwg_bytes { + return super::dwg_to_dxf_noted(bytes, limits); + } + let reply_limit = (limits.max_dxf_bytes as u64).saturating_add(MAX_NOTES_BYTES); + match call(&request(DECODE, limits, ""), bytes, reply_limit, "reading this file") { + Some(reply) => reply?, + None if in_process_allowed() => super::dwg_to_dxf_noted(bytes, limits), + None => Err(NO_WORKER.to_string()), + } +} + +/// [`super::dxf_to_dwg_as`] in the worker. Refused without a worker unless [`allow_in_process`] +/// was called. +pub fn encode_as(dxf: &[u8], version: &str) -> Result, String> { + encode_noted(dxf, version).map(|(dwg, _)| dwg) +} + +/// [`super::dxf_to_dwg_noted`] in the worker. What the writer left out comes back as notes of +/// `TYPEcount`. +pub fn encode_noted(dxf: &[u8], version: &str) -> Result<(Vec, super::Unwritten), String> { + encode_kept_noted(dxf, version, None) +} + +/// [`encode_noted`] for a DXF whose carried-over groups were read from a file of version `kept` +/// ([`super::dxf_to_dwg_kept_noted`]); `None` is [`encode_noted`]. +pub fn encode_kept_noted(dxf: &[u8], version: &str, kept: Option<&str>) -> Result<(Vec, super::Unwritten), String> { + if u8::try_from(version.len()).is_err() { + return Err(super::unsupported(version)); + } + // A kept version too long to send is unsupported by this bridge and changes nothing. + let kept = kept.filter(|k| u8::try_from(k.len()).is_ok()); + let mut header = request(ENCODE, Limits::DEFAULT, version); + push_str(&mut header, kept.unwrap_or("")); + match call(&header, dxf, MAX_DWG_REPLY, "writing this file") { + Some(reply) => reply?.map(|(dwg, notes)| (dwg, unwritten_from(¬es))), + None if in_process_allowed() => encode_here(dxf, version, kept), + None => Err(NO_WORKER.to_string()), + } +} + +/// The conversion in this process. +fn encode_here(dxf: &[u8], version: &str, kept: Option<&str>) -> Result<(Vec, super::Unwritten), String> { + match kept { + Some(kept) => super::dxf_to_dwg_kept_noted(dxf, version, kept), + None => super::dxf_to_dwg_noted(dxf, version), + } +} + +fn unwritten_from(notes: &[String]) -> super::Unwritten { + notes.iter().filter_map(|n| n.split_once('\t')).filter_map(|(kind, n)| Some((kind.to_string(), n.parse::().ok()?))).fold( + super::Unwritten::new(), + |mut m, (kind, n)| { + *m.entry(kind).or_insert(0) = n; + m + }, + ) +} + +/// Make the worker abort, as the third-party code can: what the tests prove isolation with. +#[doc(hidden)] +pub fn crash_worker_for_test() -> Result<(), String> { + call(&request(CRASH, Limits::DEFAULT, ""), &[], 1024, "testing").ok_or("no worker")??.map(|_| ()) +} + +fn request(op: u8, limits: Limits, version: &str) -> Vec { + let mut r = vec![PROTOCOL, op]; + r.extend((limits.max_dwg_bytes as u64).to_le_bytes()); + r.extend((limits.max_dxf_bytes as u64).to_le_bytes()); + push_str(&mut r, version); + r +} + +/// `text` with a one-byte length before it; callers keep it under 256 bytes. +fn push_str(r: &mut Vec, text: &str) { + r.push(u8::try_from(text.len()).unwrap_or(0)); + r.extend(text.as_bytes()); +} + +type Reply = Result<(Vec, Vec), String>; + +/// Run one request in a worker. `None`: no worker is configured, so the caller decides what to do +/// (convert here only with [`allow_in_process`], or refuse). `Err`: a configured worker could not +/// start or went wrong, and `doing` says at what, for the user. +fn call(header: &[u8], payload: &[u8], reply_limit: u64, doing: &str) -> Option> { + let mut command = { + let worker = WORKER.lock().ok()?; + let worker = worker.as_ref()?; + let mut c = Command::new(&worker.program); + c.args(&worker.args); + c + }; + command.stdin(Stdio::piped()).stdout(Stdio::piped()).stderr(Stdio::null()); + #[cfg(windows)] + { + use std::os::windows::process::CommandExt; + const CREATE_NO_WINDOW: u32 = 0x0800_0000; + command.creation_flags(CREATE_NO_WINDOW); + } + let mut child = match command.spawn() { + Ok(c) => c, + Err(e) => { + // A worker is configured but cannot run: never fall back to decoding here. + log::warn!("DWG worker: can't start it: {e}"); + return Some(Err(format!("the DWG worker could not start ({e}); nothing else is affected"))); + } + }; + let (Some(mut to_child), Some(from_child)) = (child.stdin.take(), child.stdout.take()) else { + let _ = child.kill(); + let _ = child.wait(); + return Some(Err("the DWG worker could not start (no pipes); nothing else is affected".to_string())); + }; + let received = std::thread::scope(|scope| { + // Written beside the read: a worker that replies before it has read everything, or dies, + // must block neither side. A failed write shows up as the reply going wrong. + scope.spawn(move || { + let _ = to_child.write_all(header).and_then(|()| to_child.write_all(payload)); + }); + let (tx, rx) = channel(); + scope.spawn(move || { + let mut reply = Vec::new(); + let read = from_child.take(reply_limit).read_to_end(&mut reply); + let _ = tx.send(read.map(|_| reply)); + }); + let received = rx.recv_timeout(TIMEOUT); + if matches!(received, Err(RecvTimeoutError::Timeout)) { + let _ = child.kill(); + } + received + }); + let status = child.wait(); + let stopped = || format!("the DWG worker stopped while {doing}; nothing else is affected"); + Some(match (received, status) { + (Err(RecvTimeoutError::Timeout), _) => Err(format!("{doing} took too long, so the DWG worker was stopped; nothing else is affected")), + (Ok(Ok(reply)), Ok(status)) if status.success() => parse_reply(reply).ok_or_else(stopped), + _ => Err(stopped()), + }) +} + +/// Take `n` bytes off the front of `b`. +fn take<'a>(b: &mut &'a [u8], n: usize) -> Option<&'a [u8]> { + let (head, tail) = b.split_at_checked(n)?; + *b = tail; + Some(head) +} + +fn take_u32(b: &mut &[u8]) -> Option { + let bytes: [u8; 4] = take(b, 4)?.try_into().ok()?; + usize::try_from(u32::from_le_bytes(bytes)).ok() +} + +fn take_u64(b: &mut &[u8]) -> Option { + Some(u64::from_le_bytes(take(b, 8)?.try_into().ok()?)) +} + +/// A reply: 0, the note count, each note with its length, then the bytes with their length; or 1 +/// and a message. `None` when it isn't one, whole. Takes the reply, because the bytes can be most +/// of the memory the app may use: they are kept, not copied. +fn parse_reply(mut reply: Vec) -> Option { + let total = reply.len(); + let mut b = &reply[..]; + match *take(&mut b, 1)?.first()? { + 0 => { + let count = take_u32(&mut b)?; + let mut notes = Vec::new(); + for _ in 0..count { + let n = take_u32(&mut b)?; + notes.push(String::from_utf8_lossy(take(&mut b, n)?).into_owned()); + } + let n = usize::try_from(take_u64(&mut b)?).ok()?; + if b.len() != n { + return None; + } + reply.drain(..total - n); + Some(Ok((reply, notes))) + } + 1 => Some(Err(String::from_utf8_lossy(b).into_owned())), + _ => None, + } +} + +fn write_reply(out: &mut impl Write, reply: Reply) -> std::io::Result<()> { + match reply { + Ok((bytes, notes)) => { + out.write_all(&[0])?; + out.write_all(&u32::try_from(notes.len()).unwrap_or(0).to_le_bytes())?; + for note in ¬es { + out.write_all(&u32::try_from(note.len()).unwrap_or(0).to_le_bytes())?; + if u32::try_from(note.len()).is_ok() { + out.write_all(note.as_bytes())?; + } + } + out.write_all(&(bytes.len() as u64).to_le_bytes())?; + out.write_all(&bytes)?; + } + Err(e) => { + out.write_all(&[1])?; + out.write_all(e.as_bytes())?; + } + } + out.flush() +} + +/// What a request asks for, or why it can't be served. +fn handle(mut b: &[u8]) -> Reply { + let bad = || "the DWG worker got a request it can't read (a different bridge version?)".to_string(); + if take(&mut b, 1).and_then(<[u8]>::first) != Some(&PROTOCOL) { + return Err(bad()); + } + let op = take(&mut b, 1).and_then(<[u8]>::first).copied().ok_or_else(bad)?; + let max_dwg_bytes = take_u64(&mut b).and_then(|n| usize::try_from(n).ok()).ok_or_else(bad)?; + let max_dxf_bytes = take_u64(&mut b).and_then(|n| usize::try_from(n).ok()).ok_or_else(bad)?; + let version_len = take(&mut b, 1).and_then(<[u8]>::first).copied().ok_or_else(bad)?; + let version = take(&mut b, usize::from(version_len)).and_then(|v| std::str::from_utf8(v).ok()).ok_or_else(bad)?; + match op { + DECODE => super::dwg_to_dxf_noted(b, Limits { max_dwg_bytes, max_dxf_bytes }), + ENCODE => { + // The version the DXF's carried-over groups were read from; empty for none. + let kept_len = take(&mut b, 1).and_then(<[u8]>::first).copied().ok_or_else(bad)?; + let kept = take(&mut b, usize::from(kept_len)).and_then(|v| std::str::from_utf8(v).ok()).ok_or_else(bad)?; + encode_here(b, version, (!kept.is_empty()).then_some(kept)) + .map(|(dwg, left_out)| (dwg, left_out.iter().map(|(kind, n)| format!("{kind}\t{n}")).collect())) + } + CRASH => std::process::abort(), + _ => Err(bad()), + } +} + +/// Serve the one request on `input`. The exit code is 0 unless the reply couldn't be sent. +fn serve(input: impl Read, mut output: impl Write) -> i32 { + let mut request = Vec::new(); + let reply = match input.take(MAX_REQUEST).read_to_end(&mut request) { + Ok(_) => handle(&request), + Err(e) => Err(format!("the DWG worker could not read its request: {e}")), + }; + i32::from(write_reply(&mut output, reply).is_err()) +} + +#[cfg(test)] +mod tests { + use super::*; + + fn wire(reply: Reply) -> Vec { + let mut out = Vec::new(); + write_reply(&mut out, reply).unwrap(); + out + } + + #[test] + fn what_the_writer_left_out_survives_the_reply() { + let notes = vec!["ACME_WIDGET\t2".to_string(), "not a count".to_string(), "X\tmany".to_string()]; + assert_eq!(unwritten_from(¬es), [("ACME_WIDGET".to_string(), 2)].into_iter().collect()); + } + + #[test] + fn replies_round_trip() { + let ok: Reply = Ok((b"bytes".to_vec(), vec!["one".into(), "two Ă©".into()])); + assert_eq!(parse_reply(wire(ok.clone())), Some(ok)); + let failed: Reply = Err("DWG: no".into()); + assert_eq!(parse_reply(wire(failed.clone())), Some(failed)); + } + + #[test] + fn a_reply_cut_short_or_made_up_is_refused() { + let whole = wire(Ok((b"bytes".to_vec(), vec!["note".into()]))); + for n in 0..whole.len() { + assert_eq!(parse_reply(whole[..n].to_vec()), None, "cut at {n}"); + } + assert_eq!(parse_reply(vec![7, 0, 0]), None); + // A note count far beyond what follows ends at the first missing note. + assert_eq!(parse_reply(vec![0, 0xff, 0xff, 0xff, 0xff]), None); + assert_eq!(parse_reply(vec![0, 1, 0, 0, 0, 0xff, 0xff, 0xff, 0xff]), None); + } + + #[test] + fn a_request_it_cannot_read_is_an_error_not_a_crash() { + for request in [ + &[][..], + &[PROTOCOL + 1, DECODE], + &[PROTOCOL], + &[PROTOCOL, DECODE, 1, 2, 3], + &[PROTOCOL, 9, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], + ] { + assert!(handle(request).is_err(), "{request:?}"); + } + } + + /// An encode request carries the kept version after the target one; a request without it is + /// one it can't read, and one with it reaches the converter. + #[test] + fn an_encode_request_carries_the_kept_version() { + let bad = "the DWG worker got a request it can't read (a different bridge version?)"; + let short = self::request(ENCODE, Limits::DEFAULT, "AC1032"); + assert_eq!(handle(&short).unwrap_err(), bad); + for kept in ["", "AC1032"] { + let mut request = self::request(ENCODE, Limits::DEFAULT, "AC1032"); + push_str(&mut request, kept); + request.extend(b"not a drawing"); + let err = handle(&request).unwrap_err(); + assert_ne!(err, bad, "kept {kept:?}"); + } + } + + /// A conversion is refused rather than run here, and a worker that cannot start is an error: + /// neither case may turn into decoding a (possibly hostile) file in this process. + #[test] + fn a_conversion_is_refused_without_a_worker_and_never_falls_back_to_this_process() { + // No worker is configured and nothing opted in. + assert_eq!(decode(b"AC1032", Limits::DEFAULT).unwrap_err(), NO_WORKER); + assert_eq!(encode_as(b"not a dxf", "AC1032").unwrap_err(), NO_WORKER); + // Trusted input may run here, and the reader still reports a file it cannot read. + allow_in_process(); + assert!(in_process_allowed()); + assert!(decode(b"AC1032", Limits::DEFAULT).is_err()); + assert!(encode_as(b"not a dxf", "AC1032").is_err()); + // A configured worker that cannot start is an error even after the opt-in: the conversion + // never runs here instead. + use_worker(PathBuf::from("/nonexistent/craft-dwg-worker"), Vec::new()); + assert!(decode(b"AC1032", Limits::DEFAULT).unwrap_err().contains("could not start")); + assert!(encode_as(b"not a dxf", "AC1032").unwrap_err().contains("could not start")); + } + + #[test] + fn serving_reports_a_decode_failure_as_a_reply() { + let mut request = self::request(DECODE, Limits::DEFAULT, ""); + request.extend(b"AC1032 not really a drawing"); + let mut out = Vec::new(); + assert_eq!(serve(&request[..], &mut out), 0); + assert!(matches!(parse_reply(out.clone()), Some(Err(_))), "{out:?}"); + } +} diff --git a/crates/dxf/Cargo.toml b/crates/dxf/Cargo.toml index 70c4f1234..11de1c377 100644 --- a/crates/dxf/Cargo.toml +++ b/crates/dxf/Cargo.toml @@ -15,3 +15,4 @@ craft-codepage = { workspace = true } thiserror = { workspace = true } log = "0.4" encoding_rs = "0.8.35" +kurbo = { workspace = true } diff --git a/crates/dxf/src/drawing.rs b/crates/dxf/src/drawing.rs new file mode 100644 index 000000000..0b52c337c --- /dev/null +++ b/crates/dxf/src/drawing.rs @@ -0,0 +1,774 @@ +//! Incremental ASCII DXF file writer used by VectorCraft. +//! +//! This is the file skeleton around a pre-written `ENTITIES` section: group codes, handles, and +//! the HEADER / TABLES / BLOCKS / OBJECTS sections CAD apps expect. It is a different API from +//! [`crate::model::write`], which serializes a parsed [`crate::model`] drawing as tags. +//! +//! Moved from `vectorcraft-cad` (`crates/cad/src/writer.rs`). The bytes a [`Drawing`] writes must +//! stay the same: VectorCraft's export tests treat this as the format oracle. +//! +//! DXF R13 and later give every object a handle and subclass markers; R2000 and later add layouts +//! and plot styles. + +use std::fmt::Write as _; + +pub use crate::model::write::Format; + +/// An object handle (written in hex). +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct Handle(pub u32); + +/// The handles of the fixed objects every drawing has. Entities and table entries count up from +/// [`FIRST_FREE`]. +pub mod fixed { + use super::Handle; + pub const LAYER_TABLE: Handle = Handle(0x1); + pub const LTYPE_TABLE: Handle = Handle(0x2); + pub const APPID_TABLE: Handle = Handle(0x3); + pub const DIMSTYLE_TABLE: Handle = Handle(0x4); + pub const STYLE_TABLE: Handle = Handle(0x5); + pub const UCS_TABLE: Handle = Handle(0x6); + pub const VIEW_TABLE: Handle = Handle(0x7); + pub const VPORT_TABLE: Handle = Handle(0x8); + pub const BLOCK_RECORD_TABLE: Handle = Handle(0x9); + pub const ROOT: Handle = Handle(0xA); + pub const GROUPS: Handle = Handle(0xB); + pub const LAYOUTS: Handle = Handle(0xC); + pub const MODEL_RECORD: Handle = Handle(0xD); + pub const PAPER_RECORD: Handle = Handle(0xE); + pub const MODEL_BLOCK: Handle = Handle(0xF); + pub const MODEL_END: Handle = Handle(0x10); + pub const PAPER_BLOCK: Handle = Handle(0x11); + pub const PAPER_END: Handle = Handle(0x12); + pub const MODEL_LAYOUT: Handle = Handle(0x13); + pub const PAPER_LAYOUT: Handle = Handle(0x14); + pub const ACTIVE_VPORT: Handle = Handle(0x15); + pub const ACAD_APPID: Handle = Handle(0x16); + pub const STANDARD_DIMSTYLE: Handle = Handle(0x17); + pub const IMAGE_DICT: Handle = Handle(0x18); + pub const IMAGE_VARS: Handle = Handle(0x19); + pub const PLOT_STYLES: Handle = Handle(0x1A); + pub const PLOT_NORMAL: Handle = Handle(0x1B); + pub const FIRST_FREE: u32 = 0x30; +} + +/// A number as DXF writes it: up to 6 decimals, no trailing zeros, 0 for anything not finite. +pub fn num(v: f64) -> String { + if !v.is_finite() { + return "0".into(); + } + let s = format!("{v:.6}"); + let s = s.trim_end_matches('0').trim_end_matches('.'); + if s == "-0" || s.is_empty() { "0".into() } else { s.into() } +} + +/// Text as a DXF string value: one line, `^` escaped, and before DXF 2007 (which reads UTF-8) +/// every character outside ASCII as `\U+XXXX`. +pub fn text(s: &str, utf8: bool) -> String { + let mut out = String::with_capacity(s.len()); + for c in s.chars() { + match c { + '\r' | '\n' | '\t' => out.push(' '), + c if c.is_control() => {} + '^' => out.push_str("^ "), + c if !utf8 && !c.is_ascii() => match u16::try_from(u32::from(c)) { + Ok(u) => _ = write!(out, "\\U+{u:04X}"), + Err(_) => out.push('?'), + }, + c => out.push(c), + } + } + out +} + +/// Group-code/value pairs in the order they are written. +pub struct Codes { + pub v: Format, + out: String, +} + +impl Codes { + pub fn new(v: Format) -> Self { + Self { v, out: String::new() } + } + + pub fn into_string(self) -> String { + self.out + } + + pub fn append(&mut self, other: &str) { + self.out.push_str(other); + } + + /// A value written as it is (names and keywords the writer itself chose). + pub fn raw(&mut self, code: u16, value: &str) { + _ = writeln!(self.out, "{code:>3}\n{value}"); + } + + /// Text from the document, encoded for the version (see [`text`]). + pub fn str(&mut self, code: u16, value: &str) { + let t = text(value, self.v.utf8()); + self.raw(code, &t); + } + + pub fn int(&mut self, code: u16, value: i64) { + _ = writeln!(self.out, "{code:>3}\n{value}"); + } + + pub fn num(&mut self, code: u16, value: f64) { + _ = writeln!(self.out, "{code:>3}\n{}", num(value)); + } + + pub fn handle(&mut self, code: u16, h: Handle) { + _ = writeln!(self.out, "{code:>3}\n{:X}", h.0); + } + + /// A 2D point at `code` (x), `code + 10` (y). + pub fn xy(&mut self, code: u16, x: f64, y: f64) { + self.num(code, x); + self.num(code + 10, y); + } + + /// A 3D point at `code`, `code + 10`, `code + 20` (z = 0). + pub fn xyz(&mut self, code: u16, x: f64, y: f64) { + self.xy(code, x, y); + self.num(code + 20, 0.0); + } + + /// The extrusion direction of planar entities: the z axis. + pub fn normal(&mut self) { + self.num(210, 0.0); + self.num(220, 0.0); + self.num(230, 1.0); + } + + /// The start of an object: its type, and from R13 its handle, owner and first subclass. + fn object(&mut self, kind: &str, h: Handle, owner: Handle, subclass: &str) { + self.raw(0, kind); + if self.v.handles() { + self.handle(if kind == "DIMSTYLE" { 105 } else { 5 }, h); + self.handle(330, owner); + self.raw(100, subclass); + } + } + + /// A table entry: the shared record subclass, then the table's own (from R13). + fn record(&mut self, kind: &str, h: Handle, table: Handle, subclass: &str) { + self.object(kind, h, table, "AcDbSymbolTableRecord"); + if self.v.handles() { + self.raw(100, subclass); + } + } + + fn section(&mut self, name: &str) { + self.raw(0, "SECTION"); + self.raw(2, name); + } + + fn end_section(&mut self) { + self.raw(0, "ENDSEC"); + } + + fn table(&mut self, name: &str, h: Handle, count: usize) { + self.raw(0, "TABLE"); + self.raw(2, name); + if self.v.handles() { + self.handle(5, h); + self.raw(330, "0"); + self.raw(100, "AcDbSymbolTable"); + } + self.int(70, count as i64); + if name == "DIMSTYLE" && self.v.r2000() { + self.raw(100, "AcDbDimStyleTable"); + } + } + + fn end_table(&mut self) { + self.raw(0, "ENDTAB"); + } +} + +/// How an entity is drawn: its colour index, true colour, lineweight, linetype and transparency. +#[derive(Clone, Debug, Default)] +pub struct EntityStyle { + pub aci: u8, + /// `0xRRGGBB` (DXF 2004 and later). + pub rgb: Option, + /// Hundredths of a millimetre (DXF 2000 and later). + pub lineweight: Option, + pub linetype: Option, + /// Opacity 0–255 (DXF 2004 and later; opaque is left out). + pub alpha: Option, +} + +impl Codes { + /// The start of an entity of `kind` owned by `owner` (model space, or a polyline for its + /// vertices) on `layer` with its style, ending with `subclass` (none when empty). + pub fn entity(&mut self, kind: &str, h: Handle, owner: Handle, layer: &str, style: &EntityStyle, subclass: &str) { + self.object(kind, h, owner, "AcDbEntity"); + self.str(8, layer); + if let Some(lt) = &style.linetype { + self.str(6, lt); + } + self.int(62, i64::from(style.aci)); + if self.v.r2000() + && let Some(lw) = style.lineweight + { + self.int(370, i64::from(lw)); + } + if self.v.true_color() { + if let Some(rgb) = style.rgb { + self.int(420, i64::from(rgb)); + } + if let Some(a) = style.alpha { + self.int(440, 0x0200_0000 | i64::from(a)); + } + } + if self.v.handles() && !subclass.is_empty() { + self.raw(100, subclass); + } + } +} + +/// A DXF layer. +#[derive(Clone, Debug)] +pub struct Layer { + pub name: String, + pub aci: u8, + /// Switched off (a hidden layer). + pub off: bool, + pub locked: bool, + /// Plotted (printable). + pub plot: bool, +} + +/// A dash pattern: dash, gap, dash… lengths in drawing units. +#[derive(Clone, Debug)] +pub struct LineType { + pub name: String, + pub pattern: Vec, +} + +/// A placed image file and the image entities that show it. +#[derive(Clone, Debug)] +pub struct ImageDef { + pub name: String, + pub size: (u32, u32), + pub handle: Handle, + /// `(reactor, image entity)` per image entity. + pub reactors: Vec<(Handle, Handle)>, +} + +/// Everything around the entities, collected while they were written. +pub struct Drawing { + pub version: Format, + pub layers: Vec, + pub linetypes: Vec, + /// Text styles: (style name, font family). + pub styles: Vec<(String, String)>, + pub images: Vec, + /// The entities' group codes. + pub entities: String, + /// `[x0, y0, x1, y1]` of everything written (drawing units). + pub extents: [f64; 4], + /// The export region (drawing units). + pub limits: [f64; 4], + /// `$INSUNITS`. + pub insunits: u8, + pub metric: bool, + /// The next free handle. + pub next: u32, +} + +/// The text style every drawing has. +pub const STANDARD_STYLE: &str = "Standard"; + +impl Drawing { + /// The whole file. + pub fn write(self) -> String { + let v = self.version; + let mut c = Codes::new(v); + // Table entries get the handles after the entities'. + let mut next = self.next; + let mut take = |n: usize| { + let start = next; + next = next.saturating_add(u32::try_from(n).unwrap_or(u32::MAX)); + start + }; + let ltype_base = take(3 + self.linetypes.len()); + let layer_base = take(1 + self.layers.len()); + let style_base = take(1 + self.styles.len()); + self.header(&mut c, next); + if v.handles() { + self.classes(&mut c); + } + c.section("TABLES"); + self.vport(&mut c); + self.ltypes(&mut c, ltype_base); + self.layer_table(&mut c, layer_base); + self.style_table(&mut c, style_base); + if v.handles() { + for (name, h) in [("VIEW", fixed::VIEW_TABLE), ("UCS", fixed::UCS_TABLE)] { + c.table(name, h, 0); + c.end_table(); + } + c.table("APPID", fixed::APPID_TABLE, 1); + c.record("APPID", fixed::ACAD_APPID, fixed::APPID_TABLE, "AcDbRegAppTableRecord"); + c.raw(2, "ACAD"); + c.int(70, 0); + c.end_table(); + c.table("DIMSTYLE", fixed::DIMSTYLE_TABLE, 1); + c.record("DIMSTYLE", fixed::STANDARD_DIMSTYLE, fixed::DIMSTYLE_TABLE, "AcDbDimStyleTableRecord"); + c.raw(2, "Standard"); + c.int(70, 0); + c.end_table(); + c.table("BLOCK_RECORD", fixed::BLOCK_RECORD_TABLE, 2); + for (name, h, layout) in + [("*Model_Space", fixed::MODEL_RECORD, fixed::MODEL_LAYOUT), ("*Paper_Space", fixed::PAPER_RECORD, fixed::PAPER_LAYOUT)] + { + c.record("BLOCK_RECORD", h, fixed::BLOCK_RECORD_TABLE, "AcDbBlockTableRecord"); + c.raw(2, name); + if v.r2000() { + c.handle(340, layout); + c.int(70, 0); + c.int(280, 1); + c.int(281, 0); + } + } + c.end_table(); + } + c.end_section(); + if v.handles() { + c.section("BLOCKS"); + for (name, record, begin, end) in [ + ("*Model_Space", fixed::MODEL_RECORD, fixed::MODEL_BLOCK, fixed::MODEL_END), + ("*Paper_Space", fixed::PAPER_RECORD, fixed::PAPER_BLOCK, fixed::PAPER_END), + ] { + c.object("BLOCK", begin, record, "AcDbEntity"); + c.raw(8, "0"); + c.raw(100, "AcDbBlockBegin"); + c.raw(2, name); + c.int(70, 0); + c.xyz(10, 0.0, 0.0); + c.raw(3, name); + c.raw(1, ""); + c.object("ENDBLK", end, record, "AcDbEntity"); + c.raw(8, "0"); + c.raw(100, "AcDbBlockEnd"); + } + c.end_section(); + } + c.section("ENTITIES"); + c.append(&self.entities); + c.end_section(); + if v.handles() { + self.objects(&mut c); + } + c.raw(0, "EOF"); + c.into_string() + } + + fn header(&self, c: &mut Codes, handseed: u32) { + let v = self.version; + let var = |c: &mut Codes, name: &str| c.raw(9, name); + c.section("HEADER"); + var(c, "$ACADVER"); + c.raw(1, v.acadver()); + var(c, "$DWGCODEPAGE"); + c.raw(3, "ANSI_1252"); + var(c, "$INSBASE"); + c.xyz(10, 0.0, 0.0); + let [x0, y0, x1, y1] = self.extents; + var(c, "$EXTMIN"); + c.xyz(10, x0, y0); + var(c, "$EXTMAX"); + c.xyz(10, x1, y1); + let [lx0, ly0, lx1, ly1] = self.limits; + var(c, "$LIMMIN"); + c.xy(10, lx0, ly0); + var(c, "$LIMMAX"); + c.xy(10, lx1, ly1); + if v.r2000() { + var(c, "$LWDISPLAY"); + c.int(290, 1); + var(c, "$INSUNITS"); + c.int(70, i64::from(self.insunits)); + var(c, "$MEASUREMENT"); + c.int(70, i64::from(self.metric)); + } + if v.handles() { + var(c, "$HANDSEED"); + c.handle(5, Handle(handseed)); + } + c.end_section(); + } + + fn classes(&self, c: &mut Codes) { + let mut classes: Vec<(&str, &str, &str, i64, bool)> = vec![]; + if self.version.r2000() { + classes.extend([ + ("ACDBDICTIONARYWDFLT", "AcDbDictionaryWithDefault", "ObjectDBX Classes", 0, false), + ("ACDBPLACEHOLDER", "AcDbPlaceHolder", "ObjectDBX Classes", 0, false), + ("LAYOUT", "AcDbLayout", "ObjectDBX Classes", 0, false), + ]); + } + if !self.images.is_empty() { + classes.extend([ + ("RASTERVARIABLES", "AcDbRasterVariables", "ISM", 0, false), + ("IMAGE", "AcDbRasterImage", "ISM", 2175, true), + ("IMAGEDEF", "AcDbRasterImageDef", "ISM", 0, false), + ("IMAGEDEF_REACTOR", "AcDbRasterImageDefReactor", "ISM", 1, false), + ]); + } + c.section("CLASSES"); + for (name, cpp, app, flags, entity) in classes { + c.raw(0, "CLASS"); + c.raw(1, name); + c.raw(2, cpp); + c.raw(3, app); + c.int(90, flags); + if self.version.true_color() { + c.int(91, 0); + } + c.int(280, 0); + c.int(281, i64::from(entity)); + } + c.end_section(); + } + + /// The active viewport, zoomed to the drawing's extents. + fn vport(&self, c: &mut Codes) { + let [x0, y0, x1, y1] = self.extents; + c.table("VPORT", fixed::VPORT_TABLE, 1); + c.record("VPORT", fixed::ACTIVE_VPORT, fixed::VPORT_TABLE, "AcDbViewportTableRecord"); + c.raw(2, if self.version.handles() { "*Active" } else { "*ACTIVE" }); + c.int(70, 0); + c.xy(10, 0.0, 0.0); + c.xy(11, 1.0, 1.0); + c.xy(12, (x0 + x1) / 2.0, (y0 + y1) / 2.0); + c.xy(13, 0.0, 0.0); + c.xy(14, 1.0, 1.0); + c.xy(15, 10.0, 10.0); + // The view direction looks down the z axis: (0, 0, 1), each code once. + c.xy(16, 0.0, 0.0); + c.num(36, 1.0); + c.xyz(17, 0.0, 0.0); + let (w, h) = ((x1 - x0).abs(), (y1 - y0).abs()); + // The view's height, with a margin; its width follows from the aspect ratio. + c.num(40, (h.max(w / 1.5) * 1.1).max(1.0)); + c.num(41, 1.5); + c.num(42, 50.0); + c.num(43, 0.0); + c.num(44, 0.0); + c.num(50, 0.0); + c.num(51, 0.0); + for (code, value) in [(71, 0), (72, 1000), (73, 1), (74, 3), (75, 0), (76, 0), (77, 0), (78, 0)] { + c.int(code, value); + } + if self.version.r2000() { + c.int(281, 0); + c.int(65, 0); + c.num(146, 0.0); + } + c.end_table(); + } + + fn ltypes(&self, c: &mut Codes, base: u32) { + c.table("LTYPE", fixed::LTYPE_TABLE, 3 + self.linetypes.len()); + let solid = [("ByBlock", ""), ("ByLayer", ""), ("Continuous", "Solid line")]; + let rows = solid.iter().map(|(n, d)| (*n, *d, &[][..])).chain(self.linetypes.iter().map(|l| (l.name.as_str(), "", l.pattern.as_slice()))); + for (i, (name, desc, pattern)) in rows.enumerate() { + c.record("LTYPE", Handle(base.saturating_add(i as u32)), fixed::LTYPE_TABLE, "AcDbLinetypeTableRecord"); + c.str(2, name); + c.int(70, 0); + c.str(3, if pattern.is_empty() { desc } else { "Dashed" }); + c.int(72, 65); + c.int(73, pattern.len() as i64); + c.num(40, pattern.iter().map(|v| v.abs()).sum()); + for v in pattern { + c.num(49, *v); + if self.version.handles() { + c.int(74, 0); + } + } + } + c.end_table(); + } + + fn layer_table(&self, c: &mut Codes, base: u32) { + let zero = Layer { name: "0".into(), aci: 7, off: false, locked: false, plot: true }; + c.table("LAYER", fixed::LAYER_TABLE, 1 + self.layers.len()); + for (i, l) in std::iter::once(&zero).chain(&self.layers).enumerate() { + c.record("LAYER", Handle(base.saturating_add(i as u32)), fixed::LAYER_TABLE, "AcDbLayerTableRecord"); + c.str(2, &l.name); + c.int(70, if l.locked { 4 } else { 0 }); + // A negative colour switches the layer off. + let aci = i64::from(l.aci.max(1)); + c.int(62, if l.off { -aci } else { aci }); + c.raw(6, "Continuous"); + if self.version.r2000() { + c.int(290, i64::from(l.plot)); + c.int(370, -3); + c.handle(390, fixed::PLOT_NORMAL); + } + } + c.end_table(); + } + + fn style_table(&self, c: &mut Codes, base: u32) { + c.table("STYLE", fixed::STYLE_TABLE, 1 + self.styles.len()); + let standard = (STANDARD_STYLE.to_string(), String::new()); + for (i, (name, family)) in std::iter::once(&standard).chain(&self.styles).enumerate() { + c.record("STYLE", Handle(base.saturating_add(i as u32)), fixed::STYLE_TABLE, "AcDbTextStyleTableRecord"); + c.str(2, name); + c.int(70, 0); + c.num(40, 0.0); + c.num(41, 1.0); + c.num(50, 0.0); + c.int(71, 0); + c.num(42, 2.5); + // The font file; from 2000 the family name too, which is how CAD apps find a TrueType font. + if family.is_empty() { + c.raw(3, "txt"); + } else { + c.str(3, &format!("{family}.ttf")); + } + c.raw(4, ""); + if self.version.r2000() && !family.is_empty() { + c.raw(1001, "ACAD"); + c.str(1000, family); + c.int(1071, 0); + } + } + c.end_table(); + } + + fn objects(&self, c: &mut Codes) { + let v = self.version; + let images = !self.images.is_empty(); + c.section("OBJECTS"); + let dict = |c: &mut Codes, h: Handle, owner: Handle, entries: &[(&str, Handle)]| { + c.object("DICTIONARY", h, owner, "AcDbDictionary"); + c.int(281, 1); + for (name, e) in entries { + c.str(3, name); + c.handle(350, *e); + } + }; + let mut root = vec![("ACAD_GROUP", fixed::GROUPS)]; + if images { + root.extend([("ACAD_IMAGE_DICT", fixed::IMAGE_DICT), ("ACAD_IMAGE_VARS", fixed::IMAGE_VARS)]); + } + if v.r2000() { + root.extend([("ACAD_LAYOUT", fixed::LAYOUTS), ("ACAD_PLOTSTYLENAME", fixed::PLOT_STYLES)]); + } + dict(c, fixed::ROOT, Handle(0), &root); + dict(c, fixed::GROUPS, fixed::ROOT, &[]); + if v.r2000() { + dict(c, fixed::LAYOUTS, fixed::ROOT, &[("Layout1", fixed::PAPER_LAYOUT), ("Model", fixed::MODEL_LAYOUT)]); + c.object("ACDBDICTIONARYWDFLT", fixed::PLOT_STYLES, fixed::ROOT, "AcDbDictionary"); + c.int(281, 1); + c.raw(3, "Normal"); + c.handle(350, fixed::PLOT_NORMAL); + c.raw(100, "AcDbDictionaryWithDefault"); + c.handle(340, fixed::PLOT_NORMAL); + // A placeholder has no data and no subclass marker of its own. + c.raw(0, "ACDBPLACEHOLDER"); + c.handle(5, fixed::PLOT_NORMAL); + c.handle(330, fixed::PLOT_STYLES); + self.layout(c, fixed::MODEL_LAYOUT, "Model", 0, fixed::MODEL_RECORD); + self.layout(c, fixed::PAPER_LAYOUT, "Layout1", 1, fixed::PAPER_RECORD); + } + if images { + c.object("RASTERVARIABLES", fixed::IMAGE_VARS, fixed::ROOT, "AcDbRasterVariables"); + c.int(90, 0); + c.int(70, 0); + c.int(71, 1); + c.int(72, 0); + let keys: Vec = self.images.iter().map(|i| format!("IMAGE_{:X}", i.handle.0)).collect(); + let entries: Vec<(&str, Handle)> = keys.iter().zip(&self.images).map(|(key, image)| (key.as_str(), image.handle)).collect(); + dict(c, fixed::IMAGE_DICT, fixed::ROOT, &entries); + for img in &self.images { + c.raw(0, "IMAGEDEF"); + c.handle(5, img.handle); + c.raw(102, "{ACAD_REACTORS"); + c.handle(330, fixed::IMAGE_DICT); + for (reactor, _) in &img.reactors { + c.handle(330, *reactor); + } + c.raw(102, "}"); + c.handle(330, fixed::IMAGE_DICT); + c.raw(100, "AcDbRasterImageDef"); + c.int(90, 0); + c.str(1, &img.name); + c.xy(10, f64::from(img.size.0), f64::from(img.size.1)); + c.xy(11, 1.0, 1.0); + c.int(280, 1); + c.int(281, 0); + for (reactor, image) in &img.reactors { + c.object("IMAGEDEF_REACTOR", *reactor, *image, "AcDbRasterImageDefReactor"); + c.int(90, 2); + c.handle(330, *image); + } + } + } + c.end_section(); + } + + /// A layout (DXF 2000 and later): its plot settings, then the layout itself. + fn layout(&self, c: &mut Codes, h: Handle, name: &str, tab: i64, record: Handle) { + let [x0, y0, x1, y1] = self.extents; + c.object("LAYOUT", h, fixed::LAYOUTS, "AcDbPlotSettings"); + c.raw(1, ""); + c.raw(2, "none_device"); + c.raw(4, ""); + c.raw(6, ""); + for code in 40..=49 { + c.num(code, 0.0); + } + c.num(140, 0.0); + c.num(141, 0.0); + c.num(142, 1.0); + c.num(143, 1.0); + // Plot flags; 1024 marks the model space layout. + c.int(70, if record == fixed::MODEL_RECORD { 688 | 1024 } else { 688 }); + c.int(72, 1); + c.int(73, 0); + c.int(74, 5); + c.raw(7, ""); + c.int(75, 16); + if self.version.true_color() { + // Shade plot mode (as displayed), resolution level (normal) and custom DPI. + c.int(76, 0); + c.int(77, 2); + c.int(78, 300); + } + c.num(147, 1.0); + // The paper image origin is 148 (x) and 149 (y), not a point at 148 and 158. + c.num(148, 0.0); + c.num(149, 0.0); + c.raw(100, "AcDbLayout"); + c.raw(1, name); + c.int(70, 1); + c.int(71, tab); + let [lx0, ly0, lx1, ly1] = self.limits; + c.xy(10, lx0, ly0); + c.xy(11, lx1, ly1); + c.xyz(12, 0.0, 0.0); + c.xyz(14, x0, y0); + c.xyz(15, x1, y1); + c.num(146, 0.0); + c.xyz(13, 0.0, 0.0); + c.xyz(16, 1.0, 0.0); + c.xyz(17, 0.0, 1.0); + c.int(76, 1); + c.handle(330, record); + } +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn numbers_and_text_are_written_plainly() { + assert_eq!(num(1.5), "1.5"); + assert_eq!(num(-0.0000001), "0"); + assert_eq!(num(2.0), "2"); + assert_eq!(num(f64::NAN), "0"); + assert_eq!(text("a^b\nc", true), "a^ b c"); + assert_eq!(text("GrĂ¶ĂŸe đŸ˜€", false), "Gr\\U+00F6\\U+00DFe ?"); + assert_eq!(text("GrĂ¶ĂŸe", true), "GrĂ¶ĂŸe"); + } + + /// A line from (0, 0) to (10, 20) on layer `Layer1`, in the style scene.rs writes entities. + fn line_entities(v: Format) -> String { + let mut e = Codes::new(v); + let style = EntityStyle { aci: 1, rgb: Some(0xFF0000), lineweight: Some(25), linetype: None, alpha: None }; + e.entity("LINE", Handle(fixed::FIRST_FREE), fixed::MODEL_RECORD, "Layer1", &style, "AcDbLine"); + e.xyz(10, 0.0, 0.0); + e.xyz(11, 10.0, 20.0); + e.into_string() + } + + fn sample(v: Format, with_line: bool) -> Drawing { + Drawing { + version: v, + layers: vec![Layer { name: "Layer1".into(), aci: 1, off: false, locked: false, plot: true }], + linetypes: vec![], + styles: vec![], + images: vec![], + entities: if with_line { line_entities(v) } else { String::new() }, + extents: [0.0, 0.0, 200.0, 100.0], + limits: [0.0, 0.0, 200.0, 100.0], + insunits: 4, + metric: true, + next: fixed::FIRST_FREE + if with_line { 1 } else { 0 }, + } + } + + fn assert_golden(name: &str, got: &str) { + let path = PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("tests/data").join(name); + if std::env::var_os("WRITE_DXF_GOLDENS").is_some() { + std::fs::create_dir_all(path.parent().expect("parent")).expect("mkdir"); + std::fs::write(&path, got).expect("write golden"); + return; + } + let expected = std::fs::read_to_string(&path).unwrap_or_else(|e| panic!("read {}: {e}", path.display())); + assert_eq!(got, expected, "bytes changed versus {}", path.display()); + } + + use std::path::PathBuf; + + #[test] + fn empty_and_line_drawings_match_goldens() { + for (name, v, line) in [ + ("drawing_empty_r12.dxf", Format::R12, false), + ("drawing_empty_r2018.dxf", Format::R2018, false), + ("drawing_line_r12.dxf", Format::R12, true), + ("drawing_line_r2000.dxf", Format::R2000, true), + ("drawing_line_r2018.dxf", Format::R2018, true), + ] { + assert_golden(name, &sample(v, line).write()); + } + } + + #[test] + fn entity_preserves_byblock_colour() { + for v in Format::ALL { + let mut c = Codes::new(v); + c.entity("LINE", Handle(fixed::FIRST_FREE), fixed::MODEL_RECORD, "0", &EntityStyle::default(), "AcDbLine"); + assert!(c.into_string().contains(" 62\n0\n"), "{v:?}"); + } + } + + #[test] + fn image_dictionary_keys_are_independent_of_filenames() { + let mut drawing = sample(Format::R2018, false); + for (index, name) in [r"C:\images\scan,1.png", "scan,1.png"].into_iter().enumerate() { + drawing.images.push(ImageDef { name: name.into(), size: (10, 20), handle: Handle(fixed::FIRST_FREE + index as u32), reactors: vec![] }); + } + drawing.next = fixed::FIRST_FREE + 2; + let images = drawing.images.clone(); + let output = drawing.write(); + for image in &images { + assert!(output.contains(&format!(" 3\nIMAGE_{:X}\n350\n{:X}\n", image.handle.0, image.handle.0))); + assert!(output.contains(&format!(" 1\n{}\n", image.name))); + assert!(!output.contains(&format!(" 3\n{}\n", image.name))); + } + } + + #[test] + fn every_format_writes_its_acadver_and_ends_at_eof() { + for v in Format::ALL { + let s = sample(v, true).write(); + assert!(s.contains(&format!(" 1\n{}\n", v.acadver())), "{v:?}"); + assert!(s.ends_with(" 0\nEOF\n"), "{v:?}"); + if v.handles() { + assert!(s.contains(" 0\nSECTION\n 2\nCLASSES\n"), "{v:?}"); + } else { + assert!(!s.contains("\nCLASSES\n"), "{v:?}"); + } + } + } +} diff --git a/crates/dxf/src/lib.rs b/crates/dxf/src/lib.rs index 0109e9040..c4595bef9 100644 --- a/crates/dxf/src/lib.rs +++ b/crates/dxf/src/lib.rs @@ -3,14 +3,19 @@ //! DXF is a sequence of (group code, value) pairs. This crate parses ASCII and binary DXF into a //! flat list of [`Tag`]s, splits them into sections and entities, and writes tags back. [`model`] //! reads the tags as a drawing (header variables, tables, blocks, entities, objects); mapping -//! that onto an application's own document is left to the caller. Written from Autodesk's +//! that onto an application's own document is left to the caller. [`drawing`] writes an ASCII +//! file around a pre-built `ENTITIES` section (VectorCraft's writer). Written from Autodesk's //! published DXF Reference; depends on no other Craft crate. Moved from CADCraft's `cadcraft-dxf`. #![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] #![forbid(unsafe_code)] use std::io::Write as _; +pub mod drawing; pub mod model; +pub mod mtext; +pub mod mtext_tokens; +pub mod paths; mod version; pub use version::Version; @@ -194,8 +199,9 @@ pub use codepage::decode_code_page; /// # Errors /// /// Returns an error for empty input, an invalid group code, an incomplete group-code/value pair, -/// truncated binary data, or more than [`MAX_TAGS`] group codes. For ASCII input, invalid numeric -/// text becomes `0.0` for real values or `0` for integer values, and an EOF tag is not required. +/// truncated binary data, or more group codes than the limit returned by [`max_tags()`], which is +/// [`MAX_TAGS`] by default. For ASCII input, invalid numeric text becomes `0.0` for real values or +/// `0` for integer values, and an EOF tag is not required. pub fn parse(bytes: &[u8]) -> Result> { parse_with_limit(bytes, max_tags()) } diff --git a/crates/dxf/src/model/write.rs b/crates/dxf/src/model/write.rs index 3ee234153..6d2879cd9 100644 --- a/crates/dxf/src/model/write.rs +++ b/crates/dxf/src/model/write.rs @@ -1478,6 +1478,13 @@ const MODEL: &str = "*Model_Space"; const PAPER: &str = "*Paper_Space"; impl Drawing<'_> { + /// Set `$LUNITS` without making callers construct low-level group tags. + pub fn with_linear_units(mut self, units: i64) -> Self { + self.header.retain(|(name, _)| !name.eq_ignore_ascii_case("$LUNITS")); + self.header.push(("$LUNITS".into(), vec![Tag::i(70, units)])); + self + } + /// The whole file as tags, ending with `0 EOF`, and how many entities were left out /// ([`EntityKind::Other`], and a MULTILEADER before 2007 or an ACAD_TABLE before 2004, which it has no groups for: pass /// those in `raw_entities`). diff --git a/crates/dxf/src/mtext.rs b/crates/dxf/src/mtext.rs new file mode 100644 index 000000000..9b113aa6c --- /dev/null +++ b/crates/dxf/src/mtext.rs @@ -0,0 +1,103 @@ +//! MTEXT inline-escape handling independent of drawing and text layout. +//! +//! Unicode and MIF escapes are decoded to literal characters, with markup-significant +//! characters re-escaped so subsequent MTEXT parsers cannot treat them as commands. + +/// Whether an inline command takes an argument terminated by `;`. +pub fn takes_arg(code: char) -> bool { + matches!(code, 'f' | 'F' | 'H' | 'W' | 'Q' | 'T' | 'A' | 'C' | 'c' | 'p') +} + +/// Four hex digits at `chars[at..at + 4]`. +fn hex4(chars: &[char], at: usize) -> Option { + let digits = chars.get(at..at.checked_add(4)?)?; + digits.iter().try_fold(0u32, |acc, c| Some(acc * 16 + c.to_digit(16)?)) +} + +/// Push a decoded character so that the tokenizers read it as text, never as markup. +fn push_literal(out: &mut String, c: char) { + if matches!(c, '\\' | '{' | '}') { + out.push('\\'); + } + out.push(c); +} + +/// Decode `\U+hhhh` (one UTF-16 unit; a surrogate pair gives one character, a lone surrogate +/// U+FFFD) and `\M+nhhhh` (double-byte East Asian characters) to the characters they stand +/// for. Malformed escapes remain unchanged for the caller to interpret. +/// An escaped backslash (`\\U+0041`) is not an escape. +pub fn unescape_codes(s: &str) -> std::borrow::Cow<'_, str> { + if !s.contains("\\U+") && !s.contains("\\M+") { + return std::borrow::Cow::Borrowed(s); + } + let chars: Vec = s.chars().collect(); + let mut out = String::with_capacity(s.len()); + let mut i = 0; + while let Some(&c) = chars.get(i) { + if c != '\\' { + out.push(c); + i += 1; + continue; + } + let (n, plus) = (chars.get(i + 1).copied(), chars.get(i + 2).copied()); + if n == Some('U') && plus == Some('+') { + if let Some(unit) = hex4(&chars, i + 3) { + i += 7; + let low = |at: usize| { + (chars.get(at) == Some(&'\\') && chars.get(at + 1) == Some(&'U') && chars.get(at + 2) == Some(&'+')) + .then(|| hex4(&chars, at + 3)) + .flatten() + .filter(|u| (0xDC00..0xE000).contains(u)) + }; + let ch = if (0xD800..0xDC00).contains(&unit) { + match low(i) { + Some(lo) => { + i += 7; + char::from_u32(0x10000 + ((unit - 0xD800) << 10) + (lo - 0xDC00)) + } + None => None, + } + } else { + char::from_u32(unit) + }; + push_literal(&mut out, ch.unwrap_or('\u{fffd}')); + continue; + } + } else if n == Some('M') && plus == Some('+') && chars.get(i + 3).is_some_and(char::is_ascii_digit) && hex4(&chars, i + 4).is_some() { + push_literal(&mut out, crate::mif_escape_at(&chars, i).unwrap_or('\u{fffd}')); + i += 8; + continue; + } + // Any other backslash pair stays as it is (and an escaped backslash stays one). + out.push('\\'); + i += 1; + if let Some(&next) = chars.get(i) { + out.push(next); + i += 1; + } + } + std::borrow::Cow::Owned(out) +} + +#[cfg(test)] +mod tests { + use super::*; + #[test] + fn unicode_and_mif_escapes_remain_literal_markup() { + assert_eq!(unescape_codes(r"a\U+0041\U+D83D\U+DE00"), "aAđŸ˜€"); + assert_eq!(unescape_codes(r"\M+5BCFE"), "ä»¶"); + assert_eq!(unescape_codes(r"\U+005C\U+007B\U+007D"), r"\\\{\}"); + assert_eq!(unescape_codes(r"\\U+0041"), r"\\U+0041"); + assert_eq!(unescape_codes(r"\U+12"), r"\U+12"); + assert_eq!(unescape_codes(r"\U+D800"), "�"); + } + #[test] + fn parameterized_commands_are_classified_without_relaxing_others() { + for c in "fFHWQTACcp".chars() { + assert!(takes_arg(c)); + } + for c in ['S', 'P', 'Z', '\\'] { + assert!(!takes_arg(c)); + } + } +} diff --git a/crates/dxf/src/mtext_tokens.rs b/crates/dxf/src/mtext_tokens.rs new file mode 100644 index 000000000..b07a9c2e0 --- /dev/null +++ b/crates/dxf/src/mtext_tokens.rs @@ -0,0 +1,132 @@ +//! Streaming lexical tokens for MTEXT inline markup. +//! +//! Apply [`crate::mtext::unescape_codes`] first when Unicode/MIF escapes should become literal +//! characters. Tokens borrow their source and scanning allocates no memory. The consumer owns +//! formatting, line layout, paragraph policy and how unknown commands are displayed. + +/// How a stacked-text argument finds its closing semicolon. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum StackMode { + /// The first semicolon ends the stack, as plain-text extraction historically did. + Plain, + /// A backslash quotes the next character, as the structured layout parser does. + Structured, +} +/// One lexical unit. Unknown commands retain their letter so the caller chooses compatibility. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum Token<'a> { + Char(char), + OpenGroup, + CloseGroup, + Control(char), + Argument(char, &'a str), + Stack(&'a str), + Unknown(char), +} +/// A bounded-memory walk of UTF-8 MTEXT markup. +pub struct Tokens<'a> { + source: &'a str, + offset: usize, + stack_mode: StackMode, +} +impl<'a> Tokens<'a> { + pub fn new(source: &'a str, stack_mode: StackMode) -> Self { + Self { source, offset: 0, stack_mode } + } + fn next_char(&mut self) -> Option { + let c = self.source.get(self.offset..)?.chars().next()?; + self.offset += c.len_utf8(); + Some(c) + } + fn argument(&mut self, structured: bool) -> &'a str { + let start = self.offset; + let mut end = start; + while let Some(c) = self.next_char() { + if c == ';' { + return self.source.get(start..end).unwrap_or_default(); + } + if structured && c == '\\' { + let _ = self.next_char(); + } + end = self.offset; + } + self.source.get(start..end).unwrap_or_default() + } +} +impl<'a> Iterator for Tokens<'a> { + type Item = Token<'a>; + fn next(&mut self) -> Option { + let c = self.next_char()?; + Some(match c { + '{' => Token::OpenGroup, + '}' => Token::CloseGroup, + '\\' => { + let code = self.next_char()?; + match code { + 'S' => Token::Stack(self.argument(self.stack_mode == StackMode::Structured)), + c if crate::mtext::takes_arg(c) => Token::Argument(c, self.argument(false)), + c @ ('P' | 'N' | '~' | '\\' | '{' | '}' | 'L' | 'l' | 'O' | 'o' | 'K' | 'k') => Token::Control(c), + c => Token::Unknown(c), + } + } + c => Token::Char(c), + }) + } +} +/// Split a structured stack into top, bottom and separator, unquoting escaped characters. +pub fn split_stack(raw: &str) -> (String, String, Option) { + let (mut top, mut bottom, mut separator) = (String::new(), String::new(), None); + let mut chars = raw.chars(); + while let Some(c) = chars.next() { + if c == '\\' { + if let Some(c) = chars.next() { + if separator.is_none() { + top.push(c); + } else { + bottom.push(c); + } + } + } else if separator.is_none() && matches!(c, '/' | '#' | '^') { + separator = Some(c); + } else if separator.is_none() { + top.push(c); + } else { + bottom.push(c); + } + } + (top, bottom, separator) +} +#[cfg(test)] +mod tests { + use super::*; + #[test] + fn tokens_borrow_text_and_preserve_unknown_commands() { + let got: Vec<_> = Tokens::new(r"A\Z\P{\H2x;x}\S1/2;", StackMode::Structured).collect(); + assert_eq!( + got, + [ + Token::Char('A'), + Token::Unknown('Z'), + Token::Control('P'), + Token::OpenGroup, + Token::Argument('H', "2x"), + Token::Char('x'), + Token::CloseGroup, + Token::Stack("1/2") + ] + ); + assert_eq!(split_stack("1/2"), ("1".into(), "2".into(), Some('/'))); + } + #[test] + fn stacked_text_uses_the_requested_terminator_rules() { + let text = r"\Sa\;b/c;X"; + assert_eq!(Tokens::new(text, StackMode::Plain).next(), Some(Token::Stack(r"a\"))); + assert_eq!(Tokens::new(text, StackMode::Structured).next(), Some(Token::Stack(r"a\;b/c"))); + assert_eq!(split_stack(r"a\;b/c"), ("a;b".into(), "c".into(), Some('/'))); + } + #[test] + fn missing_terminator_and_trailing_backslash_are_bounded() { + assert_eq!(Tokens::new("\\H2x", StackMode::Structured).collect::>(), [Token::Argument('H', "2x")]); + assert_eq!(Tokens::new("a\\", StackMode::Structured).collect::>(), [Token::Char('a')]); + } +} diff --git a/crates/dxf/src/paths.rs b/crates/dxf/src/paths.rs new file mode 100644 index 000000000..7e4f18f0f --- /dev/null +++ b/crates/dxf/src/paths.rs @@ -0,0 +1,859 @@ +//! Bounded conversion of common planar DXF entities into neutral BĂ©zier paths. +//! +//! This is geometry extraction, not a full CAD scene renderer: unsupported entities such as +//! text, dimensions, meshes, and block inserts are reported as warnings and omitted. + +use kurbo::{BezPath, Point, Rect, Shape as _}; + +/// Limits on the number of paths returned by [`import`]. Parsing also uses the crate's tag cap. +pub const MAX_PATHS: usize = 200_000; +/// Aggregate number of BĂ©zier elements a drawing may generate during conversion. +pub const MAX_PATH_ELEMENTS: usize = 2_000_000; + +#[derive(Debug, thiserror::Error)] +pub enum Error { + #[error("DXF parse failed: {0}")] + Parse(#[from] crate::DxfError), + #[error("DXF contains more than {MAX_PATHS} paths")] + TooManyPaths, + #[error("DXF curve exceeds the subdivision limit")] + TooManySegments, + #[error("DXF contains a non-finite path coordinate")] + NonFinite, +} + +/// Geometry in source DXF coordinates. `unit_to_mm` maps a source coordinate to millimetres; +/// unitless drawings use `1.0`, as the coordinates are then interpreted as millimetres. +#[derive(Debug, Clone)] +pub struct Imported { + pub paths: Vec, + pub bounds: Option, + pub unit_to_mm: f64, + pub warnings: Vec, +} + +/// Parse bounded DXF input and extract visible planar linework. +pub fn import(bytes: &[u8]) -> Result { + let tags = crate::parse_with_limit(bytes, crate::max_tags())?; + let drawing = crate::model::Document::read(&tags); + let unit_to_mm = units_to_mm(drawing.header.insunits(), drawing.header.measurement()); + let layer_visibility: std::collections::HashMap<_, _> = + drawing.layers().into_iter().map(|layer| (layer.name, layer.on && !layer.frozen)).collect(); + let mut paths = Vec::new(); + let mut warnings = Vec::new(); + let mut unsupported = 0usize; + let mut elements = 0usize; + for record in &drawing.entities { + let entity = crate::model::Entity::read(record); + if !entity.common.visible || record.paper_space() || layer_visibility.get(&entity.common.layer) == Some(&false) { + continue; + } + if excessive_sweep(&entity.kind) { + return Err(Error::TooManySegments); + } + let remaining = MAX_PATH_ELEMENTS.saturating_sub(elements); + if estimated_elements(&entity.kind) > remaining { + return Err(Error::TooManySegments); + } + let converted = if let crate::model::EntityKind::Spline(spline) = &entity.kind { + spline_path(spline)?.into_iter().collect() + } else { + entity_paths(&entity.kind, entity.common.extrusion) + }; + if converted.iter().flat_map(|path| path.elements()).any(|element| { + let finite = |p: Point| p.x.is_finite() && p.y.is_finite(); + match *element { + kurbo::PathEl::MoveTo(p) | kurbo::PathEl::LineTo(p) => !finite(p), + kurbo::PathEl::QuadTo(a, b) => !finite(a) || !finite(b), + kurbo::PathEl::CurveTo(a, b, c) => !finite(a) || !finite(b) || !finite(c), + kurbo::PathEl::ClosePath => false, + } + }) { + return Err(Error::NonFinite); + } + if converted.is_empty() { + unsupported = unsupported.saturating_add(1); + continue; + } + elements = elements.saturating_add(converted.iter().map(|path| path.elements().len()).sum::()); + if elements > MAX_PATH_ELEMENTS { + return Err(Error::TooManySegments); + } + if paths.len().saturating_add(converted.len()) > MAX_PATHS { + return Err(Error::TooManyPaths); + } + paths.extend(converted); + } + if unsupported > 0 { + warnings.push(format!("{unsupported} visible DXF entities are unsupported and were omitted")); + } + if drawing.skipped.iter().any(|(_, n)| *n > 0) { + warnings.push("some unsupported or stray DXF records were skipped".into()); + } + let bounds = paths.iter().map(|path| path.bounding_box()).reduce(|a, b| a.union(b)); + Ok(Imported { paths, bounds, unit_to_mm, warnings }) +} + +fn estimated_elements(kind: &crate::model::EntityKind) -> usize { + use crate::model::EntityKind as E; + let arc_count = |sweep: f64| (sweep.abs() / std::f64::consts::FRAC_PI_2).ceil().max(1.0) as usize; + let bulge_count = |bulge: f64| { + if bulge.is_finite() && bulge.abs() > 1e-12 { arc_count(4.0 * bulge.atan()) } else { 1 } + }; + let polyline_count = |vertices: &[(Point, f64)], closed: bool| { + if vertices.is_empty() { + return 0; + } + let mut count = 1usize; + for pair in vertices.windows(2) { + count = count.saturating_add(bulge_count(pair[0].1)); + } + if closed && vertices.len() > 1 { + count = count.saturating_add(bulge_count(vertices.last().map_or(0.0, |vertex| vertex.1))); + count = count.saturating_add(1); + } + count + }; + match kind { + E::Line { .. } => 2, + E::Circle { .. } => 6, + E::Arc { start_angle, end_angle, .. } => 1usize.saturating_add(arc_count((end_angle - start_angle).to_radians())), + E::LwPolyline(poly) => { + let vertices: Vec<_> = poly.vertices.iter().map(|v| (Point::new(v.x, v.y), v.bulge)).collect(); + polyline_count(&vertices, poly.closed) + } + E::Polyline(poly) if !poly.is_mesh() && !poly.is_polyface() => { + let vertices: Vec<_> = poly + .vertices + .iter() + .filter(|vertex| !vertex.is_face()) + .map(|vertex| (Point::new(vertex.location[0], vertex.location[1]), vertex.bulge)) + .collect(); + polyline_count(&vertices, poly.closed()) + } + E::Ellipse { start, end, .. } => 1usize.saturating_add(arc_count(end - start)), + // `spline_path` already caps subdivision at MAX_SPLINE_ELEMENTS. Its actual output is + // checked against the drawing-wide budget after conversion. + E::Spline(_) => 0, + _ => 0, + } +} + +fn units_to_mm(code: i64, _measurement: Option) -> f64 { + match code { + 0 => 1.0, + 1 => 25.4, + 2 => 304.8, + 3 => 1_609_344.0, + 5 => 10.0, + 6 => 1_000.0, + 7 => 1_000_000.0, + 8 => 0.000_025_4, + 9 => 0.0254, + 10 => 914.4, + 11 => 1e-7, + 12 => 1e-6, + 13 => 0.001, + 14 => 100.0, + 15 => 10_000.0, + 16 => 100_000.0, + 17 => 1e12, + 18 => 149_597_870_700_000.0, + 19 => 9.460_730_472_580_8e18, + 20 => 3.085_677_581_491_367e19, + 21 => 1200.0 / 3937.0 * 1000.0, + 22 => 1200.0 / 3937.0 * 1000.0 / 12.0, + 23 => 1200.0 / 3937.0 * 1000.0 * 3.0, + 24 => 1200.0 / 3937.0 * 1000.0 * 5280.0, + _ => 1.0, + } +} + +fn entity_paths(kind: &crate::model::EntityKind, normal: crate::model::Point) -> Vec { + use crate::model::EntityKind as E; + match kind { + E::Line { start, end } => vec![line(*start, *end)], + E::Circle { center, radius } if radius.is_finite() && *radius > 0.0 => { + vec![map_ocs_path(circle(Point::new(center[0], center[1]), *radius), normal, center[2])] + } + E::Arc { center, radius, start_angle, end_angle } + if radius.is_finite() && *radius > 0.0 && start_angle.is_finite() && end_angle.is_finite() => + { + vec![map_ocs_path(arc(Point::new(center[0], center[1]), *radius, start_angle.to_radians(), end_angle.to_radians()), normal, center[2])] + } + E::LwPolyline(poly) => { + vec![map_ocs_path(polyline(poly.vertices.iter().map(|v| (Point::new(v.x, v.y), v.bulge)), poly.closed), normal, poly.elevation)] + } + E::Polyline(poly) if !poly.is_mesh() && !poly.is_polyface() => { + let is_3d = poly.flags & 8 != 0; + let z = if is_3d { 0.0 } else { poly.elevation }; + let path = + polyline(poly.vertices.iter().filter(|v| !v.is_face()).map(|v| (Point::new(v.location[0], v.location[1]), v.bulge)), poly.closed()); + vec![if is_3d { path } else { map_ocs_path(path, normal, z) }] + } + E::Ellipse { center, major, ratio, start, end } if ratio.is_finite() && *ratio > 0.0 => { + vec![ellipse(*center, *major, normal, *ratio, *start, *end)] + } + _ => Vec::new(), + } +} + +const MAX_SPLINE_ELEMENTS: usize = 200_000; +const SPLINE_TOLERANCE: f64 = 1e-6; +const MAX_SPLINE_DEPTH: u32 = 12; +const MAX_SPLINE_DEGREE: usize = 11; +const MIN_RELATIVE_WEIGHT: f64 = 1e-100; + +#[derive(Clone, Copy)] +struct Homogeneous { + x: f64, + y: f64, + w: f64, +} + +impl Homogeneous { + fn lerp(self, other: Self, t: f64) -> Self { + let s = 1.0 - t; + Self { x: self.x * s + other.x * t, y: self.y * s + other.y * t, w: self.w * s + other.w * t } + } + fn point(self) -> Option { + (self.w.abs() > 1e-300).then(|| Point::new(self.x / self.w, self.y / self.w)).filter(|point| point.x.is_finite() && point.y.is_finite()) + } +} + +// Keep the NURBS conversion semantics of the previous importer: exact polynomial Bezier +// spans through cubic degree, tolerance-bounded rational/high-degree spans, invalid-weight +// fallback, and fit-point fallback when control data is malformed. +fn spline_path(spline: &crate::model::Spline) -> Result, Error> { + let weights = (spline.flags & 4 != 0 && spline.weights.len() == spline.control.len()).then_some(spline.weights.as_slice()); + let mut limit_hit = false; + let converted = usize::try_from(spline.degree).ok().and_then(|degree| nurbs(degree, &spline.knots, &spline.control, weights, &mut limit_hit)); + let path = converted + .map(|mut path| { + if spline.flags & 1 != 0 { + path.close_path(); + } + path + }) + .or_else(|| { + let closed = spline.flags & 1 != 0; + // `through` emits one move, n-1 (open) or n (closed) cubics, and + // an additional close element for closed paths. Check before allocating + // its converted point vector. + let max_points = MAX_SPLINE_ELEMENTS.saturating_sub(if closed { 2 } else { 0 }); + if spline.fit.len() > max_points { + limit_hit = true; + return None; + } + through(&spline.fit.iter().map(|p| Point::new(p[0], p[1])).collect::>(), closed) + }); + if limit_hit { + return Err(Error::TooManySegments); + } + Ok(path) +} + +fn nurbs(degree: usize, knots: &[f64], control: &[[f64; 3]], weights: Option<&[f64]>, limit_hit: &mut bool) -> Option { + let n = control.len(); + if !(1..=MAX_SPLINE_DEGREE).contains(°ree) + || n <= degree + || knots.len() != n + degree + 1 + || knots.windows(2).any(|w| w[1] < w[0]) + || control.iter().any(|p| !p[0].is_finite() || !p[1].is_finite()) + || knots.iter().any(|k| !k.is_finite()) + { + return None; + } + let raw = |i: usize| weights.and_then(|w| w.get(i)).copied().filter(|w| *w > 0.0 && w.is_finite()).unwrap_or(1.0); + let max_w = (0..n).map(raw).fold(0.0, f64::max); + if max_w == 0.0 { + return None; + } + let weight = |i: usize| (raw(i) / max_w).max(MIN_RELATIVE_WEIGHT); + let rational = (0..n).any(|i| (weight(i) - weight(0)).abs() > 1e-12); + let (lo, hi) = control.iter().fold((Point::new(f64::MAX, f64::MAX), Point::new(f64::MIN, f64::MIN)), |(lo, hi), p| { + (Point::new(lo.x.min(p[0]), lo.y.min(p[1])), Point::new(hi.x.max(p[0]), hi.y.max(p[1]))) + }); + let tolerance = ((hi - lo).hypot() * SPLINE_TOLERANCE).max(1e-12); + let mut path = BezPath::new(); + for span in degree..n { + let (Some(&t0), Some(&t1)) = (knots.get(span), knots.get(span + 1)) else { continue }; + if t1 <= t0 { + continue; + } + let bezier: Vec = (0..=degree) + .map(|j| { + let args: Vec<_> = (0..degree).map(|k| if k < degree - j { t0 } else { t1 }).collect(); + blossom(degree, knots, control, span, &args, &weight) + }) + .collect::>()?; + let start = bezier.first()?.point()?; + let needs_new_subpath = path_endpoint(&path).is_none_or(|end| end.distance(start) > 1e-12 * (1.0 + start.to_vec2().hypot())); + if needs_new_subpath { + path.move_to(start); + } + // A rational degree-one span changes its parameterization with its weights, but its locus + // is still the straight segment between its endpoints. Preserve that exact geometry. + let polynomial: Option> = if rational && degree > 1 { None } else { bezier.iter().map(|p| p.point()).collect() }; + match polynomial.as_deref() { + Some([_, p]) => path.line_to(*p), + Some([_, q, p]) => path.quad_to(*q, *p), + Some([_, a, b, p]) => path.curve_to(*a, *b, *p), + _ => fit_spline(&mut path, &bezier, tolerance, 0, limit_hit)?, + } + if path.elements().len() > MAX_SPLINE_ELEMENTS { + *limit_hit = true; + return None; + } + } + (path.elements().len() > 1).then_some(path) +} + +fn path_endpoint(path: &BezPath) -> Option { + match path.elements().last()? { + kurbo::PathEl::MoveTo(p) | kurbo::PathEl::LineTo(p) => Some(*p), + kurbo::PathEl::QuadTo(_, p) => Some(*p), + kurbo::PathEl::CurveTo(_, _, p) => Some(*p), + kurbo::PathEl::ClosePath => None, + } +} + +fn blossom(degree: usize, knots: &[f64], control: &[[f64; 3]], span: usize, args: &[f64], weight: &impl Fn(usize) -> f64) -> Option { + // Only the active degree+1 control points are needed for one span. Convert this small window + // here instead of making a second vector proportional to an untrusted entity's full controls. + let first = span.checked_sub(degree)?; + let mut points: Vec<_> = control + .get(first..=span)? + .iter() + .enumerate() + .map(|(offset, p)| { + let w = weight(first + offset); + Homogeneous { x: p[0] * w, y: p[1] * w, w } + }) + .collect(); + for r in 1..=degree { + let u = *args.get(r - 1)?; + for j in (r..=degree).rev() { + let i = span - degree + j; + let (a, b) = (*knots.get(i)?, *knots.get(i + degree + 1 - r)?); + let alpha = if b - a > 1e-300 { (u - a) / (b - a) } else { 0.0 }; + let previous = *points.get(j - 1)?; + let current = points.get_mut(j)?; + *current = previous.lerp(*current, alpha); + } + } + points.get(degree).copied() +} + +fn eval_spline_bezier(bezier: &[Homogeneous], t: f64) -> Option<(Point, kurbo::Vec2)> { + let mut points = bezier.to_vec(); + let n = points.len().checked_sub(1)?; + for level in (1..=n).rev() { + if level == 1 { + let (a, b) = (*points.first()?, *points.get(1)?); + let p = a.lerp(b, t); + let point = p.point()?; + let k = n as f64; + let (dx, dy, dw) = (k * (b.x - a.x), k * (b.y - a.y), k * (b.w - a.w)); + let w2 = p.w * p.w; + return Some((point, kurbo::Vec2::new((dx * p.w - p.x * dw) / w2, (dy * p.w - p.y * dw) / w2))); + } + for i in 0..level { + let next = *points.get(i + 1)?; + let current = points.get_mut(i)?; + *current = current.lerp(next, t); + } + } + Some((points.first()?.point()?, kurbo::Vec2::ZERO)) +} + +fn fit_spline(path: &mut BezPath, bezier: &[Homogeneous], tolerance: f64, depth: u32, limit_hit: &mut bool) -> Option<()> { + let (p0, d0) = eval_spline_bezier(bezier, 0.0)?; + let (p3, d3) = eval_spline_bezier(bezier, 1.0)?; + let (c1, c2) = (p0 + d0 / 3.0, p3 - d3 / 3.0); + if ![c1, c2].iter().all(|p| p.x.is_finite() && p.y.is_finite()) { + path.line_to(p3); + return Some(()); + } + let candidate = [p0, c1, c2, p3]; + if cubic_error_bound(bezier, &candidate).is_some_and(|error| error <= tolerance) { + if path.elements().len() >= MAX_SPLINE_ELEMENTS { + *limit_hit = true; + return None; + } + path.curve_to(c1, c2, p3); + return Some(()); + } + if depth >= MAX_SPLINE_DEPTH || path.elements().len() >= MAX_SPLINE_ELEMENTS { + *limit_hit = true; + return None; + } + let (left, right) = split_homogeneous_bezier(bezier)?; + fit_spline(path, &left, tolerance, depth + 1, limit_hit)?; + fit_spline(path, &right, tolerance, depth + 1, limit_hit) +} + +fn split_homogeneous_bezier(bezier: &[Homogeneous]) -> Option<(Vec, Vec)> { + let count = bezier.len(); + if count < 2 { + return None; + } + let mut row = bezier.to_vec(); + let mut left = vec![*bezier.first()?; count]; + let mut right = vec![*bezier.last()?; count]; + left[0] = row[0]; + right[count - 1] = row[count - 1]; + for level in 1..count { + for index in 0..count - level { + row[index] = row[index].lerp(row[index + 1], 0.5); + } + left[level] = row[0]; + right[count - 1 - level] = row[count - 1 - level]; + } + Some((left, right)) +} + +// Bound the rational curve/cubic difference over the whole parameter interval. In Bernstein +// form, X(t)-W(t)C(t) has degree p+3, so its norm is bounded by the largest coefficient norm; +// W(t) is a convex combination of positive weights and is at least their minimum. +fn cubic_error_bound(bezier: &[Homogeneous], cubic: &[Point; 4]) -> Option { + let degree = bezier.len().checked_sub(1)?; + let total = degree.checked_add(3)?; + let min_weight = bezier.iter().map(|point| point.w).fold(f64::INFINITY, f64::min); + if !(min_weight > 0.0 && min_weight.is_finite()) { + return None; + } + let mut max_error_numerator: f64 = 0.0; + for k in 0..=total { + let mut x = 0.0; + let mut y = 0.0; + for (i, point) in bezier.iter().enumerate() { + let Some(j) = k.checked_sub(i).filter(|j| *j < 4) else { continue }; + let factor = binomial(degree, i) * binomial(3, j) / binomial(total, k); + x += (point.x - point.w * cubic[j].x) * factor; + y += (point.y - point.w * cubic[j].y) * factor; + } + max_error_numerator = max_error_numerator.max(x.hypot(y)); + } + Some(max_error_numerator / min_weight) +} + +fn binomial(n: usize, k: usize) -> f64 { + if k > n { + return 0.0; + } + let k = k.min(n - k); + (1..=k).fold(1.0, |value, i| value * (n + 1 - i) as f64 / i as f64) +} + +fn through(points: &[Point], closed: bool) -> Option { + let n = points.len(); + let first = *points.first()?; + if n < 2 { + return None; + } + let at = |i: isize| { + let len = n as isize; + let j = if closed { i.rem_euclid(len) } else { i.clamp(0, len - 1) }; + points.get(j as usize).copied().unwrap_or(first) + }; + let mut path = BezPath::new(); + path.move_to(first); + let segments = if closed { n } else { n - 1 }; + for i in 0..segments as isize { + let (p0, p1, p2, p3) = (at(i - 1), at(i), at(i + 1), at(i + 2)); + path.curve_to(p1 + (p2 - p0) / 6.0, p2 - (p3 - p1) / 6.0, p2); + } + if closed { + path.close_path(); + } + Some(path) +} + +fn map_ocs_path(path: BezPath, normal: crate::model::Point, z: f64) -> BezPath { + let [x, y, n] = crate::model::ocs_axes(normal); + let transform = kurbo::Affine::new([x[0], x[1], y[0], y[1], n[0] * z, n[1] * z]); + transform * path +} + +fn excessive_sweep(kind: &crate::model::EntityKind) -> bool { + let cap = 4096.0 * std::f64::consts::FRAC_PI_2; + let over = |sweep: f64| !sweep.is_finite() || sweep.abs() > cap; + match kind { + crate::model::EntityKind::Arc { start_angle, end_angle, .. } => over((end_angle - start_angle).to_radians()), + crate::model::EntityKind::Ellipse { start, end, .. } => over(end - start), + _ => false, + } +} + +fn line(start: crate::model::Point, end: crate::model::Point) -> BezPath { + let mut path = BezPath::new(); + path.move_to((start[0], start[1])); + path.line_to((end[0], end[1])); + path +} + +fn polyline(vertices: impl IntoIterator, closed: bool) -> BezPath { + let vertices = vertices.into_iter().collect::>(); + let mut path = BezPath::new(); + let Some((first, _)) = vertices.first().copied() else { return path }; + path.move_to(first); + for i in 1..vertices.len() { + let (a, bulge) = vertices[i - 1]; + let (b, _) = vertices[i]; + if bulge.is_finite() && bulge.abs() > 1e-12 { + append_bulge(&mut path, a, b, bulge); + } else { + path.line_to(b); + } + } + if closed && vertices.len() > 1 { + let (a, bulge) = *vertices.last().unwrap_or(&(first, 0.0)); + if bulge.is_finite() && bulge.abs() > 1e-12 { + append_bulge(&mut path, a, first, bulge); + } else { + path.line_to(first); + } + path.close_path(); + } + path +} + +fn append_bulge(path: &mut BezPath, a: Point, b: Point, bulge: f64) { + let chord = b - a; + let length = chord.hypot(); + if !(length.is_finite() && length > 0.0) { + path.line_to(b); + return; + } + let center = a + chord * 0.5 + kurbo::Vec2::new(-chord.y, chord.x) * ((1.0 - bulge * bulge) / (4.0 * bulge)); + let start = (a.y - center.y).atan2(a.x - center.x); + let sweep = 4.0 * bulge.atan(); + let radius = length * (1.0 + bulge * bulge) / (4.0 * bulge.abs()); + arc_segments(path, center, start, sweep, radius); +} + +fn circle(center: Point, radius: f64) -> BezPath { + let mut path = BezPath::new(); + path.move_to((center.x + radius, center.y)); + arc_segments(&mut path, center, 0.0, std::f64::consts::TAU, radius); + path.close_path(); + path +} + +fn arc(center: Point, radius: f64, start: f64, end: f64) -> BezPath { + let mut path = BezPath::new(); + path.move_to((center.x + radius * start.cos(), center.y + radius * start.sin())); + let sweep = if end >= start { end - start } else { end - start + std::f64::consts::TAU }; + arc_segments(&mut path, center, start, sweep, radius); + path +} + +fn arc_segments(path: &mut BezPath, center: Point, start: f64, sweep: f64, radius: f64) { + let count = (sweep.abs() / (std::f64::consts::FRAC_PI_2)).ceil().max(1.0) as usize; + if count > 4096 { + return; + } + let step = sweep / count as f64; + for i in 0..count { + let a = start + step * i as f64; + let b = a + step; + let k = 4.0 / 3.0 * (step / 4.0).tan(); + let p1 = Point::new(center.x + radius * (a.cos() - k * a.sin()), center.y + radius * (a.sin() + k * a.cos())); + let p2 = Point::new(center.x + radius * (b.cos() + k * b.sin()), center.y + radius * (b.sin() - k * b.cos())); + let p3 = Point::new(center.x + radius * b.cos(), center.y + radius * b.sin()); + path.curve_to(p1, p2, p3); + } +} + +fn ellipse(center: crate::model::Point, major: crate::model::Point, normal: crate::model::Point, ratio: f64, start: f64, end: f64) -> BezPath { + let minor = cross(normal, major); + let minor_len = minor.iter().map(|v| v * v).sum::().sqrt(); + if !(minor_len.is_finite() && minor_len > 0.0) { + return BezPath::new(); + } + let major_len = major.iter().map(|v| v * v).sum::().sqrt(); + let bx = [minor[0] / minor_len * major_len * ratio, minor[1] / minor_len * major_len * ratio, minor[2] / minor_len * major_len * ratio]; + let point = |t: f64| Point::new(center[0] + major[0] * t.cos() + bx[0] * t.sin(), center[1] + major[1] * t.cos() + bx[1] * t.sin()); + let mut path = BezPath::new(); + path.move_to(point(start)); + let sweep = if end >= start { end - start } else { end - start + std::f64::consts::TAU }; + let count = (sweep.abs() / std::f64::consts::FRAC_PI_2).ceil().max(1.0) as usize; + if count > 4096 { + return path; + } + let step = sweep / count as f64; + for i in 0..count { + let a = start + step * i as f64; + let b = a + step; + let k = 4.0 / 3.0 * (step / 4.0).tan(); + let derivative = |t: f64| Point::new(-major[0] * t.sin() + bx[0] * t.cos(), -major[1] * t.sin() + bx[1] * t.cos()); + let p1 = point(a) + derivative(a).to_vec2() * k; + let p2 = point(b) - derivative(b).to_vec2() * k; + path.curve_to(p1, p2, point(b)); + } + path +} + +fn cross(a: crate::model::Point, b: crate::model::Point) -> crate::model::Point { + [a[1] * b[2] - a[2] * b[1], a[2] * b[0] - a[0] * b[2], a[0] * b[1] - a[1] * b[0]] +} + +#[cfg(test)] +mod spline_tests { + use super::*; + use crate::model::Spline; + use kurbo::ParamCurveNearest; + + // The previous importer accepted cubic fits based on seven fixed probes, then bisected the + // parameter interval. Keep this test-only reference to compare real legacy output with the + // conservative fit used by craft-dxf. + fn prior_fit(path: &mut BezPath, bezier: &[Homogeneous], t0: f64, t1: f64, tolerance: f64, depth: u32) -> Option<()> { + let (p0, d0) = eval_spline_bezier(bezier, t0)?; + let (p3, d3) = eval_spline_bezier(bezier, t1)?; + let h = (t1 - t0) / 3.0; + let (c1, c2) = (p0 + d0 * h, p3 - d3 * h); + if ![c1, c2].iter().all(|point| point.x.is_finite() && point.y.is_finite()) { + path.line_to(p3); + return Some(()); + } + let cubic = kurbo::CubicBez::new(p0, c1, c2, p3); + let close = (1..8).all(|i| { + let t = f64::from(i) / 8.0; + eval_spline_bezier(bezier, t0 + (t1 - t0) * t).is_some_and(|(point, _)| kurbo::ParamCurve::eval(&cubic, t).distance(point) <= tolerance) + }); + if close || depth >= MAX_SPLINE_DEPTH || path.elements().len() >= MAX_SPLINE_ELEMENTS { + path.curve_to(c1, c2, p3); + return Some(()); + } + let midpoint = (t0 + t1) / 2.0; + prior_fit(path, bezier, t0, midpoint, tolerance, depth + 1)?; + prior_fit(path, bezier, midpoint, t1, tolerance, depth + 1) + } + + fn legacy_path(bezier: &[Homogeneous], tolerance: f64) -> BezPath { + let mut path = BezPath::new(); + let start = eval_spline_bezier(bezier, 0.0).expect("legacy start point").0; + path.move_to(start); + prior_fit(&mut path, bezier, 0.0, 1.0, tolerance, 0).expect("legacy fit succeeds"); + path + } + + fn sampled_path_distance(from: &BezPath, to: &BezPath, tolerance: f64) -> f64 { + let target: Vec<_> = to.segments().collect(); + from.segments() + .flat_map(|segment| (0..=32).map(move |i| kurbo::ParamCurve::eval(&segment, f64::from(i) / 32.0))) + .map(|point| target.iter().map(|segment| segment.nearest(point, tolerance / 100.0).distance_sq.sqrt()).fold(f64::INFINITY, f64::min)) + .fold(0.0_f64, f64::max) + } + + fn cubic() -> Spline { + Spline { + degree: 3, + flags: 8, + knots: vec![0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 1.0, 1.0], + control: vec![[0.0, 0.0, 0.0], [1.0, 2.0, 0.0], [2.0, 2.0, 0.0], [3.0, 0.0, 0.0]], + weights: Vec::new(), + fit: Vec::new(), + start_tangent: None, + end_tangent: None, + fit_tolerance: 0.0, + } + } + + fn xyz(points: &[Point]) -> Vec<[f64; 3]> { + points.iter().map(|point| [point.x, point.y, 0.0]).collect() + } + + #[test] + fn spline_path_follows_clamped_control_curve_and_bounds_work() { + let path = spline_path(&cubic()).unwrap().expect("valid cubic spline"); + assert_eq!(path.elements().len(), 2); + assert!(matches!(path.elements()[0], kurbo::PathEl::MoveTo(p) if p == Point::new(0.0, 0.0))); + assert!(matches!(path.elements()[1], kurbo::PathEl::CurveTo(a, b, p) + if a == Point::new(1.0, 2.0) && b == Point::new(2.0, 2.0) && p == Point::new(3.0, 0.0))); + assert_eq!(estimated_elements(&crate::model::EntityKind::Spline(cubic())), 0); + } + + #[test] + fn rational_closed_periodic_and_fit_point_splines_keep_compatibility() { + let mut rational_line = Spline { + degree: 1, + flags: 4, + knots: vec![0.0, 0.0, 1.0, 1.0], + control: vec![[2.0, 3.0, 0.0], [5.0, 7.0, 0.0]], + weights: vec![1.0, 100.0], + fit: Vec::new(), + start_tangent: None, + end_tangent: None, + fit_tolerance: 0.0, + }; + rational_line.flags |= 8; + let line = spline_path(&rational_line).unwrap().expect("rational degree-one spline"); + assert!(matches!(line.elements(), [kurbo::PathEl::MoveTo(a), kurbo::PathEl::LineTo(b)] + if *a == Point::new(2.0, 3.0) && *b == Point::new(5.0, 7.0))); + + let mut rational = cubic(); + rational.flags |= 4; + rational.weights = vec![1.0, 0.5, 0.5, 1.0]; + let rational_path = spline_path(&rational).unwrap().expect("rational spline"); + assert!(rational_path.elements().iter().any(|el| matches!(el, kurbo::PathEl::CurveTo(..)))); + + let mut closed = cubic(); + closed.flags |= 1; + assert!(matches!(spline_path(&closed).unwrap().unwrap().elements().last(), Some(kurbo::PathEl::ClosePath))); + let mut periodic = cubic(); + periodic.flags |= 2; + assert!(!matches!(spline_path(&periodic).unwrap().unwrap().elements().last(), Some(kurbo::PathEl::ClosePath))); + + let mut fit = cubic(); + fit.knots.clear(); + fit.fit = vec![[0.0, 0.0, 0.0], [1.0, 2.0, 0.0], [3.0, 0.0, 0.0]]; + let path = spline_path(&fit).unwrap().expect("fit-point fallback"); + assert!(matches!(path.elements()[0], kurbo::PathEl::MoveTo(p) if p == Point::new(0.0,0.0))); + assert!(matches!(path.elements().last(), Some(kurbo::PathEl::CurveTo(_, _, p)) if *p == Point::new(3.0,0.0))); + + let mut negative = cubic(); + negative.degree = -1; + negative.fit = fit.fit; + assert!(spline_path(&negative).unwrap().is_some(), "invalid degree reaches the prior fit-point fallback"); + } + + #[test] + fn spline_path_rejects_malformed_knots_and_defaults_invalid_weights() { + let mut spline = cubic(); + spline.knots[4] = f64::NAN; + assert!(spline_path(&spline).unwrap().is_none()); + let mut spline = cubic(); + spline.flags |= 4; + spline.weights = vec![1.0, 0.0, 1.0, 1.0]; + assert!(spline_path(&spline).unwrap().is_some(), "invalid weights historically fall back to unit weights"); + } + + #[test] + fn ascii_document_import_keeps_spline_as_supported_geometry() { + let bytes = b"0\nSECTION\n2\nENTITIES\n0\nSPLINE\n8\n0\n70\n8\n71\n3\n72\n8\n73\n4\n40\n0\n40\n0\n40\n0\n40\n0\n40\n1\n40\n1\n40\n1\n40\n1\n10\n0\n20\n0\n10\n1\n20\n2\n10\n2\n20\n2\n10\n3\n20\n0\n0\nENDSEC\n0\nEOF\n"; + let imported = import(bytes).expect("valid spline drawing"); + assert_eq!(imported.paths.len(), 1); + assert!(imported.warnings.is_empty()); + let bounds = imported.bounds.expect("spline bounds"); + assert!((bounds.x0 - 0.0).abs() < 1e-12 && (bounds.x1 - 3.0).abs() < 1e-12); + assert!((bounds.y0 - 0.0).abs() < 1e-12 && (bounds.y1 - 1.5).abs() < 1e-12); + } + + #[test] + fn discontinuous_nurbs_spans_remain_separate_subpaths() { + let mut limit_hit = false; + let path = nurbs( + 1, + &[0.0, 0.0, 0.5, 0.5, 1.0, 1.0], + &xyz(&[Point::new(0.0, 0.0), Point::new(1.0, 0.0), Point::new(5.0, 5.0), Point::new(6.0, 5.0)]), + None, + &mut limit_hit, + ) + .expect("piecewise linear spline"); + assert!(!limit_hit); + assert!(matches!(path.elements(), [kurbo::PathEl::MoveTo(a), kurbo::PathEl::LineTo(b), kurbo::PathEl::MoveTo(c), kurbo::PathEl::LineTo(d)] + if *a == Point::new(0.0,0.0) && *b == Point::new(1.0,0.0) && *c == Point::new(5.0,5.0) && *d == Point::new(6.0,5.0))); + } + + #[test] + fn positive_knot_spans_at_large_offsets_are_kept() { + let start = 1.0e12; + let delta = 1.0e-3; + let mut limit_hit = false; + let path = nurbs(1, &[start, start, start + delta, start + delta], &xyz(&[Point::new(2.0, 3.0), Point::new(5.0, 7.0)]), None, &mut limit_hit) + .expect("positive-width span"); + assert!(!limit_hit); + assert!(matches!(path.elements(), [kurbo::PathEl::MoveTo(a), kurbo::PathEl::LineTo(b)] + if *a == Point::new(2.0, 3.0) && *b == Point::new(5.0, 7.0))); + } + + #[test] + fn fit_point_fallback_checks_element_budget_before_collecting() { + let mut spline = cubic(); + spline.degree = -1; + spline.fit = vec![[0.0, 0.0, 0.0]; MAX_SPLINE_ELEMENTS + 1]; + assert!(matches!(spline_path(&spline), Err(Error::TooManySegments))); + } + + #[test] + fn bernstein_error_bound_covers_rational_and_high_degree_deviations() { + for control in [ + vec![Homogeneous { x: 0.0, y: 0.0, w: 1.0 }, Homogeneous { x: 0.2, y: 2.0, w: 0.5 }, Homogeneous { x: 1.0, y: 0.0, w: 1.0 }], + vec![ + Homogeneous { x: 0.0, y: 0.0, w: 1.0 }, + Homogeneous { x: 0.2, y: 2.0, w: 0.7 }, + Homogeneous { x: 0.5, y: -2.0, w: 0.9 }, + Homogeneous { x: 0.8, y: 2.0, w: 1.2 }, + Homogeneous { x: 1.0, y: 0.0, w: 1.0 }, + ], + ] { + let degree = control.len() - 1; + let bezier = [Point::new(0.0, 0.0), Point::new(0.25, 1.0), Point::new(0.75, 1.0), Point::new(1.0, 0.0)]; + let bound = cubic_error_bound(&control, &bezier).unwrap(); + let cubic = kurbo::CubicBez::new(bezier[0], bezier[1], bezier[2], bezier[3]); + for sample in 0..=1024 { + let t = sample as f64 / 1024.0; + let mut points = control.clone(); + for level in (1..=degree).rev() { + for i in 0..level { + points[i] = points[i].lerp(points[i + 1], t); + } + } + let source = points[0].point().unwrap(); + let approximation = kurbo::ParamCurve::eval(&cubic, t); + assert!(source.distance(approximation) <= bound * (1.0 + 1e-12), "sample {t}: bound {bound}"); + } + } + } + + #[test] + fn rational_quadratic_circle_matches_the_prior_importer_tolerance() { + let weight = std::f64::consts::FRAC_1_SQRT_2; + let control = xyz(&[(1.0, 0.0), (1.0, 1.0), (0.0, 1.0), (-1.0, 1.0), (-1.0, 0.0), (-1.0, -1.0), (0.0, -1.0), (1.0, -1.0), (1.0, 0.0)] + .map(|(x, y)| Point::new(200.0 + x * 100.0, 200.0 + y * 100.0))); + let knots = [0.0, 0.0, 0.0, 1.0, 1.0, 2.0, 2.0, 3.0, 3.0, 4.0, 4.0, 4.0]; + let weights = [1.0, weight, 1.0, weight, 1.0, weight, 1.0, weight, 1.0]; + let mut limit_hit = false; + let path = nurbs(2, &knots, &control, Some(&weights), &mut limit_hit).expect("circle NURBS"); + assert!(!limit_hit); + let center = Point::new(200.0, 200.0); + let max_radial_error = path + .segments() + .flat_map(|segment| (0..=32).map(move |i| kurbo::ParamCurve::eval(&segment, i as f64 / 32.0))) + .map(|point| (point.distance(center) - 100.0).abs()) + .fold(0.0_f64, f64::max); + assert!(max_radial_error < 0.01, "prior importer tolerance was 0.01; got {max_radial_error}"); + } + + #[test] + fn bounded_spline_fit_matches_prior_rational_and_high_degree_geometry() { + let weight = std::f64::consts::FRAC_1_SQRT_2; + let curves = [ + vec![Homogeneous { x: 1.0, y: 0.0, w: 1.0 }, Homogeneous { x: weight, y: weight, w: weight }, Homogeneous { x: 0.0, y: 1.0, w: 1.0 }], + vec![ + Homogeneous { x: 0.0, y: 0.0, w: 1.0 }, + Homogeneous { x: 0.2, y: 2.0, w: 0.7 }, + Homogeneous { x: 0.5, y: -2.0, w: 0.9 }, + Homogeneous { x: 0.8, y: 2.0, w: 1.2 }, + Homogeneous { x: 1.0, y: 0.0, w: 1.0 }, + ], + ]; + for bezier in curves { + let points: Vec<_> = bezier.iter().map(|point| point.point().expect("finite source point")).collect(); + let (lo, hi) = points.iter().fold((points[0], points[0]), |(lo, hi), point| { + (Point::new(lo.x.min(point.x), lo.y.min(point.y)), Point::new(hi.x.max(point.x), hi.y.max(point.y))) + }); + let tolerance = ((hi - lo).hypot() * SPLINE_TOLERANCE).max(1e-12); + let mut current = BezPath::new(); + current.move_to(points[0]); + let mut limit_hit = false; + fit_spline(&mut current, &bezier, tolerance, 0, &mut limit_hit).expect("bounded fit succeeds"); + assert!(!limit_hit); + let prior = legacy_path(&bezier, tolerance); + let difference = sampled_path_distance(&prior, ¤t, tolerance).max(sampled_path_distance(¤t, &prior, tolerance)); + assert!(difference <= 2.1 * tolerance, "legacy/new Hausdorff sample distance {difference} exceeds 2.1 Ă— tolerance {tolerance}"); + } + } +} diff --git a/crates/dxf/tests/data/drawing_empty_r12.dxf b/crates/dxf/tests/data/drawing_empty_r12.dxf new file mode 100644 index 000000000..402512d9e --- /dev/null +++ b/crates/dxf/tests/data/drawing_empty_r12.dxf @@ -0,0 +1,250 @@ + 0 +SECTION + 2 +HEADER + 9 +$ACADVER + 1 +AC1009 + 9 +$DWGCODEPAGE + 3 +ANSI_1252 + 9 +$INSBASE + 10 +0 + 20 +0 + 30 +0 + 9 +$EXTMIN + 10 +0 + 20 +0 + 30 +0 + 9 +$EXTMAX + 10 +200 + 20 +100 + 30 +0 + 9 +$LIMMIN + 10 +0 + 20 +0 + 9 +$LIMMAX + 10 +200 + 20 +100 + 0 +ENDSEC + 0 +SECTION + 2 +TABLES + 0 +TABLE + 2 +VPORT + 70 +1 + 0 +VPORT + 2 +*ACTIVE + 70 +0 + 10 +0 + 20 +0 + 11 +1 + 21 +1 + 12 +100 + 22 +50 + 13 +0 + 23 +0 + 14 +1 + 24 +1 + 15 +10 + 25 +10 + 16 +0 + 26 +0 + 36 +1 + 17 +0 + 27 +0 + 37 +0 + 40 +146.666667 + 41 +1.5 + 42 +50 + 43 +0 + 44 +0 + 50 +0 + 51 +0 + 71 +0 + 72 +1000 + 73 +1 + 74 +3 + 75 +0 + 76 +0 + 77 +0 + 78 +0 + 0 +ENDTAB + 0 +TABLE + 2 +LTYPE + 70 +3 + 0 +LTYPE + 2 +ByBlock + 70 +0 + 3 + + 72 +65 + 73 +0 + 40 +0 + 0 +LTYPE + 2 +ByLayer + 70 +0 + 3 + + 72 +65 + 73 +0 + 40 +0 + 0 +LTYPE + 2 +Continuous + 70 +0 + 3 +Solid line + 72 +65 + 73 +0 + 40 +0 + 0 +ENDTAB + 0 +TABLE + 2 +LAYER + 70 +2 + 0 +LAYER + 2 +0 + 70 +0 + 62 +7 + 6 +Continuous + 0 +LAYER + 2 +Layer1 + 70 +0 + 62 +1 + 6 +Continuous + 0 +ENDTAB + 0 +TABLE + 2 +STYLE + 70 +1 + 0 +STYLE + 2 +Standard + 70 +0 + 40 +0 + 41 +1 + 50 +0 + 71 +0 + 42 +2.5 + 3 +txt + 4 + + 0 +ENDTAB + 0 +ENDSEC + 0 +SECTION + 2 +ENTITIES + 0 +ENDSEC + 0 +EOF diff --git a/crates/dxf/tests/data/drawing_empty_r2018.dxf b/crates/dxf/tests/data/drawing_empty_r2018.dxf new file mode 100644 index 000000000..01e9d07b3 --- /dev/null +++ b/crates/dxf/tests/data/drawing_empty_r2018.dxf @@ -0,0 +1,972 @@ + 0 +SECTION + 2 +HEADER + 9 +$ACADVER + 1 +AC1032 + 9 +$DWGCODEPAGE + 3 +ANSI_1252 + 9 +$INSBASE + 10 +0 + 20 +0 + 30 +0 + 9 +$EXTMIN + 10 +0 + 20 +0 + 30 +0 + 9 +$EXTMAX + 10 +200 + 20 +100 + 30 +0 + 9 +$LIMMIN + 10 +0 + 20 +0 + 9 +$LIMMAX + 10 +200 + 20 +100 + 9 +$LWDISPLAY +290 +1 + 9 +$INSUNITS + 70 +4 + 9 +$MEASUREMENT + 70 +1 + 9 +$HANDSEED + 5 +36 + 0 +ENDSEC + 0 +SECTION + 2 +CLASSES + 0 +CLASS + 1 +ACDBDICTIONARYWDFLT + 2 +AcDbDictionaryWithDefault + 3 +ObjectDBX Classes + 90 +0 + 91 +0 +280 +0 +281 +0 + 0 +CLASS + 1 +ACDBPLACEHOLDER + 2 +AcDbPlaceHolder + 3 +ObjectDBX Classes + 90 +0 + 91 +0 +280 +0 +281 +0 + 0 +CLASS + 1 +LAYOUT + 2 +AcDbLayout + 3 +ObjectDBX Classes + 90 +0 + 91 +0 +280 +0 +281 +0 + 0 +ENDSEC + 0 +SECTION + 2 +TABLES + 0 +TABLE + 2 +VPORT + 5 +8 +330 +0 +100 +AcDbSymbolTable + 70 +1 + 0 +VPORT + 5 +15 +330 +8 +100 +AcDbSymbolTableRecord +100 +AcDbViewportTableRecord + 2 +*Active + 70 +0 + 10 +0 + 20 +0 + 11 +1 + 21 +1 + 12 +100 + 22 +50 + 13 +0 + 23 +0 + 14 +1 + 24 +1 + 15 +10 + 25 +10 + 16 +0 + 26 +0 + 36 +1 + 17 +0 + 27 +0 + 37 +0 + 40 +146.666667 + 41 +1.5 + 42 +50 + 43 +0 + 44 +0 + 50 +0 + 51 +0 + 71 +0 + 72 +1000 + 73 +1 + 74 +3 + 75 +0 + 76 +0 + 77 +0 + 78 +0 +281 +0 + 65 +0 +146 +0 + 0 +ENDTAB + 0 +TABLE + 2 +LTYPE + 5 +2 +330 +0 +100 +AcDbSymbolTable + 70 +3 + 0 +LTYPE + 5 +30 +330 +2 +100 +AcDbSymbolTableRecord +100 +AcDbLinetypeTableRecord + 2 +ByBlock + 70 +0 + 3 + + 72 +65 + 73 +0 + 40 +0 + 0 +LTYPE + 5 +31 +330 +2 +100 +AcDbSymbolTableRecord +100 +AcDbLinetypeTableRecord + 2 +ByLayer + 70 +0 + 3 + + 72 +65 + 73 +0 + 40 +0 + 0 +LTYPE + 5 +32 +330 +2 +100 +AcDbSymbolTableRecord +100 +AcDbLinetypeTableRecord + 2 +Continuous + 70 +0 + 3 +Solid line + 72 +65 + 73 +0 + 40 +0 + 0 +ENDTAB + 0 +TABLE + 2 +LAYER + 5 +1 +330 +0 +100 +AcDbSymbolTable + 70 +2 + 0 +LAYER + 5 +33 +330 +1 +100 +AcDbSymbolTableRecord +100 +AcDbLayerTableRecord + 2 +0 + 70 +0 + 62 +7 + 6 +Continuous +290 +1 +370 +-3 +390 +1B + 0 +LAYER + 5 +34 +330 +1 +100 +AcDbSymbolTableRecord +100 +AcDbLayerTableRecord + 2 +Layer1 + 70 +0 + 62 +1 + 6 +Continuous +290 +1 +370 +-3 +390 +1B + 0 +ENDTAB + 0 +TABLE + 2 +STYLE + 5 +5 +330 +0 +100 +AcDbSymbolTable + 70 +1 + 0 +STYLE + 5 +35 +330 +5 +100 +AcDbSymbolTableRecord +100 +AcDbTextStyleTableRecord + 2 +Standard + 70 +0 + 40 +0 + 41 +1 + 50 +0 + 71 +0 + 42 +2.5 + 3 +txt + 4 + + 0 +ENDTAB + 0 +TABLE + 2 +VIEW + 5 +7 +330 +0 +100 +AcDbSymbolTable + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +UCS + 5 +6 +330 +0 +100 +AcDbSymbolTable + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +APPID + 5 +3 +330 +0 +100 +AcDbSymbolTable + 70 +1 + 0 +APPID + 5 +16 +330 +3 +100 +AcDbSymbolTableRecord +100 +AcDbRegAppTableRecord + 2 +ACAD + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +DIMSTYLE + 5 +4 +330 +0 +100 +AcDbSymbolTable + 70 +1 +100 +AcDbDimStyleTable + 0 +DIMSTYLE +105 +17 +330 +4 +100 +AcDbSymbolTableRecord +100 +AcDbDimStyleTableRecord + 2 +Standard + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +BLOCK_RECORD + 5 +9 +330 +0 +100 +AcDbSymbolTable + 70 +2 + 0 +BLOCK_RECORD + 5 +D +330 +9 +100 +AcDbSymbolTableRecord +100 +AcDbBlockTableRecord + 2 +*Model_Space +340 +13 + 70 +0 +280 +1 +281 +0 + 0 +BLOCK_RECORD + 5 +E +330 +9 +100 +AcDbSymbolTableRecord +100 +AcDbBlockTableRecord + 2 +*Paper_Space +340 +14 + 70 +0 +280 +1 +281 +0 + 0 +ENDTAB + 0 +ENDSEC + 0 +SECTION + 2 +BLOCKS + 0 +BLOCK + 5 +F +330 +D +100 +AcDbEntity + 8 +0 +100 +AcDbBlockBegin + 2 +*Model_Space + 70 +0 + 10 +0 + 20 +0 + 30 +0 + 3 +*Model_Space + 1 + + 0 +ENDBLK + 5 +10 +330 +D +100 +AcDbEntity + 8 +0 +100 +AcDbBlockEnd + 0 +BLOCK + 5 +11 +330 +E +100 +AcDbEntity + 8 +0 +100 +AcDbBlockBegin + 2 +*Paper_Space + 70 +0 + 10 +0 + 20 +0 + 30 +0 + 3 +*Paper_Space + 1 + + 0 +ENDBLK + 5 +12 +330 +E +100 +AcDbEntity + 8 +0 +100 +AcDbBlockEnd + 0 +ENDSEC + 0 +SECTION + 2 +ENTITIES + 0 +ENDSEC + 0 +SECTION + 2 +OBJECTS + 0 +DICTIONARY + 5 +A +330 +0 +100 +AcDbDictionary +281 +1 + 3 +ACAD_GROUP +350 +B + 3 +ACAD_LAYOUT +350 +C + 3 +ACAD_PLOTSTYLENAME +350 +1A + 0 +DICTIONARY + 5 +B +330 +A +100 +AcDbDictionary +281 +1 + 0 +DICTIONARY + 5 +C +330 +A +100 +AcDbDictionary +281 +1 + 3 +Layout1 +350 +14 + 3 +Model +350 +13 + 0 +ACDBDICTIONARYWDFLT + 5 +1A +330 +A +100 +AcDbDictionary +281 +1 + 3 +Normal +350 +1B +100 +AcDbDictionaryWithDefault +340 +1B + 0 +ACDBPLACEHOLDER + 5 +1B +330 +1A + 0 +LAYOUT + 5 +13 +330 +C +100 +AcDbPlotSettings + 1 + + 2 +none_device + 4 + + 6 + + 40 +0 + 41 +0 + 42 +0 + 43 +0 + 44 +0 + 45 +0 + 46 +0 + 47 +0 + 48 +0 + 49 +0 +140 +0 +141 +0 +142 +1 +143 +1 + 70 +1712 + 72 +1 + 73 +0 + 74 +5 + 7 + + 75 +16 + 76 +0 + 77 +2 + 78 +300 +147 +1 +148 +0 +149 +0 +100 +AcDbLayout + 1 +Model + 70 +1 + 71 +0 + 10 +0 + 20 +0 + 11 +200 + 21 +100 + 12 +0 + 22 +0 + 32 +0 + 14 +0 + 24 +0 + 34 +0 + 15 +200 + 25 +100 + 35 +0 +146 +0 + 13 +0 + 23 +0 + 33 +0 + 16 +1 + 26 +0 + 36 +0 + 17 +0 + 27 +1 + 37 +0 + 76 +1 +330 +D + 0 +LAYOUT + 5 +14 +330 +C +100 +AcDbPlotSettings + 1 + + 2 +none_device + 4 + + 6 + + 40 +0 + 41 +0 + 42 +0 + 43 +0 + 44 +0 + 45 +0 + 46 +0 + 47 +0 + 48 +0 + 49 +0 +140 +0 +141 +0 +142 +1 +143 +1 + 70 +688 + 72 +1 + 73 +0 + 74 +5 + 7 + + 75 +16 + 76 +0 + 77 +2 + 78 +300 +147 +1 +148 +0 +149 +0 +100 +AcDbLayout + 1 +Layout1 + 70 +1 + 71 +1 + 10 +0 + 20 +0 + 11 +200 + 21 +100 + 12 +0 + 22 +0 + 32 +0 + 14 +0 + 24 +0 + 34 +0 + 15 +200 + 25 +100 + 35 +0 +146 +0 + 13 +0 + 23 +0 + 33 +0 + 16 +1 + 26 +0 + 36 +0 + 17 +0 + 27 +1 + 37 +0 + 76 +1 +330 +E + 0 +ENDSEC + 0 +EOF diff --git a/crates/dxf/tests/data/drawing_line_r12.dxf b/crates/dxf/tests/data/drawing_line_r12.dxf new file mode 100644 index 000000000..6810c07cb --- /dev/null +++ b/crates/dxf/tests/data/drawing_line_r12.dxf @@ -0,0 +1,268 @@ + 0 +SECTION + 2 +HEADER + 9 +$ACADVER + 1 +AC1009 + 9 +$DWGCODEPAGE + 3 +ANSI_1252 + 9 +$INSBASE + 10 +0 + 20 +0 + 30 +0 + 9 +$EXTMIN + 10 +0 + 20 +0 + 30 +0 + 9 +$EXTMAX + 10 +200 + 20 +100 + 30 +0 + 9 +$LIMMIN + 10 +0 + 20 +0 + 9 +$LIMMAX + 10 +200 + 20 +100 + 0 +ENDSEC + 0 +SECTION + 2 +TABLES + 0 +TABLE + 2 +VPORT + 70 +1 + 0 +VPORT + 2 +*ACTIVE + 70 +0 + 10 +0 + 20 +0 + 11 +1 + 21 +1 + 12 +100 + 22 +50 + 13 +0 + 23 +0 + 14 +1 + 24 +1 + 15 +10 + 25 +10 + 16 +0 + 26 +0 + 36 +1 + 17 +0 + 27 +0 + 37 +0 + 40 +146.666667 + 41 +1.5 + 42 +50 + 43 +0 + 44 +0 + 50 +0 + 51 +0 + 71 +0 + 72 +1000 + 73 +1 + 74 +3 + 75 +0 + 76 +0 + 77 +0 + 78 +0 + 0 +ENDTAB + 0 +TABLE + 2 +LTYPE + 70 +3 + 0 +LTYPE + 2 +ByBlock + 70 +0 + 3 + + 72 +65 + 73 +0 + 40 +0 + 0 +LTYPE + 2 +ByLayer + 70 +0 + 3 + + 72 +65 + 73 +0 + 40 +0 + 0 +LTYPE + 2 +Continuous + 70 +0 + 3 +Solid line + 72 +65 + 73 +0 + 40 +0 + 0 +ENDTAB + 0 +TABLE + 2 +LAYER + 70 +2 + 0 +LAYER + 2 +0 + 70 +0 + 62 +7 + 6 +Continuous + 0 +LAYER + 2 +Layer1 + 70 +0 + 62 +1 + 6 +Continuous + 0 +ENDTAB + 0 +TABLE + 2 +STYLE + 70 +1 + 0 +STYLE + 2 +Standard + 70 +0 + 40 +0 + 41 +1 + 50 +0 + 71 +0 + 42 +2.5 + 3 +txt + 4 + + 0 +ENDTAB + 0 +ENDSEC + 0 +SECTION + 2 +ENTITIES + 0 +LINE + 8 +Layer1 + 62 +1 + 10 +0 + 20 +0 + 30 +0 + 11 +10 + 21 +20 + 31 +0 + 0 +ENDSEC + 0 +EOF diff --git a/crates/dxf/tests/data/drawing_line_r2000.dxf b/crates/dxf/tests/data/drawing_line_r2000.dxf new file mode 100644 index 000000000..173a9b427 --- /dev/null +++ b/crates/dxf/tests/data/drawing_line_r2000.dxf @@ -0,0 +1,982 @@ + 0 +SECTION + 2 +HEADER + 9 +$ACADVER + 1 +AC1015 + 9 +$DWGCODEPAGE + 3 +ANSI_1252 + 9 +$INSBASE + 10 +0 + 20 +0 + 30 +0 + 9 +$EXTMIN + 10 +0 + 20 +0 + 30 +0 + 9 +$EXTMAX + 10 +200 + 20 +100 + 30 +0 + 9 +$LIMMIN + 10 +0 + 20 +0 + 9 +$LIMMAX + 10 +200 + 20 +100 + 9 +$LWDISPLAY +290 +1 + 9 +$INSUNITS + 70 +4 + 9 +$MEASUREMENT + 70 +1 + 9 +$HANDSEED + 5 +37 + 0 +ENDSEC + 0 +SECTION + 2 +CLASSES + 0 +CLASS + 1 +ACDBDICTIONARYWDFLT + 2 +AcDbDictionaryWithDefault + 3 +ObjectDBX Classes + 90 +0 +280 +0 +281 +0 + 0 +CLASS + 1 +ACDBPLACEHOLDER + 2 +AcDbPlaceHolder + 3 +ObjectDBX Classes + 90 +0 +280 +0 +281 +0 + 0 +CLASS + 1 +LAYOUT + 2 +AcDbLayout + 3 +ObjectDBX Classes + 90 +0 +280 +0 +281 +0 + 0 +ENDSEC + 0 +SECTION + 2 +TABLES + 0 +TABLE + 2 +VPORT + 5 +8 +330 +0 +100 +AcDbSymbolTable + 70 +1 + 0 +VPORT + 5 +15 +330 +8 +100 +AcDbSymbolTableRecord +100 +AcDbViewportTableRecord + 2 +*Active + 70 +0 + 10 +0 + 20 +0 + 11 +1 + 21 +1 + 12 +100 + 22 +50 + 13 +0 + 23 +0 + 14 +1 + 24 +1 + 15 +10 + 25 +10 + 16 +0 + 26 +0 + 36 +1 + 17 +0 + 27 +0 + 37 +0 + 40 +146.666667 + 41 +1.5 + 42 +50 + 43 +0 + 44 +0 + 50 +0 + 51 +0 + 71 +0 + 72 +1000 + 73 +1 + 74 +3 + 75 +0 + 76 +0 + 77 +0 + 78 +0 +281 +0 + 65 +0 +146 +0 + 0 +ENDTAB + 0 +TABLE + 2 +LTYPE + 5 +2 +330 +0 +100 +AcDbSymbolTable + 70 +3 + 0 +LTYPE + 5 +31 +330 +2 +100 +AcDbSymbolTableRecord +100 +AcDbLinetypeTableRecord + 2 +ByBlock + 70 +0 + 3 + + 72 +65 + 73 +0 + 40 +0 + 0 +LTYPE + 5 +32 +330 +2 +100 +AcDbSymbolTableRecord +100 +AcDbLinetypeTableRecord + 2 +ByLayer + 70 +0 + 3 + + 72 +65 + 73 +0 + 40 +0 + 0 +LTYPE + 5 +33 +330 +2 +100 +AcDbSymbolTableRecord +100 +AcDbLinetypeTableRecord + 2 +Continuous + 70 +0 + 3 +Solid line + 72 +65 + 73 +0 + 40 +0 + 0 +ENDTAB + 0 +TABLE + 2 +LAYER + 5 +1 +330 +0 +100 +AcDbSymbolTable + 70 +2 + 0 +LAYER + 5 +34 +330 +1 +100 +AcDbSymbolTableRecord +100 +AcDbLayerTableRecord + 2 +0 + 70 +0 + 62 +7 + 6 +Continuous +290 +1 +370 +-3 +390 +1B + 0 +LAYER + 5 +35 +330 +1 +100 +AcDbSymbolTableRecord +100 +AcDbLayerTableRecord + 2 +Layer1 + 70 +0 + 62 +1 + 6 +Continuous +290 +1 +370 +-3 +390 +1B + 0 +ENDTAB + 0 +TABLE + 2 +STYLE + 5 +5 +330 +0 +100 +AcDbSymbolTable + 70 +1 + 0 +STYLE + 5 +36 +330 +5 +100 +AcDbSymbolTableRecord +100 +AcDbTextStyleTableRecord + 2 +Standard + 70 +0 + 40 +0 + 41 +1 + 50 +0 + 71 +0 + 42 +2.5 + 3 +txt + 4 + + 0 +ENDTAB + 0 +TABLE + 2 +VIEW + 5 +7 +330 +0 +100 +AcDbSymbolTable + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +UCS + 5 +6 +330 +0 +100 +AcDbSymbolTable + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +APPID + 5 +3 +330 +0 +100 +AcDbSymbolTable + 70 +1 + 0 +APPID + 5 +16 +330 +3 +100 +AcDbSymbolTableRecord +100 +AcDbRegAppTableRecord + 2 +ACAD + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +DIMSTYLE + 5 +4 +330 +0 +100 +AcDbSymbolTable + 70 +1 +100 +AcDbDimStyleTable + 0 +DIMSTYLE +105 +17 +330 +4 +100 +AcDbSymbolTableRecord +100 +AcDbDimStyleTableRecord + 2 +Standard + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +BLOCK_RECORD + 5 +9 +330 +0 +100 +AcDbSymbolTable + 70 +2 + 0 +BLOCK_RECORD + 5 +D +330 +9 +100 +AcDbSymbolTableRecord +100 +AcDbBlockTableRecord + 2 +*Model_Space +340 +13 + 70 +0 +280 +1 +281 +0 + 0 +BLOCK_RECORD + 5 +E +330 +9 +100 +AcDbSymbolTableRecord +100 +AcDbBlockTableRecord + 2 +*Paper_Space +340 +14 + 70 +0 +280 +1 +281 +0 + 0 +ENDTAB + 0 +ENDSEC + 0 +SECTION + 2 +BLOCKS + 0 +BLOCK + 5 +F +330 +D +100 +AcDbEntity + 8 +0 +100 +AcDbBlockBegin + 2 +*Model_Space + 70 +0 + 10 +0 + 20 +0 + 30 +0 + 3 +*Model_Space + 1 + + 0 +ENDBLK + 5 +10 +330 +D +100 +AcDbEntity + 8 +0 +100 +AcDbBlockEnd + 0 +BLOCK + 5 +11 +330 +E +100 +AcDbEntity + 8 +0 +100 +AcDbBlockBegin + 2 +*Paper_Space + 70 +0 + 10 +0 + 20 +0 + 30 +0 + 3 +*Paper_Space + 1 + + 0 +ENDBLK + 5 +12 +330 +E +100 +AcDbEntity + 8 +0 +100 +AcDbBlockEnd + 0 +ENDSEC + 0 +SECTION + 2 +ENTITIES + 0 +LINE + 5 +30 +330 +D +100 +AcDbEntity + 8 +Layer1 + 62 +1 +370 +25 +100 +AcDbLine + 10 +0 + 20 +0 + 30 +0 + 11 +10 + 21 +20 + 31 +0 + 0 +ENDSEC + 0 +SECTION + 2 +OBJECTS + 0 +DICTIONARY + 5 +A +330 +0 +100 +AcDbDictionary +281 +1 + 3 +ACAD_GROUP +350 +B + 3 +ACAD_LAYOUT +350 +C + 3 +ACAD_PLOTSTYLENAME +350 +1A + 0 +DICTIONARY + 5 +B +330 +A +100 +AcDbDictionary +281 +1 + 0 +DICTIONARY + 5 +C +330 +A +100 +AcDbDictionary +281 +1 + 3 +Layout1 +350 +14 + 3 +Model +350 +13 + 0 +ACDBDICTIONARYWDFLT + 5 +1A +330 +A +100 +AcDbDictionary +281 +1 + 3 +Normal +350 +1B +100 +AcDbDictionaryWithDefault +340 +1B + 0 +ACDBPLACEHOLDER + 5 +1B +330 +1A + 0 +LAYOUT + 5 +13 +330 +C +100 +AcDbPlotSettings + 1 + + 2 +none_device + 4 + + 6 + + 40 +0 + 41 +0 + 42 +0 + 43 +0 + 44 +0 + 45 +0 + 46 +0 + 47 +0 + 48 +0 + 49 +0 +140 +0 +141 +0 +142 +1 +143 +1 + 70 +1712 + 72 +1 + 73 +0 + 74 +5 + 7 + + 75 +16 +147 +1 +148 +0 +149 +0 +100 +AcDbLayout + 1 +Model + 70 +1 + 71 +0 + 10 +0 + 20 +0 + 11 +200 + 21 +100 + 12 +0 + 22 +0 + 32 +0 + 14 +0 + 24 +0 + 34 +0 + 15 +200 + 25 +100 + 35 +0 +146 +0 + 13 +0 + 23 +0 + 33 +0 + 16 +1 + 26 +0 + 36 +0 + 17 +0 + 27 +1 + 37 +0 + 76 +1 +330 +D + 0 +LAYOUT + 5 +14 +330 +C +100 +AcDbPlotSettings + 1 + + 2 +none_device + 4 + + 6 + + 40 +0 + 41 +0 + 42 +0 + 43 +0 + 44 +0 + 45 +0 + 46 +0 + 47 +0 + 48 +0 + 49 +0 +140 +0 +141 +0 +142 +1 +143 +1 + 70 +688 + 72 +1 + 73 +0 + 74 +5 + 7 + + 75 +16 +147 +1 +148 +0 +149 +0 +100 +AcDbLayout + 1 +Layout1 + 70 +1 + 71 +1 + 10 +0 + 20 +0 + 11 +200 + 21 +100 + 12 +0 + 22 +0 + 32 +0 + 14 +0 + 24 +0 + 34 +0 + 15 +200 + 25 +100 + 35 +0 +146 +0 + 13 +0 + 23 +0 + 33 +0 + 16 +1 + 26 +0 + 36 +0 + 17 +0 + 27 +1 + 37 +0 + 76 +1 +330 +E + 0 +ENDSEC + 0 +EOF diff --git a/crates/dxf/tests/data/drawing_line_r2018.dxf b/crates/dxf/tests/data/drawing_line_r2018.dxf new file mode 100644 index 000000000..8318c1c5e --- /dev/null +++ b/crates/dxf/tests/data/drawing_line_r2018.dxf @@ -0,0 +1,1002 @@ + 0 +SECTION + 2 +HEADER + 9 +$ACADVER + 1 +AC1032 + 9 +$DWGCODEPAGE + 3 +ANSI_1252 + 9 +$INSBASE + 10 +0 + 20 +0 + 30 +0 + 9 +$EXTMIN + 10 +0 + 20 +0 + 30 +0 + 9 +$EXTMAX + 10 +200 + 20 +100 + 30 +0 + 9 +$LIMMIN + 10 +0 + 20 +0 + 9 +$LIMMAX + 10 +200 + 20 +100 + 9 +$LWDISPLAY +290 +1 + 9 +$INSUNITS + 70 +4 + 9 +$MEASUREMENT + 70 +1 + 9 +$HANDSEED + 5 +37 + 0 +ENDSEC + 0 +SECTION + 2 +CLASSES + 0 +CLASS + 1 +ACDBDICTIONARYWDFLT + 2 +AcDbDictionaryWithDefault + 3 +ObjectDBX Classes + 90 +0 + 91 +0 +280 +0 +281 +0 + 0 +CLASS + 1 +ACDBPLACEHOLDER + 2 +AcDbPlaceHolder + 3 +ObjectDBX Classes + 90 +0 + 91 +0 +280 +0 +281 +0 + 0 +CLASS + 1 +LAYOUT + 2 +AcDbLayout + 3 +ObjectDBX Classes + 90 +0 + 91 +0 +280 +0 +281 +0 + 0 +ENDSEC + 0 +SECTION + 2 +TABLES + 0 +TABLE + 2 +VPORT + 5 +8 +330 +0 +100 +AcDbSymbolTable + 70 +1 + 0 +VPORT + 5 +15 +330 +8 +100 +AcDbSymbolTableRecord +100 +AcDbViewportTableRecord + 2 +*Active + 70 +0 + 10 +0 + 20 +0 + 11 +1 + 21 +1 + 12 +100 + 22 +50 + 13 +0 + 23 +0 + 14 +1 + 24 +1 + 15 +10 + 25 +10 + 16 +0 + 26 +0 + 36 +1 + 17 +0 + 27 +0 + 37 +0 + 40 +146.666667 + 41 +1.5 + 42 +50 + 43 +0 + 44 +0 + 50 +0 + 51 +0 + 71 +0 + 72 +1000 + 73 +1 + 74 +3 + 75 +0 + 76 +0 + 77 +0 + 78 +0 +281 +0 + 65 +0 +146 +0 + 0 +ENDTAB + 0 +TABLE + 2 +LTYPE + 5 +2 +330 +0 +100 +AcDbSymbolTable + 70 +3 + 0 +LTYPE + 5 +31 +330 +2 +100 +AcDbSymbolTableRecord +100 +AcDbLinetypeTableRecord + 2 +ByBlock + 70 +0 + 3 + + 72 +65 + 73 +0 + 40 +0 + 0 +LTYPE + 5 +32 +330 +2 +100 +AcDbSymbolTableRecord +100 +AcDbLinetypeTableRecord + 2 +ByLayer + 70 +0 + 3 + + 72 +65 + 73 +0 + 40 +0 + 0 +LTYPE + 5 +33 +330 +2 +100 +AcDbSymbolTableRecord +100 +AcDbLinetypeTableRecord + 2 +Continuous + 70 +0 + 3 +Solid line + 72 +65 + 73 +0 + 40 +0 + 0 +ENDTAB + 0 +TABLE + 2 +LAYER + 5 +1 +330 +0 +100 +AcDbSymbolTable + 70 +2 + 0 +LAYER + 5 +34 +330 +1 +100 +AcDbSymbolTableRecord +100 +AcDbLayerTableRecord + 2 +0 + 70 +0 + 62 +7 + 6 +Continuous +290 +1 +370 +-3 +390 +1B + 0 +LAYER + 5 +35 +330 +1 +100 +AcDbSymbolTableRecord +100 +AcDbLayerTableRecord + 2 +Layer1 + 70 +0 + 62 +1 + 6 +Continuous +290 +1 +370 +-3 +390 +1B + 0 +ENDTAB + 0 +TABLE + 2 +STYLE + 5 +5 +330 +0 +100 +AcDbSymbolTable + 70 +1 + 0 +STYLE + 5 +36 +330 +5 +100 +AcDbSymbolTableRecord +100 +AcDbTextStyleTableRecord + 2 +Standard + 70 +0 + 40 +0 + 41 +1 + 50 +0 + 71 +0 + 42 +2.5 + 3 +txt + 4 + + 0 +ENDTAB + 0 +TABLE + 2 +VIEW + 5 +7 +330 +0 +100 +AcDbSymbolTable + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +UCS + 5 +6 +330 +0 +100 +AcDbSymbolTable + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +APPID + 5 +3 +330 +0 +100 +AcDbSymbolTable + 70 +1 + 0 +APPID + 5 +16 +330 +3 +100 +AcDbSymbolTableRecord +100 +AcDbRegAppTableRecord + 2 +ACAD + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +DIMSTYLE + 5 +4 +330 +0 +100 +AcDbSymbolTable + 70 +1 +100 +AcDbDimStyleTable + 0 +DIMSTYLE +105 +17 +330 +4 +100 +AcDbSymbolTableRecord +100 +AcDbDimStyleTableRecord + 2 +Standard + 70 +0 + 0 +ENDTAB + 0 +TABLE + 2 +BLOCK_RECORD + 5 +9 +330 +0 +100 +AcDbSymbolTable + 70 +2 + 0 +BLOCK_RECORD + 5 +D +330 +9 +100 +AcDbSymbolTableRecord +100 +AcDbBlockTableRecord + 2 +*Model_Space +340 +13 + 70 +0 +280 +1 +281 +0 + 0 +BLOCK_RECORD + 5 +E +330 +9 +100 +AcDbSymbolTableRecord +100 +AcDbBlockTableRecord + 2 +*Paper_Space +340 +14 + 70 +0 +280 +1 +281 +0 + 0 +ENDTAB + 0 +ENDSEC + 0 +SECTION + 2 +BLOCKS + 0 +BLOCK + 5 +F +330 +D +100 +AcDbEntity + 8 +0 +100 +AcDbBlockBegin + 2 +*Model_Space + 70 +0 + 10 +0 + 20 +0 + 30 +0 + 3 +*Model_Space + 1 + + 0 +ENDBLK + 5 +10 +330 +D +100 +AcDbEntity + 8 +0 +100 +AcDbBlockEnd + 0 +BLOCK + 5 +11 +330 +E +100 +AcDbEntity + 8 +0 +100 +AcDbBlockBegin + 2 +*Paper_Space + 70 +0 + 10 +0 + 20 +0 + 30 +0 + 3 +*Paper_Space + 1 + + 0 +ENDBLK + 5 +12 +330 +E +100 +AcDbEntity + 8 +0 +100 +AcDbBlockEnd + 0 +ENDSEC + 0 +SECTION + 2 +ENTITIES + 0 +LINE + 5 +30 +330 +D +100 +AcDbEntity + 8 +Layer1 + 62 +1 +370 +25 +420 +16711680 +100 +AcDbLine + 10 +0 + 20 +0 + 30 +0 + 11 +10 + 21 +20 + 31 +0 + 0 +ENDSEC + 0 +SECTION + 2 +OBJECTS + 0 +DICTIONARY + 5 +A +330 +0 +100 +AcDbDictionary +281 +1 + 3 +ACAD_GROUP +350 +B + 3 +ACAD_LAYOUT +350 +C + 3 +ACAD_PLOTSTYLENAME +350 +1A + 0 +DICTIONARY + 5 +B +330 +A +100 +AcDbDictionary +281 +1 + 0 +DICTIONARY + 5 +C +330 +A +100 +AcDbDictionary +281 +1 + 3 +Layout1 +350 +14 + 3 +Model +350 +13 + 0 +ACDBDICTIONARYWDFLT + 5 +1A +330 +A +100 +AcDbDictionary +281 +1 + 3 +Normal +350 +1B +100 +AcDbDictionaryWithDefault +340 +1B + 0 +ACDBPLACEHOLDER + 5 +1B +330 +1A + 0 +LAYOUT + 5 +13 +330 +C +100 +AcDbPlotSettings + 1 + + 2 +none_device + 4 + + 6 + + 40 +0 + 41 +0 + 42 +0 + 43 +0 + 44 +0 + 45 +0 + 46 +0 + 47 +0 + 48 +0 + 49 +0 +140 +0 +141 +0 +142 +1 +143 +1 + 70 +1712 + 72 +1 + 73 +0 + 74 +5 + 7 + + 75 +16 + 76 +0 + 77 +2 + 78 +300 +147 +1 +148 +0 +149 +0 +100 +AcDbLayout + 1 +Model + 70 +1 + 71 +0 + 10 +0 + 20 +0 + 11 +200 + 21 +100 + 12 +0 + 22 +0 + 32 +0 + 14 +0 + 24 +0 + 34 +0 + 15 +200 + 25 +100 + 35 +0 +146 +0 + 13 +0 + 23 +0 + 33 +0 + 16 +1 + 26 +0 + 36 +0 + 17 +0 + 27 +1 + 37 +0 + 76 +1 +330 +D + 0 +LAYOUT + 5 +14 +330 +C +100 +AcDbPlotSettings + 1 + + 2 +none_device + 4 + + 6 + + 40 +0 + 41 +0 + 42 +0 + 43 +0 + 44 +0 + 45 +0 + 46 +0 + 47 +0 + 48 +0 + 49 +0 +140 +0 +141 +0 +142 +1 +143 +1 + 70 +688 + 72 +1 + 73 +0 + 74 +5 + 7 + + 75 +16 + 76 +0 + 77 +2 + 78 +300 +147 +1 +148 +0 +149 +0 +100 +AcDbLayout + 1 +Layout1 + 70 +1 + 71 +1 + 10 +0 + 20 +0 + 11 +200 + 21 +100 + 12 +0 + 22 +0 + 32 +0 + 14 +0 + 24 +0 + 34 +0 + 15 +200 + 25 +100 + 35 +0 +146 +0 + 13 +0 + 23 +0 + 33 +0 + 16 +1 + 26 +0 + 36 +0 + 17 +0 + 27 +1 + 37 +0 + 76 +1 +330 +E + 0 +ENDSEC + 0 +EOF diff --git a/crates/dxf/tests/model_write.rs b/crates/dxf/tests/model_write.rs index 1f1df5fcb..8ece9c5bb 100644 --- a/crates/dxf/tests/model_write.rs +++ b/crates/dxf/tests/model_write.rs @@ -596,7 +596,8 @@ fn the_file_around_the_content() { raw_objects: vec![Tag::s(0, "DICTIONARY"), Tag::s(5, "901"), Tag::s(330, "A")], raw_classes: vec![Tag::s(0, "CLASS"), Tag::s(1, "IMAGEDEF")], ..Drawing::default() - }; + } + .with_linear_units(2); let (tags, skipped) = drawing.tags(); assert_eq!(skipped, 1, "the entity with no groups to write"); assert_eq!(tags.last().map(|t| (t.code, t.str())), Some((0, "EOF".to_string()))); @@ -617,6 +618,7 @@ fn the_file_around_the_content() { assert_eq!((d.header.acadver().as_deref(), d.header.ltscale(), d.header.insunits()), (Some("AC1032"), 2.0, 6)); assert_eq!(d.header.vars.iter().filter(|(n, _)| *n == "$INSUNITS" || *n == "$ACADVER" || *n == "$HANDSEED").count(), 3); assert_eq!(d.header.point("$EXTMAX"), Some([100.0, 50.0, 0.0])); + assert_eq!(d.header.vars.iter().find(|(name, _)| *name == "$LUNITS").map(|(_, tags)| tags[0].i64()), Some(2)); // Blocks, their records and the layouts that name them. let names = d.block_names(); let block = d.block("DOOR").expect("the block"); diff --git a/crates/eps/Cargo.toml b/crates/eps/Cargo.toml index fbe69c26f..6984fb576 100644 --- a/crates/eps/Cargo.toml +++ b/crates/eps/Cargo.toml @@ -11,3 +11,11 @@ repository.workspace = true workspace = true [dependencies] +kurbo = { workspace = true, optional = true } +vectorcraft-doc = { workspace = true, optional = true } +vectorcraft-eps = { workspace = true, optional = true } +thiserror = { workspace = true, optional = true } + +[features] +default = [] +paths = ["dep:kurbo", "dep:thiserror", "dep:vectorcraft-doc", "dep:vectorcraft-eps"] diff --git a/crates/eps/README.md b/crates/eps/README.md index f2abd0b2c..25c8a7eda 100644 --- a/crates/eps/README.md +++ b/crates/eps/README.md @@ -8,8 +8,10 @@ nothing. |---|---| | `sniff`, `is_eps` | EPS (binary header, or `%!PS-Adobe-… EPSF-…`), plain PostScript or PDF, and whether it is an Illustrator file: PostScript-based (`%AI…` comments, or Illustrator as `%%Creator`) or PDF-based (the `AIPrivateData` / `AIPDFPrivateData` keys near the start or end of the PDF, a hint only) | | `sections` | The parts behind the binary header ("DOS EPS"): the PostScript, the TIFF preview, the Windows metafile preview. A header that points outside the file is an error | +| `postscript_section` | The PostScript program alone; a damaged preview does not prevent a reader from using a valid program section | | `wrap` | PostScript and its previews behind a binary header, as `sections` reads them back | | `Comments::read` | `%%BoundingBox` and `%%HiResBoundingBox` (the first in the header; `(atend)` taken from the trailer, outside embedded documents), `%%Pages`, `%%LanguageLevel`, `%%Creator`, `%%Title` | +| `Comments::read_with_header_limit` | The same comments with a caller-selected header scan limit capped at 256 KiB; `read` defaults to 64 KiB | | `epsi` | The device-independent preview (`%%BeginPreview`, 1, 2, 4 or 8 bits) as an 8-bit grey bitmap | | `ascii85::encode`, `ascii85::decode` | The ASCII85 data encoding; lines never start with `%`, decoding is bounded | diff --git a/crates/eps/src/container.rs b/crates/eps/src/container.rs index ba721f5c2..245252b83 100644 --- a/crates/eps/src/container.rs +++ b/crates/eps/src/container.rs @@ -27,6 +27,27 @@ pub struct Sections<'a> { pub metafile: Option<&'a [u8]>, } +fn word(bytes: &[u8], at: usize) -> Result { + let end = at.checked_add(4).ok_or(Error::OutsideFile)?; + let b = bytes.get(at..end).and_then(|b| <[u8; 4]>::try_from(b).ok()).ok_or(Error::OutsideFile)?; + usize::try_from(u32::from_le_bytes(b)).map_err(|_| Error::OutsideFile) +} + +fn part(bytes: &[u8], at: usize) -> Result<&[u8], Error> { + let len_at = at.checked_add(4).ok_or(Error::OutsideFile)?; + let (start, len) = (word(bytes, at)?, word(bytes, len_at)?); + start.checked_add(len).and_then(|end| bytes.get(start..end)).ok_or(Error::OutsideFile) +} + +/// The PostScript program in an EPS, ignoring preview ranges in its binary header. +/// +/// This is useful to readers that execute only PostScript and historically accepted a valid +/// program section even when a separate preview range was damaged. Use [`sections`] when every +/// declared section must be validated. +pub fn postscript_section(bytes: &[u8]) -> Result<&[u8], Error> { + if bytes.starts_with(&DOS_MAGIC) { part(bytes, 4) } else { Ok(bytes) } +} + /// The sections of an EPS file: its PostScript and previews behind a binary header, or the whole /// file as PostScript. A header that points outside the file is an error, whichever section it /// is: such a file is cut short or not an EPS. @@ -34,17 +55,12 @@ pub fn sections(bytes: &[u8]) -> Result, Error> { if !bytes.starts_with(&DOS_MAGIC) { return Ok(Sections { postscript: bytes, tiff: None, metafile: None }); } - let word = |at: usize| -> Result { - let b = bytes.get(at..at.saturating_add(4)).and_then(|b| <[u8; 4]>::try_from(b).ok()).ok_or(Error::OutsideFile)?; - usize::try_from(u32::from_le_bytes(b)).map_err(|_| Error::OutsideFile) - }; - let part = |at: usize| -> Result<&[u8], Error> { - let (start, len) = (word(at)?, word(at + 4)?); - start.checked_add(len).and_then(|end| bytes.get(start..end)).ok_or(Error::OutsideFile) - }; // A preview with no offset or no length is absent. - let preview = |at: usize| -> Result, Error> { if word(at)? == 0 || word(at + 4)? == 0 { Ok(None) } else { part(at).map(Some) } }; - Ok(Sections { postscript: part(4)?, metafile: preview(12)?, tiff: preview(20)? }) + let preview = |at: usize| -> Result, Error> { + let len_at = at.checked_add(4).ok_or(Error::OutsideFile)?; + if word(bytes, at)? == 0 || word(bytes, len_at)? == 0 { Ok(None) } else { part(bytes, at).map(Some) } + }; + Ok(Sections { postscript: postscript_section(bytes)?, metafile: preview(12)?, tiff: preview(20)? }) } /// `postscript` behind a binary header, followed by its previews. With neither preview the diff --git a/crates/eps/src/dsc.rs b/crates/eps/src/dsc.rs index 835f9b23f..2334e1d90 100644 --- a/crates/eps/src/dsc.rs +++ b/crates/eps/src/dsc.rs @@ -2,6 +2,8 @@ /// Bytes of a file's start read for its header comments. pub const MAX_HEADER: usize = 64 << 10; +/// Largest header scan available for readers preserving a wider historical bound. +pub const MAX_EXTENDED_HEADER: usize = 256 << 10; /// Bytes of a file's end read for comments its header defers there with `(atend)`. pub const MAX_TRAILER: usize = 64 << 10; @@ -34,7 +36,17 @@ impl Comments { /// A bounding box the header defers with `(atend)` is the last one of the trailer, outside /// any document the file embeds. pub fn read(ps: &[u8]) -> Self { - let head = String::from_utf8_lossy(ps.get(..MAX_HEADER).unwrap_or(ps)); + Self::read_with_header_limit(ps, MAX_HEADER) + } + + /// Read comments with a caller-selected scan limit for the header. + /// + /// The limit is capped at [`MAX_EXTENDED_HEADER`], and the trailer remains capped at + /// [`MAX_TRAILER`]. This lets readers preserve a previously supported wider scan without + /// making header or trailer work unbounded. + pub fn read_with_header_limit(ps: &[u8], header_limit: usize) -> Self { + let header_limit = header_limit.min(MAX_EXTENDED_HEADER).min(ps.len()); + let head = String::from_utf8_lossy(&ps[..header_limit]); let mut c = Comments::default(); let (mut box_atend, mut hires_atend) = (false, false); // Lines may end in `\r` alone (classic Mac files). diff --git a/crates/eps/src/embedded.rs b/crates/eps/src/embedded.rs new file mode 100644 index 000000000..8e2225993 --- /dev/null +++ b/crates/eps/src/embedded.rs @@ -0,0 +1,269 @@ +//! Embedded Type 1 and CFF font programs in PostScript / EPS source. + +/// A font program embedded in the PostScript source of an EPS file. +#[derive(Debug, Clone, PartialEq, Eq)] +pub struct EmbeddedFontProgram { + /// The PostScript `/FontName` when one is present in the source. CFF `StartData` blocks store + /// their name inside the CFF Name INDEX, which the font parser reads from `program`. + pub name: Option, + /// The bytes of the Type 1 program or CFF font set. + pub program: Vec, + /// Whether `program` is a CFF font set rather than a Type 1 program. + pub cff: bool, +} + +/// Embedded font programs in a PostScript section, in source order within each format. +/// +/// Type 1 programs are recognized from their `eexec` and `cleartomark` delimiters and are named +/// by `/FontName`. CFF `StartData` blocks are returned without a name because the name is inside +/// the binary CFF data. Malformed or truncated blocks are skipped, as in the original EPS reader. +pub fn embedded_font_programs(ps: &[u8]) -> Vec { + let mut out = Vec::new(); + let mut heads = positions(ps, b"%!PS-AdobeFont"); + heads.extend(positions(ps, b"%!FontType1")); + heads.sort_unstable(); + let names = positions(ps, b"/FontName"); + let mut from = 0; + let mut used = 0; + while let Some(ee) = find_eexec(ps, from) { + from = ee + 5; + let last_before = |v: &[usize]| v[..v.partition_point(|&p| p < ee)].last().copied().filter(|&p| p >= used); + // The fallback look-back stops at the previous font's end, so it cannot pick up its header. + let start = last_before(&heads).or_else(|| last_before(&names).map(|p| p.saturating_sub(512).max(used))); + let Some(start) = start else { continue }; + let end = find_from(ps, b"cleartomark", ee).map_or(ps.len(), |p| p + 11); + if let Some(name) = name_after(&ps[start..ee], b"/FontName") { + out.push(EmbeddedFontProgram { name: Some(name), program: ps[start..end].to_vec(), cff: false }); + } + from = end; + used = end; + } + + let mut from = 0; + while let Some(sd) = find_from(ps, b"StartData", from) { + from = sd + 9; + let head = String::from_utf8_lossy(&ps[sd.saturating_sub(64)..sd]).into_owned(); + let Some(len) = head.split_whitespace().last().and_then(|t| t.parse::().ok()) else { continue }; + let s = sd + 10; + let Some(data) = ps.get(s..s.saturating_add(len)) else { continue }; + out.push(EmbeddedFontProgram { name: None, program: data.to_vec(), cff: true }); + from = s + len; + } + out +} + +fn find_from(hay: &[u8], needle: &[u8], from: usize) -> Option { + hay.get(from..)?.windows(needle.len()).position(|w| w == needle).map(|p| p + from) +} + +/// Find an executable `eexec` operator, ignoring comments, literal strings and names. +fn find_eexec(ps: &[u8], from: usize) -> Option { + let mut pos = from.min(ps.len()); + while pos < ps.len() { + let c = ps[pos]; + if is_white(c) { + pos += 1; + continue; + } + if c == b'%' { + let start = pos; + while pos < ps.len() && !matches!(ps[pos], b'\n' | b'\r') { + pos += 1; + } + let line = &ps[start..pos]; + let marker: Option<&[u8]> = if line.starts_with(b"%%BeginData:") { + Some(b"%%EndData") + } else if line.starts_with(b"%%BeginBinary:") { + Some(b"%%EndBinary") + } else if line.starts_with(b"%%BeginPreview") { + Some(b"%%EndPreview") + } else { + None + }; + if let Some(marker) = marker { + let rest = ps.get(pos..)?; + pos = rest.windows(marker.len()).position(|window| window == marker).map_or(ps.len(), |offset| pos + offset); + } + continue; + } + if c == b'(' { + pos += 1; + let mut depth = 1usize; + while pos < ps.len() && depth > 0 { + match ps[pos] { + b'\\' => pos = pos.saturating_add(2).min(ps.len()), + b'(' => { + depth = depth.saturating_add(1); + pos += 1; + } + b')' => { + depth -= 1; + pos += 1; + } + _ => pos += 1, + } + } + continue; + } + if c == b'<' { + match ps.get(pos + 1) { + Some(b'<') => pos += 2, + Some(b'~') => { + pos += 2; + pos = ps + .get(pos..) + .and_then(|rest| { + rest.windows(2).position(|window| window == b"~>").map(|offset| pos.saturating_add(offset).saturating_add(2)) + }) + .unwrap_or(ps.len()); + } + _ => { + pos += 1; + while pos < ps.len() && ps[pos] != b'>' { + pos += 1; + } + pos = pos.saturating_add(1).min(ps.len()); + } + } + continue; + } + if c == b'/' { + pos += 1; + if ps.get(pos) == Some(&b'/') { + pos += 1; + } + while pos < ps.len() && !is_white(ps[pos]) && !is_delimiter(ps[pos]) { + pos += 1; + } + continue; + } + if is_delimiter(c) { + pos += 1; + continue; + } + let start = pos; + while pos < ps.len() && !is_white(ps[pos]) && !is_delimiter(ps[pos]) { + pos += 1; + } + if &ps[start..pos] == b"eexec" { + return Some(start); + } + } + None +} + +fn is_white(c: u8) -> bool { + matches!(c, 0 | 9 | 10 | 12 | 13 | 32) +} + +fn is_delimiter(c: u8) -> bool { + matches!(c, b'(' | b')' | b'<' | b'>' | b'[' | b']' | b'{' | b'}' | b'/' | b'%') +} + +fn positions(hay: &[u8], needle: &[u8]) -> Vec { + let mut out = vec![]; + let mut from = 0; + while let Some(p) = find_from(hay, needle, from) { + out.push(p); + from = p + needle.len(); + } + out +} + +fn name_after(b: &[u8], key: &[u8]) -> Option { + let p = find_from(b, key, 0)? + key.len(); + let rest = &b[p..]; + let s = rest.iter().position(|c| *c == b'/')? + 1; + let e = rest[s..].iter().position(|c| is_white(*c) || is_delimiter(*c)).map_or(rest.len(), |q| s + q); + (e > s).then(|| String::from_utf8_lossy(&rest[s..e]).into_owned()) +} + +#[cfg(test)] +mod tests { + #![allow(clippy::unwrap_used, clippy::expect_used, clippy::panic)] + + use super::embedded_font_programs; + + #[test] + fn finds_type1_font_programs_by_name() { + let ps = b"%!PS-Adobe-3.0 EPSF-3.0\n%!PS-AdobeFont-1.0: Test\n/FontName /Test def\ncurrentfile eexec\n0000\ncleartomark\n"; + let fonts = embedded_font_programs(ps); + assert_eq!(fonts.len(), 1); + assert_eq!(fonts[0].name.as_deref(), Some("Test")); + assert!(!fonts[0].cff); + assert!(String::from_utf8_lossy(&fonts[0].program).contains("cleartomark")); + } + + #[test] + fn finds_cff_startdata_without_interpreting_its_name_index() { + let ps = b"%!PS-Adobe-3.0 EPSF-3.0\n10 StartData\n0123456789\n"; + let fonts = embedded_font_programs(ps); + assert_eq!(fonts.len(), 1); + assert_eq!(fonts[0].name, None); + assert!(fonts[0].cff); + assert_eq!(fonts[0].program, b"0123456789"); + } + + #[test] + fn ignores_truncated_startdata() { + let fonts = embedded_font_programs(b"%!PS-Adobe-3.0 EPSF-3.0\n100 StartData\nshort"); + assert!(fonts.is_empty()); + } + + #[test] + fn many_headerless_eexec_words_do_not_hide_later_font() { + let mut ps = b"%!PS-Adobe-3.0 EPSF-3.0\n".to_vec(); + ps.extend_from_slice(b"eexec ".repeat(20_000).as_slice()); + ps.extend_from_slice(b"\n%!PS-AdobeFont-1.0: Real\n/FontName /Real def\ncurrentfile eexec\n0000\ncleartomark\n"); + let fonts = embedded_font_programs(&ps); + assert_eq!(fonts.len(), 1); + assert_eq!(fonts[0].name.as_deref(), Some("Real")); + } + + #[test] + fn headerless_font_does_not_reuse_previous_font_header() { + let first = b"%!PS-AdobeFont-1.0: A\n/FontName /A def\ncurrentfile eexec\n0000\ncleartomark\n"; + let second = b"/FontName /B def\ncurrentfile eexec\n0000\ncleartomark\n"; + let mut ps = b"%!PS-Adobe-3.0 EPSF-3.0\n".to_vec(); + ps.extend_from_slice(first); + ps.extend_from_slice(second); + let fonts = embedded_font_programs(&ps); + assert_eq!(fonts.len(), 2); + let b = fonts.iter().find(|font| font.name.as_deref() == Some("B")).unwrap(); + assert!(!String::from_utf8_lossy(&b.program).contains("AdobeFont")); + } + + #[test] + fn eexec_in_comments_strings_names_and_font_names_is_not_an_operator() { + let ps = b"%!PS-Adobe-3.0 EPSF-3.0\n%!PS-AdobeFont-1.0: Test\n%% eexec in a comment\n%%BeginBinary: 10\neexec junk\n%%EndBinary\n/FontName /eexecTest def\n(nested (ignored eexec)) pop\n<6565786563> pop\n<~eexec~> pop\n/eexec def\ncurrentfile eexec\n0000\ncleartomark\n"; + let fonts = embedded_font_programs(ps); + assert_eq!(fonts.len(), 1); + assert_eq!(fonts[0].name.as_deref(), Some("eexecTest")); + } + + #[test] + fn font_name_stops_at_postscript_delimiters_and_comments() { + let ps = b"%!PS-AdobeFont-1.0: Test\n/FontName /Test%comment\n def\ncurrentfile eexec\n0000\ncleartomark\n"; + let fonts = embedded_font_programs(ps); + assert_eq!(fonts.len(), 1); + assert_eq!(fonts[0].name.as_deref(), Some("Test")); + } + + #[test] + fn font_names_stop_at_every_postscript_delimiter() { + for delimiter in b"()<>[]{}/%" { + let mut source = b"/FontName /Test".to_vec(); + source.push(*delimiter); + source.extend_from_slice(b"ignored"); + assert_eq!(super::name_after(&source, b"/FontName").as_deref(), Some("Test")); + } + } + + #[test] + fn literal_name_backslash_does_not_escape_a_comment_delimiter() { + let ps = b"%!PS-AdobeFont-1.0: Test\n/FontName /Test def\n/unused\\%(\npop\ncurrentfile eexec\n0000\ncleartomark\n"; + let fonts = embedded_font_programs(ps); + assert_eq!(fonts.len(), 1); + assert_eq!(fonts[0].name.as_deref(), Some("Test")); + } +} diff --git a/crates/eps/src/lib.rs b/crates/eps/src/lib.rs index acffdfcf2..c9a2574f5 100644 --- a/crates/eps/src/lib.rs +++ b/crates/eps/src/lib.rs @@ -18,11 +18,18 @@ pub mod ascii85; mod container; mod dsc; +mod embedded; +#[cfg(feature = "paths")] +pub mod paths; +pub mod postscript; mod preview; +mod rectangles; -pub use container::{DOS_HEADER, DOS_MAGIC, Sections, sections, wrap}; -pub use dsc::{Comments, MAX_HEADER, MAX_TRAILER}; +pub use container::{DOS_HEADER, DOS_MAGIC, Sections, postscript_section, sections, wrap}; +pub use dsc::{Comments, MAX_EXTENDED_HEADER, MAX_HEADER, MAX_TRAILER}; +pub use embedded::{EmbeddedFontProgram, embedded_font_programs}; pub use preview::{Bitmap, MAX_EPSI, epsi}; +pub use rectangles::{MAX_SIMPLE_RECTANGLE_SAVE_DEPTH, MAX_SIMPLE_RECTANGLE_TOKENS, MAX_SIMPLE_RECTANGLES, Rectangle, simple_rectangles}; use std::fmt; diff --git a/crates/eps/src/paths.rs b/crates/eps/src/paths.rs new file mode 100644 index 000000000..271f905bd --- /dev/null +++ b/crates/eps/src/paths.rs @@ -0,0 +1,108 @@ +//! Neutral path extraction for EPS and PostScript files. +//! +//! This facade keeps the PostScript interpreter and its format decisions in shared dependencies; +//! applications receive BĂ©zier paths, page bounds and warnings for their own document adapters. + +use kurbo::{BezPath, Rect}; + +/// Limits on the extracted geometry. The interpreter has its own execution and allocation caps. +#[derive(Debug, Clone, Copy)] +pub struct Limits { + pub max_input_bytes: usize, + pub max_paths: usize, + pub max_path_elements: usize, +} + +impl Default for Limits { + fn default() -> Self { + Self { max_input_bytes: 64 * 1024 * 1024, max_paths: 200_000, max_path_elements: 2_000_000 } + } +} + +#[derive(Debug, thiserror::Error)] +pub enum Error { + #[error("PostScript input exceeds {limit} bytes")] + InputLimit { limit: usize }, + #[error("PostScript import failed: {0}")] + Parse(String), + #[error("PostScript path output exceeds configured limit")] + OutputLimit, +} + +/// First-page path geometry normalized to the top-left origin in PostScript points, plus notes. +#[derive(Debug, Clone)] +pub struct Imported { + pub paths: Vec, + /// Origin-zero page bounds in PostScript points, using the same accepted DSC box as the + /// interpreter. + pub page_bounds: Rect, + pub warnings: Vec, + pub preview_fallback: bool, +} + +/// Interpret EPS/PostScript bytes and return their editable vector path geometry. +pub fn import(bytes: &[u8], limits: Limits) -> Result { + if bytes.len() > limits.max_input_bytes { + return Err(Error::InputLimit { limit: limits.max_input_bytes }); + } + let source = vectorcraft_eps::import(bytes).map_err(Error::Parse)?; + let page_bounds = vectorcraft_eps::page(bytes).map_err(Error::Parse)?; + // Illustrator's editable copy is expressed in its art-box document coordinates. The page + // reader, in contrast, returns a page at the origin. Normalize extracted paths to match it. + let origin = source.document.artboards.first().map(|board| board.rect.origin()).unwrap_or_default(); + let to_page = kurbo::Affine::translate((-origin.x, -origin.y)); + let mut paths = Vec::new(); + let mut total_elements = 0usize; + let mut clipped = false; + let mut omitted = false; + for layer in &source.document.layers { + collect(layer, &mut paths, &mut total_elements, limits, to_page, &mut clipped, &mut omitted)?; + } + let mut warnings = source.warnings; + if clipped { + warnings.push("EPS clipping groups are returned as unclipped artwork".into()); + } + if omitted { + warnings.push("visible EPS text or image content is not returned as paths".into()); + } + Ok(Imported { paths, page_bounds, warnings, preview_fallback: source.preview }) +} + +fn collect( + node: &vectorcraft_doc::Node, + paths: &mut Vec, + elements: &mut usize, + limits: Limits, + to_page: kurbo::Affine, + clipped: &mut bool, + omitted: &mut bool, +) -> Result<(), Error> { + if !node.visible { + return Ok(()); + } + if let Some(data) = node.path_data() { + if matches!(node.kind, vectorcraft_doc::NodeKind::Path { clipping: true, .. }) { + return Ok(()); + } + let path = data.to_bezpath(); + if !path.elements().is_empty() { + *elements = elements.saturating_add(path.elements().len()); + if paths.len() >= limits.max_paths || *elements > limits.max_path_elements { + return Err(Error::OutputLimit); + } + paths.push(to_page * path); + } + } else if let Some(children) = node.children() { + let is_clip = matches!(node.kind, vectorcraft_doc::NodeKind::Group { clip: true, .. } | vectorcraft_doc::NodeKind::Layer { clip: true, .. }); + *clipped |= is_clip; + for (index, child) in children.iter().enumerate() { + if is_clip && index == 0 { + continue; + } + collect(child, paths, elements, limits, to_page, clipped, omitted)?; + } + } else { + *omitted = true; + } + Ok(()) +} diff --git a/crates/eps/src/postscript.rs b/crates/eps/src/postscript.rs new file mode 100644 index 000000000..3261b3193 --- /dev/null +++ b/crates/eps/src/postscript.rs @@ -0,0 +1,873 @@ +//! Shared PostScript format primitives for EPS programs. +//! +//! [`Scanner`] handles bounded lexical parsing. [`OperandStack`], [`ExecutionBudget`], and +//! [`NumericOperator`] provide reusable language-level VM behavior while leaving values, +//! dictionaries, graphics state, rendering, and font policy to the host interpreter. + +/// Maximum nesting depth for executable procedures in a source token. +pub const MAX_PROCEDURE_DEPTH: usize = 100; + +/// Maximum number of operand values an interpreter should retain by default. +pub const MAX_OPERAND_STACK: usize = 100_000; + +/// Maximum nested executable procedure calls supported by the bounded VM. +pub const MAX_CALL_DEPTH: usize = 256; + +/// PostScript operand-stack operators whose semantics do not depend on the value model. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum StackOperator { + Pop, + Exchange, + Duplicate, + CopyTop(usize), + Index(usize), + Roll { count: usize, shift: i64 }, + Clear, + Count, +} + +/// Result of applying an operand-stack operator. +#[derive(Debug, PartialEq, Eq)] +pub enum StackOutcome { + None, + Popped(Option), + Count(usize), + Copied(bool), +} + +/// A bounded PostScript operand stack. +/// +/// The stack is deliberately generic: an interpreter owns the concrete PostScript values and +/// graphics state, while this module provides the shared language-level stack operators. A push +/// past `limit` is ignored, matching the bounded EPS interpreter's defensive behavior. +#[derive(Clone, Debug)] +pub struct OperandStack { + values: Vec, + limit: usize, +} + +impl OperandStack { + /// Create an empty stack with a maximum number of values. + pub fn new(limit: usize) -> Self { + Self { values: Vec::new(), limit } + } + + /// Number of values currently on the stack. + pub fn len(&self) -> usize { + self.values.len() + } + + /// Whether the stack is empty. + pub fn is_empty(&self) -> bool { + self.values.is_empty() + } + + /// Borrow all stack values mutably from bottom to top. + pub fn values_mut(&mut self) -> &mut [T] { + &mut self.values + } + + /// Remove and return the top value, if present. + pub fn pop(&mut self) -> Option { + self.values.pop() + } + + /// Remove every value. + pub fn clear(&mut self) { + self.values.clear(); + } + + /// Remove values at and above `at`, returning them in bottom-to-top order. An index beyond + /// the stack length removes nothing. + pub fn split_off(&mut self, at: usize) -> Vec { + let at = at.min(self.values.len()); + self.values.split_off(at) + } + + /// Push a value if the configured limit allows it; report whether it was pushed. + pub fn push(&mut self, value: T) -> bool { + if self.values.len() >= self.limit { + return false; + } + self.values.push(value); + true + } + + /// Pop a value, or return the caller's language-specific underflow value. + pub fn pop_or(&mut self, underflow: T) -> T { + self.values.pop().unwrap_or(underflow) + } + + /// Borrow the top value. + pub fn top(&self) -> Option<&T> { + self.values.last() + } + + /// PostScript `dup`: duplicate the top value if one exists. + pub fn duplicate(&mut self) { + if let Some(value) = self.values.last().cloned() { + let _ = self.push(value); + } + } + + /// PostScript `dup` with a caller-supplied underflow value. + pub fn duplicate_or(&mut self, underflow: T) { + let value = self.values.last().cloned().unwrap_or(underflow); + let _ = self.push(value); + } + + /// PostScript `exch`: exchange the top two values when both exist. + pub fn exchange(&mut self) { + let len = self.values.len(); + if len >= 2 { + self.values.swap(len - 1, len - 2); + } + } + + /// PostScript `exch` with caller-supplied underflow values. Both operands are popped and + /// restored even if either pop underflows, matching interpreters that use a null sentinel. + pub fn exchange_or(&mut self, underflow: T) { + let b = self.pop_or(underflow.clone()); + let a = self.pop_or(underflow); + let _ = self.push(b); + let _ = self.push(a); + } + + /// PostScript `index`: copy the value `depth` below the top onto the stack. + /// + /// Invalid or negative depths are left to the caller to validate; an out-of-range depth is a + /// no-op. This keeps the helper bounded even for hostile numeric operands. + pub fn index(&mut self, depth: usize) { + let len = self.values.len(); + if depth < len && len < self.limit { + let value = self.values[len - depth - 1].clone(); + self.values.push(value); + } + } + + /// PostScript `index` with a caller-supplied underflow value, pushed even when `depth` is out + /// of range. The configured stack limit still applies. + pub fn index_or(&mut self, depth: usize, underflow: T) { + let len = self.values.len(); + let value = if depth < len { self.values[len - depth - 1].clone() } else { underflow }; + let _ = self.push(value); + } + + /// PostScript `n j roll`: rotate the top `count` values by signed `shift` places. + /// + /// Positive shifts move values toward the top; negative shifts move them toward the bottom. + /// Counts larger than the stack and empty rolls are ignored. + pub fn roll(&mut self, count: usize, shift: i64) { + let len = self.values.len(); + if count == 0 || count > len { + return; + } + let places = (i128::from(shift).rem_euclid(count as i128)) as usize; + self.values[len - count..].rotate_right(places); + } + + /// PostScript `copy` for an integer operand: append a copy of the top `count` values if it + /// fits. Returns false when the range is invalid or the configured limit would be exceeded. + pub fn copy_top(&mut self, count: usize) -> bool { + let len = self.values.len(); + if count > len || len.saturating_add(count) > self.limit { + return false; + } + let copied = self.values[len - count..].to_vec(); + self.values.extend(copied); + true + } + + /// Borrow all stack values from bottom to top. + pub fn values(&self) -> &[T] { + &self.values + } + + /// Apply a PostScript operand-stack operator. `underflow` is the interpreter's fallback value + /// for `exch`, `dup`, and `index`; the stack's configured limit remains in force. + pub fn execute(&mut self, operator: StackOperator, underflow: T) -> StackOutcome { + match operator { + StackOperator::Pop => StackOutcome::Popped(self.pop()), + StackOperator::Exchange => { + self.exchange_or(underflow); + StackOutcome::None + } + StackOperator::Duplicate => { + self.duplicate_or(underflow); + StackOutcome::None + } + StackOperator::CopyTop(count) => StackOutcome::Copied(self.copy_top(count)), + StackOperator::Index(depth) => { + self.index_or(depth, underflow); + StackOutcome::None + } + StackOperator::Roll { count, shift } => { + self.roll(count, shift); + StackOutcome::None + } + StackOperator::Clear => { + self.clear(); + StackOutcome::None + } + StackOperator::Count => StackOutcome::Count(self.len()), + } + } +} + +/// Bounded execution accounting shared by PostScript interpreters. +/// +/// EPS programs can recurse or repeatedly execute procedures. A host can use this counter to +/// stop work before it consumes unbounded CPU, independently of its operator and graphics policy. +#[derive(Clone, Debug)] +pub struct ExecutionBudget { + remaining: usize, + call_depth: usize, + max_call_depth: usize, +} + +impl ExecutionBudget { + /// Create a budget with an operation allowance and maximum nested procedure calls. + pub fn new(operations: usize, max_call_depth: usize) -> Self { + Self { remaining: operations, call_depth: 0, max_call_depth } + } + + /// Units of work still available. + pub fn remaining(&self) -> usize { + self.remaining + } + + /// Current nested procedure depth. + pub fn call_depth(&self) -> usize { + self.call_depth + } + + /// Charge work atomically. On failure the remaining budget stays unchanged. + pub fn spend(&mut self, units: usize) -> bool { + let Some(remaining) = self.remaining.checked_sub(units) else { + return false; + }; + self.remaining = remaining; + true + } + + /// Charge work that is accounted lazily: saturate at zero so the caller's next strict + /// [`Self::spend`] stops execution. This supports work whose current operator preserves the + /// app's historical behavior after exhausting its budget. + pub fn charge(&mut self, units: usize) { + self.remaining = self.remaining.saturating_sub(units); + } + + /// Enter a procedure call if it remains within the configured depth. + pub fn enter_call(&mut self) -> bool { + if self.call_depth >= self.max_call_depth { + return false; + } + self.call_depth += 1; + true + } + + /// Leave a procedure call. Extra leaves are ignored to keep cleanup paths safe. + pub fn leave_call(&mut self) { + self.call_depth = self.call_depth.saturating_sub(1); + } +} + +/// Numeric operators used by PostScript's value-independent arithmetic core. +/// +/// The host interpreter remains responsible for popping operands and selecting a result value, +/// but these operations define the shared arithmetic semantics. Non-finite results become zero, +/// matching the defensive finite-value policy used by the host interpreter. +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub enum NumericOperator { + Add, + Subtract, + Multiply, + Divide, + IntegerDivide, + Modulo, + Atan, + Exponentiate, + Maximum, + Minimum, + Negate, + Absolute, + SquareRoot, + Sine, + Cosine, + NaturalLog, + CommonLog, + ToInteger, + ToReal, + Round, + Floor, + Ceiling, + Truncate, +} + +impl NumericOperator { + /// Resolve a standard PostScript operator name to its numeric operation. + pub fn from_name(name: &str) -> Option { + Some(match name { + "add" => Self::Add, + "sub" => Self::Subtract, + "mul" => Self::Multiply, + "div" => Self::Divide, + "idiv" => Self::IntegerDivide, + "mod" => Self::Modulo, + "atan" => Self::Atan, + "exp" => Self::Exponentiate, + "max" => Self::Maximum, + "min" => Self::Minimum, + "neg" => Self::Negate, + "abs" => Self::Absolute, + "sqrt" => Self::SquareRoot, + "sin" => Self::Sine, + "cos" => Self::Cosine, + "ln" => Self::NaturalLog, + "log" => Self::CommonLog, + "cvi" => Self::ToInteger, + "cvr" => Self::ToReal, + "round" => Self::Round, + "floor" => Self::Floor, + "ceiling" => Self::Ceiling, + "truncate" => Self::Truncate, + _ => return None, + }) + } + + /// Number of operands the operation consumes. + pub const fn arity(self) -> usize { + match self { + Self::Add + | Self::Subtract + | Self::Multiply + | Self::Divide + | Self::IntegerDivide + | Self::Modulo + | Self::Atan + | Self::Exponentiate + | Self::Maximum + | Self::Minimum => 2, + _ => 1, + } + } + + /// Apply a binary operator. Returns `None` for an operator that takes one operand. + pub fn apply_binary(self, a: f64, b: f64) -> Option { + let value = match self { + Self::Add => a + b, + Self::Subtract => a - b, + Self::Multiply => a * b, + Self::Divide => { + if b == 0.0 { + 0.0 + } else { + a / b + } + } + // Rust's float-to-int cast saturates, as the app's previous VM did. checked_div/rem + // also make both zero-divisor and i64::MIN / -1 cases deterministic. + Self::IntegerDivide => (a as i64).checked_div(b as i64).unwrap_or(0) as f64, + Self::Modulo => (a as i64).checked_rem(b as i64).unwrap_or(0) as f64, + Self::Atan => a.atan2(b).to_degrees().rem_euclid(360.0), + Self::Exponentiate => a.powf(b), + Self::Maximum => a.max(b), + Self::Minimum => a.min(b), + _ => return None, + }; + Some(finite_or_zero(value)) + } + + /// Apply a unary operator. Returns `None` for an operator that takes two operands. + pub fn apply_unary(self, a: f64) -> Option { + let value = match self { + Self::Negate => -a, + Self::Absolute => a.abs(), + Self::SquareRoot => a.max(0.0).sqrt(), + Self::Sine => a.to_radians().sin(), + Self::Cosine => a.to_radians().cos(), + Self::NaturalLog => a.ln(), + Self::CommonLog => a.log10(), + Self::ToInteger | Self::Truncate => a.trunc(), + Self::ToReal => a, + Self::Round => (a + 0.5).floor(), + Self::Floor => a.floor(), + Self::Ceiling => a.ceil(), + _ => return None, + }; + Some(finite_or_zero(value)) + } +} + +fn finite_or_zero(value: f64) -> f64 { + if value.is_finite() { value } else { 0.0 } +} + +/// A lexical PostScript value. Names are left unevaluated for the caller's interpreter. +#[derive(Clone, Debug, PartialEq)] +pub enum Token { + /// A number accepted by the PostScript source reader. + Number(f64), + /// An executable name, including immediately evaluated `//name`. + Name(String), + /// A literal name such as `/name`. + LiteralName(String), + /// A parenthesized or hexadecimal byte string. + String(Vec), + /// The encoded bytes between `<~` and `~>` (including delimiters), not decoded data. + Ascii85(Vec), + /// A procedure whose nested values have already been tokenized. + Procedure(Vec), + /// A closing brace not consumed as part of a procedure, or another malformed delimiter. + Null, +} + +/// PostScript source bytes and the current byte offset. +pub struct Scanner<'a> { + /// Source bytes. Public so an interpreter can implement `currentfile` operators. + pub data: &'a [u8], + /// Current byte offset in [`Scanner::data`]. + pub pos: usize, +} + +impl<'a> Scanner<'a> { + /// Start scanning at a byte offset, clamped to the source length. + pub fn new(data: &'a [u8], pos: usize) -> Self { + Self { data, pos: pos.min(data.len()) } + } + + /// The next token; `{ … }` is returned as one procedure token. + pub fn token(&mut self) -> Option { + self.token_at_depth(0) + } + + fn token_at_depth(&mut self, depth: usize) -> Option { + self.skip_ws(); + let c = *self.data.get(self.pos)?; + Some(match c { + b'{' => { + self.pos += 1; + let mut body = Vec::new(); + loop { + self.skip_ws(); + match self.data.get(self.pos) { + None => break, + Some(b'}') => { + self.pos += 1; + break; + } + _ if depth >= MAX_PROCEDURE_DEPTH => break, + _ => match self.token_at_depth(depth + 1) { + Some(t) => body.push(t), + None => break, + }, + } + } + Token::Procedure(body) + } + b'}' => { + self.pos += 1; + Token::Null + } + b'/' => { + self.pos += 1; + let immediately_evaluated = self.data.get(self.pos) == Some(&b'/'); + if immediately_evaluated { + self.pos += 1; + } + let name = String::from_utf8_lossy(self.word()).into_owned(); + if immediately_evaluated { Token::Name(name) } else { Token::LiteralName(name) } + } + b'(' => Token::String(self.parenthesized()), + b'<' if self.data.get(self.pos + 1) == Some(&b'<') => { + self.pos += 2; + Token::Name("<<".to_owned()) + } + b'>' if self.data.get(self.pos + 1) == Some(&b'>') => { + self.pos += 2; + Token::Name(">>".to_owned()) + } + b'<' if self.data.get(self.pos + 1) == Some(&b'~') => self.ascii85(), + b'<' => Token::String(self.hex_string()), + b'[' | b']' => { + self.pos += 1; + Token::Name(char::from(c).to_string()) + } + b')' | b'>' => { + self.pos += 1; + Token::Null + } + _ => { + let word = self.word(); + if word.is_empty() { + self.pos += 1; + return Some(Token::Null); + } + match parse_number(word) { + Some(n) => Token::Number(n), + None => { + let name = String::from_utf8_lossy(word).into_owned(); + if let Some((radix, digits)) = name.split_once('#') { + let radix = radix.parse::().ok().filter(|r| (2..=36).contains(r)); + if let Some(number) = radix.and_then(|r| i64::from_str_radix(digits, r).ok()) { + return Some(Token::Number(number as f64)); + } + } + Token::Name(name) + } + } + } + }) + } + + /// Step over inline image data that follows an `image` operator reading `currentfile`. + pub fn skip_image_data(&mut self) { + self.skip_ws(); + let rest = self.data.get(self.pos..).unwrap_or_default(); + if rest.starts_with(b"<~") || self.data.get(self.pos).is_some_and(|c| (b'!'..=b'u').contains(c) && !c.is_ascii_hexdigit()) { + self.pos = match rest.windows(2).position(|w| w == b"~>") { + Some(p) => self.pos.saturating_add(p).saturating_add(2).min(self.data.len()), + None => self.data.len(), + }; + return; + } + while self.pos < self.data.len() && (self.data[self.pos].is_ascii_hexdigit() || is_white(self.data[self.pos])) { + self.pos += 1; + } + if self.data.get(self.pos) == Some(&b'>') { + self.pos += 1; + } + } + + fn skip_ws(&mut self) { + while self.pos < self.data.len() { + let c = self.data[self.pos]; + if is_white(c) { + self.pos += 1; + } else if c == b'%' { + let start = self.pos; + while self.pos < self.data.len() && !matches!(self.data[self.pos], b'\n' | b'\r') { + self.pos += 1; + } + let line = &self.data[start..self.pos]; + let end_marker: Option<&[u8]> = if line.starts_with(b"%%BeginData:") { + Some(b"%%EndData") + } else if line.starts_with(b"%%BeginBinary:") { + Some(b"%%EndBinary") + } else if line.starts_with(b"%%BeginPreview") { + Some(b"%%EndPreview") + } else { + None + }; + if let Some(marker) = end_marker { + self.pos = match self.data.get(self.pos..).unwrap_or_default().windows(marker.len()).position(|w| w == marker) { + Some(offset) => self.pos.saturating_add(offset).min(self.data.len()), + None => self.data.len(), + }; + } + } else { + break; + } + } + } + + fn word(&mut self) -> &[u8] { + let start = self.pos; + while self.pos < self.data.len() { + let c = self.data[self.pos]; + if is_white(c) || matches!(c, b'(' | b')' | b'<' | b'>' | b'[' | b']' | b'{' | b'}' | b'/' | b'%') { + break; + } + self.pos += 1; + } + &self.data[start..self.pos] + } + + fn parenthesized(&mut self) -> Vec { + self.pos += 1; + let mut out = Vec::new(); + let mut nesting = 1usize; + while self.pos < self.data.len() { + let c = self.data[self.pos]; + self.pos += 1; + match c { + b'\\' => { + let e = self.data.get(self.pos).copied().unwrap_or(b'\\'); + self.pos = self.pos.saturating_add(1).min(self.data.len()); + match e { + b'n' => out.push(b'\n'), + b'r' => out.push(b'\r'), + b't' => out.push(b'\t'), + b'0'..=b'7' => { + let mut value = u32::from(e - b'0'); + for _ in 0..2 { + match self.data.get(self.pos) { + Some(d @ b'0'..=b'7') => { + value = value * 8 + u32::from(*d - b'0'); + self.pos += 1; + } + _ => break, + } + } + out.push(value as u8); + } + b'\n' | b'\r' => {} + other => out.push(other), + } + } + b'(' => { + nesting = nesting.saturating_add(1); + out.push(c); + } + b')' => { + nesting = nesting.saturating_sub(1); + if nesting == 0 { + break; + } + out.push(c); + } + _ => out.push(c), + } + } + out + } + + fn hex_string(&mut self) -> Vec { + self.pos += 1; + let mut hex = Vec::new(); + while self.pos < self.data.len() && self.data[self.pos] != b'>' { + if self.data[self.pos].is_ascii_hexdigit() { + hex.push(self.data[self.pos]); + } + self.pos += 1; + } + if self.data.get(self.pos) == Some(&b'>') { + self.pos += 1; + } + if hex.len() % 2 == 1 { + hex.push(b'0'); + } + hex.chunks(2).filter_map(|pair| Some((hex_value(*pair.first()?) << 4) | hex_value(*pair.get(1)?))).collect() + } + + fn ascii85(&mut self) -> Token { + let start = self.pos; + self.pos = self.pos.saturating_add(2).min(self.data.len()); + while self.pos + 1 < self.data.len() && !(self.data[self.pos] == b'~' && self.data[self.pos + 1] == b'>') { + self.pos += 1; + } + self.pos = self.pos.saturating_add(2).min(self.data.len()); + Token::Ascii85(self.data[start..self.pos].to_vec()) + } +} + +fn is_white(c: u8) -> bool { + matches!(c, 0 | 9 | 10 | 12 | 13 | 32) +} + +fn hex_value(c: u8) -> u8 { + (c as char).to_digit(16).unwrap_or(0) as u8 +} + +fn parse_number(word: &[u8]) -> Option { + if word.is_empty() || !word.iter().all(|c| c.is_ascii_digit() || matches!(c, b'+' | b'-' | b'.')) || !word.iter().any(u8::is_ascii_digit) { + return None; + } + let text = std::str::from_utf8(word).ok()?; + text.parse::() + .ok() + .or_else(|| text.trim_start_matches(['+', '-']).parse::().ok().map(|value| if text.starts_with('-') { -value } else { value })) +} + +#[cfg(test)] +mod stack_tests { + use super::{ExecutionBudget, OperandStack, StackOperator, StackOutcome}; + + #[test] + fn stack_operators_follow_postscript_order_and_respect_limit() { + let mut stack = OperandStack::new(8); + assert!(stack.push(1)); + assert!(stack.push(2)); + assert!(stack.push(3)); + stack.exchange(); + assert_eq!(stack.values(), &[1, 3, 2]); + stack.duplicate(); + assert_eq!(stack.values(), &[1, 3, 2, 2]); + stack.roll(4, 1); + assert_eq!(stack.values(), &[2, 1, 3, 2]); + stack.roll(4, -1); + assert_eq!(stack.values(), &[1, 3, 2, 2]); + stack.index(2); + assert_eq!(stack.values(), &[1, 3, 2, 2, 3]); + assert!(stack.copy_top(2)); + assert_eq!(stack.values(), &[1, 3, 2, 2, 3, 2, 3]); + stack.duplicate(); + assert_eq!(stack.len(), 8); + assert!(!stack.push(7)); + } + + #[test] + fn stack_underflow_and_invalid_rolls_are_safe() { + let mut stack = OperandStack::new(2); + assert_eq!(stack.pop_or(99), 99); + stack.duplicate_or(99); + assert_eq!(stack.values(), &[99]); + stack.clear(); + stack.exchange_or(99); + assert_eq!(stack.values(), &[99, 99]); + stack.clear(); + stack.roll(1, i64::MAX); + stack.index(usize::MAX); + assert!(stack.is_empty()); + assert!(stack.copy_top(0)); + assert!(stack.is_empty()); + } + + #[test] + fn stack_mutation_and_underflow_index_are_bounded() { + let mut stack = OperandStack::new(4); + assert!(stack.push(10)); + assert!(stack.push(20)); + stack.index_or(1, -1); + stack.index_or(9, -1); + assert_eq!(stack.values(), &[10, 20, 10, -1]); + stack.index_or(0, -1); + assert_eq!(stack.values(), &[10, 20, 10, -1]); // full stack silently drops the push + + assert_eq!(stack.top(), Some(&-1)); + stack.values_mut()[3] = 40; + stack.roll(3, i64::MIN); + assert_eq!(stack.values(), &[10, 40, 20, 10]); + assert!(stack.split_off(usize::MAX).is_empty()); + assert_eq!(stack.split_off(1), [40, 20, 10]); + assert_eq!(stack.values(), &[10]); + assert_eq!(stack.pop(), Some(10)); + assert_eq!(stack.pop(), None); + stack.clear(); + assert!(stack.is_empty()); + } + + #[test] + fn shared_stack_operator_dispatch_reports_only_value_results() { + let mut stack = OperandStack::new(4); + assert!(stack.push(1)); + assert!(stack.push(2)); + assert_eq!(stack.execute(StackOperator::Count, 0), StackOutcome::Count(2)); + assert_eq!(stack.execute(StackOperator::Exchange, 0), StackOutcome::None); + assert_eq!(stack.values(), &[2, 1]); + assert_eq!(stack.execute(StackOperator::Duplicate, 0), StackOutcome::None); + assert_eq!(stack.execute(StackOperator::Index(2), 0), StackOutcome::None); + assert_eq!(stack.values(), &[2, 1, 1, 2]); + assert_eq!(stack.execute(StackOperator::Roll { count: 3, shift: 1 }, 0), StackOutcome::None); + assert_eq!(stack.values(), &[2, 2, 1, 1]); + assert_eq!(stack.execute(StackOperator::CopyTop(2), 0), StackOutcome::Copied(false)); + assert_eq!(stack.execute(StackOperator::Pop, 0), StackOutcome::Popped(Some(1))); + assert_eq!(stack.execute(StackOperator::Clear, 0), StackOutcome::None); + assert!(stack.is_empty()); + } + + #[test] + fn execution_budget_charges_atomically_and_bounds_call_depth() { + let mut budget = ExecutionBudget::new(3, 1); + assert!(budget.spend(2)); + assert_eq!(budget.remaining(), 1); + assert!(!budget.spend(2)); + assert_eq!(budget.remaining(), 1); + budget.charge(2); + assert_eq!(budget.remaining(), 0); + assert!(!budget.spend(1)); + assert!(budget.enter_call()); + assert_eq!(budget.call_depth(), 1); + assert!(!budget.enter_call()); + budget.leave_call(); + budget.leave_call(); + assert_eq!(budget.call_depth(), 0); + assert!(budget.enter_call()); + } +} + +#[cfg(test)] +mod numeric_operator_tests { + use super::NumericOperator as N; + + #[test] + fn binary_numeric_operator_results_match_the_shared_vm_contract() { + assert_eq!(N::from_name("add"), Some(N::Add)); + assert_eq!(N::from_name("idiv").map(N::arity), Some(2)); + assert_eq!(N::from_name("sqrt").map(N::arity), Some(1)); + assert_eq!(N::from_name("setgray"), None); + assert_eq!(N::Add.apply_binary(2.0, 3.0), Some(5.0)); + assert_eq!(N::Subtract.apply_binary(2.0, 3.0), Some(-1.0)); + assert_eq!(N::Multiply.apply_binary(2.0, 3.0), Some(6.0)); + assert_eq!(N::Divide.apply_binary(7.0, 2.0), Some(3.5)); + assert_eq!(N::Divide.apply_binary(7.0, 0.0), Some(0.0)); + assert_eq!(N::IntegerDivide.apply_binary(-7.0, 2.0), Some(-3.0)); + assert_eq!(N::IntegerDivide.apply_binary(7.0, 0.0), Some(0.0)); + assert_eq!(N::IntegerDivide.apply_binary(i64::MIN as f64, -1.0), Some(0.0)); + assert_eq!(N::Modulo.apply_binary(-7.0, 2.0), Some(-1.0)); + assert_eq!(N::Modulo.apply_binary(7.0, 0.0), Some(0.0)); + assert_eq!(N::Atan.apply_binary(0.0, -1.0), Some(180.0)); + assert_eq!(N::Exponentiate.apply_binary(-2.0, 3.0), Some(-8.0)); + assert_eq!(N::Exponentiate.apply_binary(-2.0, 0.5), Some(0.0)); + assert_eq!(N::Maximum.apply_binary(2.0, 3.0), Some(3.0)); + assert_eq!(N::Minimum.apply_binary(2.0, 3.0), Some(2.0)); + assert_eq!(N::Negate.apply_binary(1.0, 2.0), None); + } + + #[test] + fn unary_numeric_operator_results_match_the_shared_vm_contract() { + assert_eq!(N::Negate.apply_unary(2.0), Some(-2.0)); + assert_eq!(N::Absolute.apply_unary(-2.0), Some(2.0)); + assert_eq!(N::SquareRoot.apply_unary(-2.0), Some(0.0)); + assert_eq!(N::Sine.apply_unary(30.0).map(|v| (v - 0.5).abs() < 1e-12), Some(true)); + assert_eq!(N::Cosine.apply_unary(60.0).map(|v| (v - 0.5).abs() < 1e-12), Some(true)); + assert_eq!(N::NaturalLog.apply_unary(-1.0), Some(0.0)); + assert_eq!(N::CommonLog.apply_unary(100.0), Some(2.0)); + assert_eq!(N::ToInteger.apply_unary(-2.9), Some(-2.0)); + assert_eq!(N::ToReal.apply_unary(2.5), Some(2.5)); + assert_eq!(N::Round.apply_unary(-1.5), Some(-1.0)); + assert_eq!(N::Floor.apply_unary(1.9), Some(1.0)); + assert_eq!(N::Ceiling.apply_unary(1.1), Some(2.0)); + assert_eq!(N::Truncate.apply_unary(-2.9), Some(-2.0)); + assert_eq!(N::Add.apply_unary(1.0), None); + } +} + +#[cfg(test)] +mod tests { + use super::{Scanner, Token}; + + fn scan(source: &[u8]) -> Vec { + let mut scanner = Scanner::new(source, 0); + let mut tokens = Vec::new(); + while let Some(token) = scanner.token() { + tokens.push(token); + } + tokens + } + + #[test] + fn scans_names_numbers_strings_and_procedures() { + assert_eq!( + scan(b"/x //y 16#ff { 2 (a\\n(b)) }"), + vec![ + Token::LiteralName("x".into()), + Token::Name("y".into()), + Token::Number(255.0), + Token::Procedure(vec![Token::Number(2.0), Token::String(b"a\n(b)".to_vec())]), + ] + ); + } + + #[test] + fn skips_dsc_binary_data_and_preserves_inline_ascii85() { + assert_eq!(scan(b"%%BeginBinary: 4\nxxxx%%EndBinary\n<~z~>"), vec![Token::Ascii85(b"<~z~>".to_vec())]); + } + + #[test] + fn malformed_delimiters_terminate_without_advancing_past_source() { + let mut scanner = Scanner::new(b" impl Iterator { + text.split(['\n', '\r']).flat_map(|line| { + line.split('%').next().unwrap_or("").split_whitespace().flat_map(|word| word.split_inclusive(['{', '}'])).flat_map(|piece| { + let (body, brace) = match piece.char_indices().last() { + Some((at, '{' | '}')) => (piece.get(..at).unwrap_or(""), piece.get(at..)), + _ => (piece, None), + }; + [Some(body).filter(|body| !body.is_empty()), brace].into_iter().flatten() + }) + }) +} + +/// Read solid rectangle fills and their sRGB colours from the restricted PostScript subset used +/// by simple EPS previews. None means the program uses an unsupported operator or exceeds a +/// fixed work bound. The function never executes PostScript. +pub fn simple_rectangles(ps: &[u8]) -> Option> { + let text = std::str::from_utf8(ps).ok()?; + let mut aliases: HashMap = HashMap::new(); + let mut stack: Vec = Vec::new(); + let mut color = [0u8, 0, 0, 255]; + let mut saved: Vec<[u8; 4]> = Vec::new(); + let mut rectangles = Vec::new(); + let mut input = tokens(text); + let mut count = 0usize; + while let Some(token) = input.next() { + count += 1; + if count > MAX_SIMPLE_RECTANGLE_TOKENS { + return None; + } + if let Ok(value) = token.parse::() { + if !value.is_finite() { + return None; + } + stack.push(value); + continue; + } + if let Some(name) = token.strip_prefix('/') { + let mut body: Vec<&str> = Vec::new(); + while body.len() < 5 { + count += 1; + if count > MAX_SIMPLE_RECTANGLE_TOKENS { + return None; + } + match input.next() { + Some("def") | None => break, + Some(token) => body.push(token), + } + } + let op = match body.as_slice() { + ["{", op, "}"] | ["{", op, "}", _] => (*op).to_string(), + _ => return None, + }; + aliases.insert(name.to_string(), op); + continue; + } + let op = aliases.get(token).map_or(token, String::as_str); + let mut pop = |n: usize| -> Option> { (stack.len() >= n).then(|| stack.split_off(stack.len() - n)) }; + match op { + "setcmykcolor" => { + let values = pop(4)?; + let black = values[3].clamp(0.0, 1.0); + let channel = |value: f64| ((1.0 - value.clamp(0.0, 1.0)) * (1.0 - black) * 255.0).round() as u8; + color = [channel(values[0]), channel(values[1]), channel(values[2]), 255]; + } + "setrgbcolor" => { + let values = pop(3)?; + let channel = |value: f64| (value.clamp(0.0, 1.0) * 255.0).round() as u8; + color = [channel(values[0]), channel(values[1]), channel(values[2]), 255]; + } + "setgray" => { + let values = pop(1)?; + let gray = (values[0].clamp(0.0, 1.0) * 255.0).round() as u8; + color = [gray, gray, gray, 255]; + } + "rectfill" => { + let values = pop(4)?; + if rectangles.len() >= MAX_SIMPLE_RECTANGLES { + return None; + } + let x1 = values[0] + values[2]; + let y1 = values[1] + values[3]; + if !x1.is_finite() || !y1.is_finite() { + return None; + } + rectangles.push(Rectangle { + x: values[0].min(x1), + y: values[1].min(y1), + width: (x1 - values[0]).abs(), + height: (y1 - values[1]).abs(), + color, + }); + } + "gsave" | "save" => { + if saved.len() >= MAX_SIMPLE_RECTANGLE_SAVE_DEPTH { + return None; + } + saved.push(color); + } + "grestore" | "restore" => { + if let Some(previous) = saved.pop() { + color = previous; + } + } + "pop" => { + pop(1)?; + } + "bind" | "newpath" | "showpage" | "end" | "begin" => {} + _ => return None, + } + } + (!rectangles.is_empty()).then_some(rectangles) +} diff --git a/crates/eps/src/tests.rs b/crates/eps/src/tests.rs index 7ac4c84ca..7a26ed09a 100644 --- a/crates/eps/src/tests.rs +++ b/crates/eps/src/tests.rs @@ -29,6 +29,24 @@ fn sections_of_plain_and_binary_files() { assert_eq!(sections(&file), Ok(Sections { postscript: EPS, tiff: Some(b"TIFF"), metafile: Some(b"WMF") })); } +#[test] +fn postscript_section_survives_a_damaged_preview_range() { + let ps = b"%!PS-Adobe-3.0 EPSF-3.0\n%%BoundingBox: 0 0 10 10\n"; + let mut bytes = wrap(ps, Some(b"tiff"), None).unwrap(); + bytes[20..24].copy_from_slice(&u32::MAX.to_le_bytes()); + bytes[24..28].copy_from_slice(&1u32.to_le_bytes()); + assert_eq!(postscript_section(&bytes), Ok(ps.as_slice())); + assert_eq!(sections(&bytes), Err(Error::OutsideFile)); +} + +#[test] +fn extended_comment_scan_keeps_its_upper_bound() { + let mut ps = b"%!PS-Adobe-3.0 EPSF-3.0\n".to_vec(); + ps.resize(ps.len() + MAX_EXTENDED_HEADER + 1, b' '); + ps.extend_from_slice(b"\n%%BoundingBox: 0 0 20 10\n%%EndComments\n"); + assert_eq!(Comments::read_with_header_limit(&ps, usize::MAX).page_box(), None); +} + #[test] fn a_header_that_points_outside_the_file_is_refused() { let file = dos(EPS, b"WMF", b"TIFF"); diff --git a/crates/exr/src/decode.rs b/crates/exr/src/decode.rs index 2b47f42eb..d8cb32827 100644 --- a/crates/exr/src/decode.rs +++ b/crates/exr/src/decode.rs @@ -29,7 +29,7 @@ use exr::prelude::{ }; pub use half::f16; -use crate::channels::color_channels; +use crate::channels::{color_channels, part_channel_name}; /// Why a file could not be read or written. #[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)] @@ -402,6 +402,233 @@ impl Part { } } +/// A channel placed in its part's display window. Samples are `f32`; samples outside the data +/// window are filled with the caller's per-channel value. +#[derive(Debug, Clone, PartialEq)] +pub struct DisplayChannel { + pub name: String, + pub samples: Vec, +} + +/// A decoded part whose channels are placed in the display window. +#[derive(Debug, Clone, PartialEq)] +pub struct DisplayPart { + pub info: PartInfo, + pub channels: Vec, +} + +/// Decodes the kept flat channels of one part and places them in its display window. +/// +/// Channels are decoded in their stored sample type, converted to `f32`, and clipped or padded +/// according to their data and display windows. `keep` and `fill` receive the part-qualified name +/// from [`part_channel_name`], which is also returned in [`DisplayChannel::name`]. Both decoded +/// block data and display-window buffers are bounded by `limits` before allocation. +pub fn decode_display_part( + bytes: &[u8], + index: usize, + keep: &dyn Fn(&str) -> bool, + fill: &dyn Fn(&str) -> f32, + limits: &Limits, +) -> Result { + // Only the caller-selected display planes are allocated here. Keep the shared header and + // expansion-ratio checks, but do not charge ignored channel data against the output budget. + let mut header_limits = *limits; + header_limits.max_width = u32::MAX; + header_limits.max_height = u32::MAX; + header_limits.max_pixels = u64::MAX; + header_limits.max_bytes = u64::MAX; + let meta = read_meta(bytes, &header_limits)?; + let info = meta.headers.get(index).map(|header| part_info(index, header)).ok_or_else(|| malformed(format!("part {index}")))?; + if info.deep { + return Err(Error::Unsupported("deep data (use the deep reader)".into())); + } + let display = info.display_window; + let pixels = u64::from(display.width) + .checked_mul(u64::from(display.height)) + .ok_or_else(|| Error::LimitExceeded("display window dimensions overflow".into()))?; + if pixels > limits.max_pixels { + return Err(Error::LimitExceeded(format!("display window of {pixels} pixels exceeds {}", limits.max_pixels))); + } + let kept = info.channels.iter().filter(|channel| keep(&part_channel_name(info.name.as_deref(), &channel.name))).count(); + let bytes_needed = pixels + .checked_mul(4) + .and_then(|bytes_per_channel| bytes_per_channel.checked_mul(kept as u64)) + .ok_or_else(|| Error::LimitExceeded("display window byte size overflow".into()))?; + if bytes_needed > limits.max_bytes || usize::try_from(bytes_needed).is_err() { + return Err(Error::LimitExceeded(format!("{bytes_needed} display-window bytes exceed {}", limits.max_bytes))); + } + let block_limit = std::cell::Cell::new(None); + let select = DisplaySelect { index, keep, fill, limits, seen: std::cell::Cell::new(0), block_limit: &block_limit }; + let image = ReadImage::new(ReadAllLayers { read_channels: select }, |_| {}) + .from_buffered(Cursor::new(bytes)) + .map_err(|error| block_limit.take().map(Error::LimitExceeded).unwrap_or_else(|| map_exr(error)))?; + let channels = image + .layer_data + .into_iter() + .nth(index) + .ok_or_else(|| malformed("the part was not read"))? + .channel_data + .into_iter() + .flatten() + .map(|(name, samples)| DisplayChannel { name, samples }) + .collect(); + Ok(DisplayPart { info, channels }) +} + +/// Filtered channel reader that writes each decoded block directly into a bounded display plane. +struct DisplaySelect<'a> { + index: usize, + keep: &'a dyn Fn(&str) -> bool, + fill: &'a dyn Fn(&str) -> f32, + limits: &'a Limits, + seen: std::cell::Cell, + block_limit: &'a std::cell::Cell>, +} + +struct DisplaySelectReader { + planes: Vec)>>, + offset: (i64, i64), + layer: (usize, usize), + display: (usize, usize), + f16s: Vec, + f32s: Vec, + u32s: Vec, +} + +impl<'s> ReadChannels<'s> for DisplaySelect<'_> { + type Reader = DisplaySelectReader; + + fn create_channels_reader(&'s self, header: &Header) -> exr::error::Result { + let part = self.seen.get(); + self.seen.set(part.saturating_add(1)); + let (layer_w, layer_h) = (header.layer_size.0, header.layer_size.1); + let display_window = header.shared_attributes.display_window; + let (display_w, display_h) = (display_window.size.0, display_window.size.1); + if part != self.index { + return Ok(DisplaySelectReader { + planes: header.channels.list.iter().map(|_| None).collect(), + offset: (0, 0), + layer: (layer_w, layer_h), + display: (display_w, display_h), + f16s: vec![], + f32s: vec![], + u32s: vec![], + }); + } + // The EXR crate decompresses each accepted block, with every channel, before `read_block` + // can discard the ignored ones, so bound that all-channel block separately from the + // selected output planes. Blocks are decompressed one at a time, so the bound is the + // largest block (a tile within the data window, or a scan-line group across it), not the + // whole part: a part whose full channel set is larger than `max_bytes` still decodes when + // the kept planes and each block fit. + let block_pixels = match header.blocks { + exr::meta::BlockDescription::Tiles(tiles) => tiles.tile_size.0.min(layer_w).checked_mul(tiles.tile_size.1.min(layer_h)), + // a part shorter than its compression's scan-line group has only its own rows + exr::meta::BlockDescription::ScanLines => header.compression.scan_lines_per_block().min(layer_h).checked_mul(layer_w), + }; + let Some(block_bytes) = block_pixels.and_then(|pixels| pixels.checked_mul(header.channels.bytes_per_pixel.max(1))) else { + self.block_limit.set(Some(format!("data-window block size exceeds {}", self.limits.max_bytes))); + return Err(exr::error::Error::Invalid("data-window block size overflow".into())); + }; + if u64::try_from(block_bytes).map_or(true, |bytes| bytes > self.limits.max_bytes) || block_bytes > isize::MAX as usize { + self.block_limit.set(Some(format!("{block_bytes} bytes in one data-window block exceed {}", self.limits.max_bytes))); + return Err(exr::error::Error::Invalid("data-window blocks exceed limit".into())); + } + let pixels = display_w.checked_mul(display_h).filter(|n| *n > 0).ok_or(exr::error::Error::Invalid("display window".into()))?; + let prefix = header.own_attributes.layer_name.as_ref().map(|name| name.to_string()).filter(|name| !name.is_empty()); + let kept = + header.channels.list.iter().filter(|channel| (self.keep)(&part_channel_name(prefix.as_deref(), &channel.name.to_string()))).count(); + let bytes = pixels + .checked_mul(4) + .and_then(|n| n.checked_mul(kept)) + .ok_or_else(|| exr::error::Error::Invalid("display planes exceed limit".into()))?; + if pixels as u64 > self.limits.max_pixels || bytes as u64 > self.limits.max_bytes || bytes > isize::MAX as usize { + return Err(exr::error::Error::Invalid("display planes exceed limit".into())); + } + let mut planes = Vec::with_capacity(header.channels.list.len()); + for channel in header.channels.list.iter() { + let name = part_channel_name(prefix.as_deref(), &channel.name.to_string()); + let plane = if (self.keep)(&name) { + if channel.sampling.0 != 1 || channel.sampling.1 != 1 { + return Err(exr::error::Error::NotSupported("subsampled channels".into())); + } + Some((name.clone(), vec![(self.fill)(&name); pixels])) + } else { + None + }; + planes.push(plane); + } + let position = header.own_attributes.layer_position; + Ok(DisplaySelectReader { + planes, + offset: (i64::from(position.0) - i64::from(display_window.position.0), i64::from(position.1) - i64::from(display_window.position.1)), + layer: (layer_w, layer_h), + display: (display_w, display_h), + f16s: vec![], + f32s: vec![], + u32s: vec![], + }) + } +} + +impl ChannelsReader for DisplaySelectReader { + type Channels = Vec)>>; + + fn filter_block(&self, tile: TileCoordinates) -> bool { + tile.is_largest_resolution_level() && self.planes.iter().any(Option::is_some) + } + + fn read_block(&mut self, header: &Header, block: UncompressedBlock) -> exr::error::UnitResult { + let ((layer_w, layer_h), (display_w, display_h)) = (self.layer, self.display); + for line in block.lines(&header.channels) { + let Some(Some((_, dst_plane))) = self.planes.get_mut(line.location.channel) else { continue }; + let (x, y, count) = (line.location.position.0, line.location.position.1, line.location.sample_count); + if y >= layer_h || x.checked_add(count).is_none_or(|end| end > layer_w) { + continue; + } + let Some(ty) = usize::try_from(y as i64 + self.offset.1).ok().filter(|ty| *ty < display_h) else { continue }; + let tx0 = x as i64 + self.offset.0; + let skip = usize::try_from(-tx0).unwrap_or(0).min(count); + let tx = usize::try_from(tx0.max(0)).unwrap_or(usize::MAX); + let take = (count - skip).min(display_w.saturating_sub(tx)); + if take == 0 { + continue; + } + let start = ty.checked_mul(display_w).and_then(|row| row.checked_add(tx)); + let Some(dst) = start.and_then(|s| dst_plane.get_mut(s..s.checked_add(take)?)) else { continue }; + match header.channels.list.get(line.location.channel).map(|c| c.sample_type) { + Some(ExrSample::F16) => { + self.f16s.resize(count, f16::ZERO); + line.read_samples_into_slice(&mut self.f16s)?; + for (d, s) in dst.iter_mut().zip(self.f16s.iter().skip(skip)) { + *d = s.to_f32(); + } + } + Some(ExrSample::F32) => { + self.f32s.resize(count, 0.0); + line.read_samples_into_slice(&mut self.f32s)?; + for (d, s) in dst.iter_mut().zip(self.f32s.iter().skip(skip)) { + *d = *s; + } + } + Some(ExrSample::U32) => { + self.u32s.resize(count, 0); + line.read_samples_into_slice(&mut self.u32s)?; + for (d, s) in dst.iter_mut().zip(self.u32s.iter().skip(skip)) { + *d = *s as f32; + } + } + None => continue, + } + } + Ok(()) + } + + fn into_channels(self) -> Self::Channels { + self.planes + } +} + /// Places a data-window plane in a display window: samples outside the display window are cut, /// the display window's pixels the data window does not cover are `fill`. `None` when the plane /// is not `data.width Ă— data.height` samples or the display window has more than `max_pixels` diff --git a/crates/exr/src/decode_tests.rs b/crates/exr/src/decode_tests.rs index 69d8d07ac..66a64cfb4 100644 --- a/crates/exr/src/decode_tests.rs +++ b/crates/exr/src/decode_tests.rs @@ -118,6 +118,66 @@ fn display_window_placement() { assert_eq!(to_display_window(&[1.0, 2.0, 3.0, 4.0], far, display, 0.0, 100).unwrap(), [0.0; 6]); } +#[test] +fn decoded_display_part_places_selected_channels_and_checks_total_budget() { + let bytes = file(vec![layer(None, 2, 2, &["R", "G"], false, exr::prelude::Compression::ZIP1)], (3, 2)); + let part = decode_display_part(&bytes, 0, &|name| name == "R", &|_| 9.0, &Limits::default()).unwrap(); + assert_eq!(part.channels.len(), 1); + assert_eq!(part.channels[0].name, "R"); + assert_eq!(part.channels[0].samples, [0.0, 1.0, 9.0, 2.0, 3.0, 9.0]); + + let limits = Limits { max_bytes: 6 * 4 - 1, ..Limits::default() }; + assert!(matches!(decode_display_part(&bytes, 0, &all, &|_| 0.0, &limits), Err(Error::LimitExceeded(_)))); + + // One selected display plane fits exactly. Both data channels together (2x2x2 f32, 32 bytes) + // don't, but they are decompressed one scan-line block (2 pixels x 2 channels) at a time. + let block_limits = Limits { max_bytes: 6 * 4, ..Limits::default() }; + let part = decode_display_part(&bytes, 0, &|name| name == "R", &|_| 9.0, &block_limits).unwrap(); + assert_eq!(part.channels[0].samples, [0.0, 1.0, 9.0, 2.0, 3.0, 9.0]); +} + +#[test] +fn decoded_display_part_bounds_each_all_channel_block() { + // The kept R plane is 4 pixels of f32 (16 bytes), but each scan-line block carries all four + // channels (64 bytes) before the ignored ones are discarded. + let bytes = file(vec![layer(None, 4, 1, &["R", "G", "B", "A"], false, exr::prelude::Compression::ZIP1)], (4, 1)); + let r = |name: &str| name == "R"; + assert!(matches!(decode_display_part(&bytes, 0, &r, &|_| 0.0, &Limits { max_bytes: 16, ..Limits::default() }), Err(Error::LimitExceeded(_)))); + assert_eq!(decode_display_part(&bytes, 0, &r, &|_| 0.0, &Limits { max_bytes: 64, ..Limits::default() }).unwrap().channels.len(), 1); + // ZIP groups 16 scan lines, but a one-row part's block is that row: still 64 bytes, not 1024 + let zip16 = file(vec![layer(None, 4, 1, &["R", "G", "B", "A"], false, exr::prelude::Compression::ZIP16)], (4, 1)); + assert!(matches!(decode_display_part(&zip16, 0, &r, &|_| 0.0, &Limits { max_bytes: 63, ..Limits::default() }), Err(Error::LimitExceeded(_)))); + assert_eq!(decode_display_part(&zip16, 0, &r, &|_| 0.0, &Limits { max_bytes: 64, ..Limits::default() }).unwrap().channels.len(), 1); + // A tile is the block of a tiled part: a 4x4 tile of four f32 channels is 256 bytes, while + // the whole 8x8 part is 1024 and the kept R plane of the 2x2 display window only 16. + let tiled = file(vec![layer(None, 8, 8, &["R", "G", "B", "A"], true, exr::prelude::Compression::ZIP1)], (2, 2)); + assert!(matches!(decode_display_part(&tiled, 0, &r, &|_| 0.0, &Limits { max_bytes: 255, ..Limits::default() }), Err(Error::LimitExceeded(_)))); + assert_eq!(decode_display_part(&tiled, 0, &r, &|_| 0.0, &Limits { max_bytes: 256, ..Limits::default() }).unwrap().channels[0].samples.len(), 4); +} + +#[test] +fn decoded_display_part_clips_a_wide_data_window_to_the_small_display() { + let width = (Limits::default().max_width as usize) + 1; + let bytes = file(vec![layer(None, width, 1, &["R"], false, exr::prelude::Compression::ZIP1)], (1, 1)); + let part = decode_display_part(&bytes, 0, &all, &|_| 9.0, &Limits::default()).unwrap(); + assert_eq!(part.channels.len(), 1); + assert_eq!(part.channels[0].samples, [0.0]); +} + +#[test] +fn decoded_display_part_rejects_kept_subsampled_channels() { + let mut bytes = file(vec![layer(None, 4, 1, &["R"], false, exr::prelude::Compression::Uncompressed)], (4, 1)); + // exr's writer doesn't emit subsampling yet, so patch the standard chlist sampling field in a + // valid uncompressed file. The reader must reject it before looking at the block payload. + let tag = b"channels\0chlist\0"; + let attr = bytes.windows(tag.len()).position(|window| window == tag).unwrap(); + let payload = attr + tag.len() + 4; + let name_end = payload + bytes[payload..].iter().position(|byte| *byte == 0).unwrap(); + let x_sampling = name_end + 1 + 4 + 1 + 3; + bytes[x_sampling..x_sampling + 4].copy_from_slice(&2u32.to_le_bytes()); + assert!(matches!(decode_display_part(&bytes, 0, &|name| name == "R", &|_| 0.0, &Limits::default()), Err(Error::Unsupported(_)))); +} + /// A 2x2 single-part file whose layer carries a 2x1 preview. fn with_preview() -> Vec { let mut attrs = LayerAttributes::named("rgb"); diff --git a/crates/exr/src/lib.rs b/crates/exr/src/lib.rs index baaf3d3f1..78afd427e 100644 --- a/crates/exr/src/lib.rs +++ b/crates/exr/src/lib.rs @@ -18,6 +18,11 @@ //! - [`stream`]: one flat scan-line part read block by block, for pictures larger than memory. //! - [`cryptomatte_samples`]: a file's Cryptomatte layers and their per-pixel (ID, coverage) ranks. +/// Scanline OpenEXR writer using a caller-supplied linearization table. +pub mod rgba8; +/// RFC 1950/1951 compressor for ZIP-compressed EXR chunks. +pub mod zlib; + pub mod channels; pub mod cryptomatte; #[cfg(feature = "decode")] diff --git a/crates/exr/src/rgba8.rs b/crates/exr/src/rgba8.rs new file mode 100644 index 000000000..6dc4bde4c --- /dev/null +++ b/crates/exr/src/rgba8.rs @@ -0,0 +1,235 @@ +//! Single-part scanline OpenEXR output from RGBA8 pixels. + +/// Scanline chunk compression. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum Compression { + /// Uncompressed. + None, + /// OpenEXR RLE. + Rle, + /// zlib per scanline. + Zips, + /// zlib per 16 scanlines. + Zip, +} +impl Compression { + fn code(self) -> u8 { + match self { + Self::None => 0, + Self::Rle => 1, + Self::Zips => 2, + Self::Zip => 3, + } + } + fn lines(self) -> usize { + if self == Self::Zip { 16 } else { 1 } + } +} +/// Why a frame cannot be encoded. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub enum Error { + /// Empty, overflowing or mismatched RGBA dimensions. + FrameSize, + /// Dimensions exceed the OpenEXR signed-size fields. + TooLarge, +} +fn checked_len(rgba: &[u8], w: u32, h: u32) -> Result { + let n = (w as usize).checked_mul(h as usize).and_then(|n| n.checked_mul(4)).ok_or(Error::FrameSize)?; + if w == 0 || h == 0 || rgba.len() != n { + return Err(Error::FrameSize); + } + Ok(n) +} + +/// IEEE 754 binary16 bits of `v`, rounded to nearest even (overflow → infinity, NaN → quiet NaN). +pub fn half(v: f32) -> u16 { + let b = v.to_bits(); + let sign = ((b >> 16) & 0x8000) as u16; + let exp = ((b >> 23) & 0xFF) as i32; + let man = b & 0x7F_FFFF; + if exp == 0xFF { + return sign | 0x7C00 | if man != 0 { 0x200 } else { 0 }; + } + // rounds `m >> shift` to nearest, ties to even (`shift` is 1..=24) + let round = |m: u32, shift: u32| { + let (q, rem, mid) = (m >> shift, m & ((1 << shift) - 1), 1 << (shift - 1)); + if rem > mid || (rem == mid && q & 1 == 1) { q + 1 } else { q } + }; + let e = exp - 127 + 15; + if e >= 0x1F { + sign | 0x7C00 + } else if e <= 0 { + if e < -10 { + return sign; + } + // subnormal half: the 24-bit significand scaled by 2^(e - 14); rounding up may carry + // into the smallest normal, which is the right encoding + sign | round(man | 0x80_0000, (14 - e) as u32) as u16 + } else { + // a carry out of the mantissa bumps the exponent (up to infinity), as it should + sign | round(((e as u32) << 23) | man, 13) as u16 + } +} + +/// An OpenEXR file: half-float scene-linear Rec.709 RGB, or RGBA with `alpha` (premultiplied, as +/// EXR defines alpha; the frame's straight alpha is multiplied in after linearisation). +pub fn write_rgba8(rgba: &[u8], w: u32, h: u32, alpha: bool, compression: Compression, linear: &[f32; 256]) -> Result, Error> { + let n = checked_len(rgba, w, h)?; + let too_large = || Error::TooLarge; + let (Ok(x_max), Ok(y_max)) = (i32::try_from(w - 1), i32::try_from(h - 1)) else { + return Err(too_large()); + }; + // channels in the alphabetical order the file layout requires + let names: &[u8] = if alpha { b"ABGR" } else { b"BGR" }; + let line = (w as usize).checked_mul(names.len() * 2).ok_or(Error::FrameSize)?; + // a chunk's byte count is a signed 32-bit field + let lines = compression.lines(); + if line.checked_mul(lines).and_then(|c| i32::try_from(c).ok()).is_none() { + return Err(too_large()); + } + + let mut hd = Vec::with_capacity(512); + hd.extend_from_slice(&[0x76, 0x2F, 0x31, 0x01]); // magic number 20000630 + hd.extend_from_slice(&[2, 0, 0, 0]); // version 2, single-part scan-line file, short names + let attr = |hd: &mut Vec, name: &str, ty: &str, value: &[u8]| { + for s in [name, ty] { + hd.extend_from_slice(s.as_bytes()); + hd.push(0); + } + hd.extend_from_slice(&(value.len() as i32).to_le_bytes()); + hd.extend_from_slice(value); + }; + let le = |v: &[f32]| v.iter().flat_map(|f| f.to_le_bytes()).collect::>(); + let mut channels = Vec::new(); + for &c in names { + channels.extend_from_slice(&[c, 0]); + channels.extend_from_slice(&1_i32.to_le_bytes()); // pixel type HALF + channels.extend_from_slice(&[0, 0, 0, 0]); // pLinear 0, reserved + channels.extend_from_slice(&1_i32.to_le_bytes()); // x sampling + channels.extend_from_slice(&1_i32.to_le_bytes()); // y sampling + } + channels.push(0); + let window: Vec = [0, 0, x_max, y_max].iter().flat_map(|v| v.to_le_bytes()).collect(); + attr(&mut hd, "channels", "chlist", &channels); + attr(&mut hd, "chromaticities", "chromaticities", &le(&[0.64, 0.33, 0.30, 0.60, 0.15, 0.06, 0.3127, 0.3290])); + attr(&mut hd, "compression", "compression", &[compression.code()]); + attr(&mut hd, "dataWindow", "box2i", &window); + attr(&mut hd, "displayWindow", "box2i", &window); + attr(&mut hd, "lineOrder", "lineOrder", &[0]); // INCREASING_Y + attr(&mut hd, "pixelAspectRatio", "float", &le(&[1.0])); + attr(&mut hd, "screenWindowCenter", "v2f", &le(&[0.0, 0.0])); + attr(&mut hd, "screenWindowWidth", "float", &le(&[1.0])); + hd.push(0); // end of header + + // offset table (one entry per chunk), then the chunks: first y, data size, data + let chunks = (h as usize).div_ceil(lines); + let table = chunks.checked_mul(8); + let body = (h as usize).checked_mul(line + 8); + let size = table.zip(body).and_then(|(t, b)| hd.len().checked_add(t)?.checked_add(b)).ok_or(Error::FrameSize)?; + // the uncompressed size: compressed chunks only shrink it + let mut out = Vec::with_capacity(size); + out.extend_from_slice(&hd); + let table_at = out.len(); + out.resize(table_at + chunks * 8, 0); + let lin = linear; + let opaque: [u16; 256] = std::array::from_fn(|i| half(lin[i])); + let row = w as usize * 4; + let wn = w as usize; + let mut px = vec![0_u16; wn * names.len()]; + let mut raw = Vec::with_capacity(line * lines); + // the deflate hash tables, shared by the frame's chunks instead of zeroed for each one + let mut tables = crate::zlib::Tables::default(); + for (k, block) in rgba[..n].chunks(row * lines).enumerate() { + raw.clear(); + for src in block.chunks_exact(row) { + // planar per line: every pixel of the first channel, then the next channel … + for (x, &[r, g, b, a]) in src.as_chunks::<4>().0.iter().enumerate() { + let rgb = [b, g, r]; + if alpha { + let k = f32::from(a) / 255.0; + px[x] = half(k); + for (c, &v) in rgb.iter().enumerate() { + px[(c + 1) * wn + x] = if a == 255 { opaque[v as usize] } else { half(lin[v as usize] * k) }; + } + } else { + for (c, &v) in rgb.iter().enumerate() { + px[c * wn + x] = opaque[v as usize]; + } + } + } + raw.extend(px.iter().flat_map(|v| v.to_le_bytes())); + } + let packed = match compression { + Compression::None => None, + Compression::Rle => Some(rle(&predict(&raw))), + Compression::Zips | Compression::Zip => Some(crate::zlib::zlib(&predict(&raw), &mut tables)), + }; + // a chunk that does not shrink is stored as it is (readers compare the sizes) + let data = packed.as_deref().filter(|p| p.len() < raw.len()).unwrap_or(&raw); + let offset = (out.len() as u64).to_le_bytes(); + out[table_at + k * 8..table_at + k * 8 + 8].copy_from_slice(&offset); + out.extend_from_slice(&((k * lines) as i32).to_le_bytes()); + out.extend_from_slice(&(data.len() as i32).to_le_bytes()); + out.extend_from_slice(data); + } + Ok(out) +} + +/// The OpenEXR RLE / ZIP predictor: bytes split into even and odd halves, then each byte replaced +/// by its difference from the previous one, plus 128. +fn predict(raw: &[u8]) -> Vec { + let half = raw.len().div_ceil(2); + let mut t = vec![0_u8; raw.len()]; + for (i, &b) in raw.iter().enumerate() { + t[if i % 2 == 0 { i / 2 } else { half + i / 2 }] = b; + } + let mut p = t.first().copied().unwrap_or(0); + for v in t.iter_mut().skip(1) { + let d = v.wrapping_sub(p).wrapping_add(128); + p = *v; + *v = d; + } + t +} + +/// OpenEXR run-length coding: a signed count byte, then either `count + 1` copies of one byte +/// (count ≥ 0, runs of 3 to 128) or `-count` literal bytes (1 to 127). +pub fn rle(src: &[u8]) -> Vec { + let mut out = Vec::with_capacity(src.len() + src.len() / 127 + 1); + let mut i = 0; + while i < src.len() { + let b = src[i]; + let run = src[i..].iter().take(128).take_while(|&&x| x == b).count(); + if run >= 3 { + out.extend_from_slice(&[(run - 1) as u8, b]); + i += run; + continue; + } + // literals up to the next run of three (or 127 bytes) + let s = i; + while i < src.len() && i - s < 127 { + if src.get(i..i + 3).is_some_and(|t| t[0] == t[1] && t[1] == t[2]) { + break; + } + i += 1; + } + out.push(((i - s) as u8).wrapping_neg()); + out.extend_from_slice(&src[s..i]); + } + out +} + +#[cfg(test)] +mod tests { + use super::*; + #[test] + fn writes_all_scanline_compressions() { + let linear = std::array::from_fn(|i| i as f32 / 255.0); + for c in [Compression::None, Compression::Rle, Compression::Zips, Compression::Zip] { + let bytes = write_rgba8(&[128; 16 * 16 * 4], 16, 16, true, c, &linear).unwrap(); + assert_eq!(&bytes[..4], &[0x76, 0x2f, 0x31, 1]); + assert!(bytes.len() > 100); + } + assert_eq!(write_rgba8(&[], 1, 1, false, Compression::None, &linear), Err(Error::FrameSize)); + } +} diff --git a/crates/exr/src/zlib.rs b/crates/exr/src/zlib.rs new file mode 100644 index 000000000..a38bd0b46 --- /dev/null +++ b/crates/exr/src/zlib.rs @@ -0,0 +1,795 @@ +//! A small DEFLATE encoder (RFC 1951) in a zlib wrapper (RFC 1950), written from the RFCs for the +//! ZIP-compressed OpenEXR chunks. +//! +//! Greedy LZ77 over a 32 KB window (hash chains of bounded length), then each block of tokens is +//! written as whichever of stored, fixed-Huffman or dynamic-Huffman is smallest. Huffman code +//! lengths are limited (15 bits, 7 for the code-length code) by halving the symbol frequencies +//! until the tree fits. Every code has at least two symbols, so all codes are complete. The +//! output favours simplicity and correctness over ratio; worst-case expansion is the stored-block +//! overhead (5 bytes per 64 KB). + +use std::cmp::Reverse; +use std::collections::BinaryHeap; + +const WINDOW: usize = 32 * 1024; +const MIN_MATCH: usize = 3; +const MAX_MATCH: usize = 258; +/// Hash chain candidates tried per position. +const MAX_CHAIN: usize = 48; +/// Tokens per block (a new set of Huffman codes after this many). +const BLOCK_TOKENS: usize = 1 << 15; +const MAX_STORED: usize = 65_535; + +const LEN_BASE: [u16; 29] = [3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258]; +const LEN_EXTRA: [u8; 29] = [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0]; +const DIST_BASE: [u16; 30] = + [1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 8193, 12289, 16385, 24577]; +const DIST_EXTRA: [u8; 30] = [0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13]; +/// Order in which the code-length code's lengths are sent. +const CL_ORDER: [usize; 19] = [16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15]; + +/// `data` compressed as a zlib stream (CMF 0x78: deflate, 32 KB window). `tables` are the match +/// finder's hash tables, kept by the caller across the chunks of a picture; the output does not +/// depend on what they held before. +pub fn zlib(data: &[u8], tables: &mut Tables) -> Vec { + let mut w = Bits { out: Vec::with_capacity(data.len() / 2 + 64), acc: 0, n: 0 }; + w.out.extend_from_slice(&[0x78, 0x9C]); // (0x78 << 8 | 0x9C) % 31 == 0 + deflate(data, &mut w, tables); + w.align(); + w.out.extend_from_slice(&adler32(data).to_be_bytes()); + w.out +} + +/// Adler-32 (RFC 1950 §9); 5552 bytes is the longest run that cannot overflow the sums. +fn adler32(data: &[u8]) -> u32 { + let (mut a, mut b) = (1_u32, 0_u32); + for c in data.chunks(5552) { + for &x in c { + a += u32::from(x); + b += a; + } + a %= 65_521; + b %= 65_521; + } + (b << 16) | a +} + +/// LSB-first bit writer. +struct Bits { + out: Vec, + acc: u64, + n: u32, +} + +impl Bits { + /// The low `n` bits of `v` (n ≤ 32). + fn put(&mut self, v: u32, n: u32) { + let mask = (1_u64 << n) - 1; + self.acc |= (u64::from(v) & mask) << self.n; + self.n += n; + while self.n >= 8 { + self.out.push(self.acc as u8); + self.acc >>= 8; + self.n -= 8; + } + } + + fn align(&mut self) { + if self.n > 0 { + self.out.push(self.acc as u8); + } + self.acc = 0; + self.n = 0; + } +} + +#[derive(Clone, Copy)] +enum Tok { + Lit(u8), + Match { len: u16, dist: u16 }, +} + +/// Index into [`LEN_BASE`] of a match length (3..=258). +fn len_code(len: u16) -> usize { + LEN_BASE.partition_point(|&b| b <= len).saturating_sub(1) +} + +/// Index into [`DIST_BASE`] of a distance (1..=32768). +fn dist_code(dist: u16) -> usize { + DIST_BASE.partition_point(|&b| b <= dist).saturating_sub(1) +} + +/// The match finder's hash heads and chains, reused from one input to the next. +/// +/// A slot holds `base + p + 1` for position `p` of the input that wrote it; anything at or below +/// the current input's `base` was left by an earlier input and reads as empty. So the tables are +/// never cleared between inputs: an OpenEXR ZIPS frame is one input per scan line, and zeroing +/// 256 KiB of tables for each 23 KB UHD line cost more memory traffic than the data itself. +#[derive(Default)] +pub struct Tables { + head: Vec, + prev: Vec, + /// Bytes compressed through these tables since they were last cleared. + base: u32, +} + +/// LZ77 match finder: hash heads and chains over the window. +struct Matcher<'a> { + data: &'a [u8], + head: &'a mut [u32], + prev: &'a mut [u32], + bits: u32, + /// `prev` slots used by this input, minus one (a power of two minus one). + mask: usize, + base: u32, +} + +impl<'a> Matcher<'a> { + /// A matcher over `data` (shorter than `u32::MAX` bytes), or None when it is not. + fn new(data: &'a [u8], t: &'a mut Tables) -> Option { + // hash width and chain span follow the input's size, as they did when every input had + // tables of its own (the output depends on them), so short chunks stay cheap to hash + let span = data.len().clamp(256, WINDOW).next_power_of_two(); + let bits = span.trailing_zeros().clamp(8, 15); + let n = u32::try_from(data.len()).ok().filter(|&n| n < u32::MAX)?; + if t.base.checked_add(n).and_then(|b| b.checked_add(1)).is_none() { + // positions would no longer fit after `base`: start over from empty tables + t.head.fill(0); + t.prev.fill(0); + t.base = 0; + } + // grown slots are 0, which is never above `base`, so they read as empty + if t.head.len() < 1 << bits { + t.head.resize(1 << bits, 0); + } + if t.prev.len() < span { + t.prev.resize(span, 0); + } + let base = t.base; + t.base += n; + let (head, prev) = (t.head.get_mut(..1 << bits)?, t.prev.get_mut(..span)?); + Some(Matcher { data, head, prev, bits, mask: span - 1, base }) + } + + /// Position `p` of this input from a stored slot, or None for an empty or earlier input's slot. + fn pos(&self, slot: u32) -> Option { + slot.checked_sub(self.base)?.checked_sub(1).map(|p| p as usize) + } + + fn hash(&self, p: usize) -> Option { + let b = self.data.get(p..p + MIN_MATCH)?; + let v = (u32::from(b[0]) << 16) | (u32::from(b[1]) << 8) | u32::from(b[2]); + Some((v.wrapping_mul(0x9E37_79B1) >> (32 - self.bits)) as usize) + } + + /// Record position `p` (stored as `p + 1`; 0 = empty). + fn insert(&mut self, p: usize) { + let Some(h) = self.hash(p) else { return }; + let (Some(head), Ok(p32)) = (self.head.get(h).copied(), u32::try_from(p)) else { return }; + if let Some(slot) = self.prev.get_mut(p & self.mask) { + *slot = head; + } + if let Some(slot) = self.head.get_mut(h) { + // `new` reserved `base + len + 1` within u32 + *slot = self.base + p32 + 1; + } + } + + /// The longest earlier match at `p`: (length, distance), length 0 if none. + fn longest(&self, p: usize) -> (usize, usize) { + let Some(h) = self.hash(p) else { return (0, 0) }; + let max = self.data.len().saturating_sub(p).min(MAX_MATCH); + let Some(here) = self.data.get(p..p + max) else { return (0, 0) }; + let (mut best, mut best_dist) = (0, 0); + let mut cand = self.head.get(h).and_then(|&s| self.pos(s)); + for _ in 0..MAX_CHAIN { + let Some(c) = cand else { break }; + if c >= p || p - c > WINDOW { + break; + } + // a candidate that differs at `best` cannot be longer than `best`: skip comparing it + if best == 0 || self.data.get(c + best) == here.get(best) { + let l = self.data.get(c..).unwrap_or_default().iter().zip(here).take_while(|(a, b)| a == b).count(); + if l > best { + (best, best_dist) = (l, p - c); + if l == max { + break; + } + } + } + let Some(next) = self.prev.get(c & self.mask).and_then(|&s| self.pos(s)) else { break }; + // chains only go back; anything else is a stale slot + if next >= c { + break; + } + cand = Some(next); + } + (best, best_dist) + } +} + +/// The deflate blocks of `data`, the last one final. +fn deflate(data: &[u8], w: &mut Bits, tables: &mut Tables) { + if data.is_empty() { + // a final fixed-Huffman block holding only end-of-block (code 0000000) + w.put(1, 1); + w.put(1, 2); + w.put(0, 7); + return; + } + // positions are kept in u32; longer inputs (never an OpenEXR chunk) go out stored + let mut m = Matcher::new(data, tables); + let mut toks = Vec::with_capacity(BLOCK_TOKENS.min(data.len()) + 1); + let (mut start, mut i) = (0, 0); + while i < data.len() { + let (len, dist) = m.as_ref().map_or((0, 0), |m| m.longest(i)); + let step = if len >= MIN_MATCH { + toks.push(Tok::Match { len: len as u16, dist: dist as u16 }); + len + } else { + toks.push(Tok::Lit(data[i])); + 1 + }; + if let Some(m) = m.as_mut() { + for p in i..i + step { + m.insert(p); + } + } + i += step; + if toks.len() >= BLOCK_TOKENS || i == data.len() { + block(w, &toks, &data[start..i], i == data.len()); + toks.clear(); + start = i; + } + } +} + +/// Bits the tokens take with the given code lengths (EOB included). +fn data_bits(lf: &[u32], df: &[u32], ll: &[u8], dl: &[u8]) -> u64 { + let mut n = 0_u64; + for (s, (&f, &l)) in lf.iter().zip(ll).enumerate() { + let extra = s.checked_sub(257).and_then(|k| LEN_EXTRA.get(k)).copied().unwrap_or(0); + n += u64::from(f) * u64::from(l + extra); + } + for (d, (&f, &l)) in df.iter().zip(dl).enumerate() { + n += u64::from(f) * u64::from(l + DIST_EXTRA.get(d).copied().unwrap_or(0)); + } + n +} + +/// One block (split into several stored blocks when stored is smallest and `raw` is long). +fn block(w: &mut Bits, toks: &[Tok], raw: &[u8], last: bool) { + let mut lf = [0_u32; 286]; + let mut df = [0_u32; 30]; + for t in toks { + match *t { + Tok::Lit(b) => lf[b as usize] += 1, + Tok::Match { len, dist } => { + lf[257 + len_code(len)] += 1; + df[dist_code(dist)] += 1; + } + } + } + lf[256] += 1; + + // dynamic codes and their header + let ll = lengths(&lf, 15); + let dl = lengths(&df, 15); + let hlit = ll.iter().rposition(|&l| l > 0).map_or(257, |p| (p + 1).max(257)); + let hdist = dl.iter().rposition(|&l| l > 0).map_or(1, |p| p + 1); + let seq = rle_lengths(&[&ll[..hlit], &dl[..hdist]].concat()); + let mut cf = [0_u32; 19]; + for &(s, _) in &seq { + cf[s as usize] += 1; + } + let cl = lengths(&cf, 7); + let hclen = CL_ORDER.iter().rposition(|&s| cl[s] > 0).map_or(4, |p| (p + 1).max(4)); + let seq_bits: u64 = + seq.iter().map(|&(s, _)| u64::from(cl[s as usize]) + [2_u64, 3, 7].get((s as usize).wrapping_sub(16)).copied().unwrap_or(0)).sum(); + let dyn_bits = 3 + 14 + 3 * hclen as u64 + seq_bits + data_bits(&lf, &df, &ll, &dl); + + let (fll, fdl) = fixed_lengths(); + let fixed_bits = 3 + data_bits(&lf, &df, &fll, &fdl); + // per stored piece: header, at most 7 padding bits, LEN and NLEN + let pieces = raw.len().div_ceil(MAX_STORED).max(1) as u64; + let stored_bits = pieces * (3 + 7 + 32) + 8 * raw.len() as u64; + + if stored_bits <= fixed_bits.min(dyn_bits) { + let n = raw.chunks(MAX_STORED).count(); + for (k, piece) in raw.chunks(MAX_STORED).enumerate() { + w.put(u32::from(last && k + 1 == n), 1); + w.put(0, 2); + w.align(); + let len = piece.len() as u16; + w.out.extend_from_slice(&len.to_le_bytes()); + w.out.extend_from_slice(&(!len).to_le_bytes()); + w.out.extend_from_slice(piece); + } + return; + } + w.put(u32::from(last), 1); + if fixed_bits <= dyn_bits { + w.put(1, 2); + write_tokens(w, toks, &fll, &fdl); + return; + } + w.put(2, 2); + w.put((hlit - 257) as u32, 5); + w.put((hdist - 1) as u32, 5); + w.put((hclen - 4) as u32, 4); + for &s in &CL_ORDER[..hclen] { + w.put(u32::from(cl[s]), 3); + } + let cc = codes(&cl); + for &(s, x) in &seq { + w.put(u32::from(cc[s as usize]), u32::from(cl[s as usize])); + match s { + 16 => w.put(u32::from(x), 2), + 17 => w.put(u32::from(x), 3), + 18 => w.put(u32::from(x), 7), + _ => {} + } + } + write_tokens(w, toks, &ll, &dl); +} + +/// The fixed Huffman code lengths (RFC 1951 §3.2.6). +fn fixed_lengths() -> ([u8; 288], [u8; 30]) { + let mut ll = [8_u8; 288]; + ll[144..256].fill(9); + ll[256..280].fill(7); + (ll, [5; 30]) +} + +fn write_tokens(w: &mut Bits, toks: &[Tok], ll: &[u8], dl: &[u8]) { + let lc = codes(ll); + let dc = codes(dl); + let sym = |w: &mut Bits, s: usize| w.put(u32::from(lc[s]), u32::from(ll[s])); + for t in toks { + match *t { + Tok::Lit(b) => sym(w, b as usize), + Tok::Match { len, dist } => { + let k = len_code(len); + sym(w, 257 + k); + w.put(u32::from(len - LEN_BASE[k]), u32::from(LEN_EXTRA[k])); + let d = dist_code(dist); + w.put(u32::from(dc[d]), u32::from(dl[d])); + w.put(u32::from(dist - DIST_BASE[d]), u32::from(DIST_EXTRA[d])); + } + } + } + sym(w, 256); +} + +/// Canonical Huffman codes for `lens`, bit-reversed for the LSB-first writer. +fn codes(lens: &[u8]) -> Vec { + let mut count = [0_u32; 16]; + for &l in lens { + if (1..16).contains(&l) { + count[l as usize] += 1; + } + } + let mut next = [0_u32; 16]; + let mut code = 0; + for bits in 1..16 { + code = (code + count[bits - 1]) << 1; + next[bits] = code; + } + lens.iter() + .map(|&l| { + if !(1..16).contains(&l) { + return 0; + } + let c = next[l as usize]; + next[l as usize] += 1; + (c.reverse_bits() >> (32 - u32::from(l))) as u16 + }) + .collect() +} + +/// Code lengths for `freq`, none longer than `limit`; at least two symbols always get a code. +fn lengths(freq: &[u32], limit: u8) -> Vec { + let mut f: Vec = freq.iter().map(|&x| u64::from(x)).collect(); + let mut used = f.iter().filter(|&&x| x > 0).count(); + for x in f.iter_mut() { + if used >= 2 { + break; + } + if *x == 0 { + *x = 1; + used += 1; + } + } + // halving converges: once every used frequency is 1 the tree is balanced (depth ≤ 9 here) + for _ in 0..66 { + let l = huffman(&f); + if l.iter().all(|&d| d <= limit) { + return l; + } + for x in f.iter_mut().filter(|x| **x > 0) { + *x = (*x >> 1).max(1); + } + } + huffman(&f) +} + +/// Huffman tree depths of the symbols with non-zero frequency (0 for the rest). +fn huffman(f: &[u64]) -> Vec { + let n = f.len(); + let mut heap: BinaryHeap> = f.iter().enumerate().filter(|(_, w)| **w > 0).map(|(i, &w)| Reverse((w, i))).collect(); + let mut parent = vec![usize::MAX; 2 * n]; + let mut next = n; + while heap.len() > 1 { + let (Some(Reverse((wa, a))), Some(Reverse((wb, b)))) = (heap.pop(), heap.pop()) else { break }; + parent[a] = next; + parent[b] = next; + heap.push(Reverse((wa + wb, next))); + next += 1; + } + // parents always come after their children, so walk down from the root + let mut depth = vec![0_u32; next]; + for i in (0..next).rev() { + if let Some(&p) = parent.get(i).filter(|&&p| p < next) { + depth[i] = depth[p].saturating_add(1); + } + } + depth[..n].iter().map(|&d| d.min(u32::from(u8::MAX)) as u8).collect() +} + +/// Code lengths run-length coded with the code-length alphabet: (symbol, extra-bits value). +fn rle_lengths(lens: &[u8]) -> Vec<(u8, u8)> { + let mut out = Vec::new(); + let mut i = 0; + while i < lens.len() { + let l = lens[i]; + let run = lens[i..].iter().take_while(|&&x| x == l).count(); + let mut r = run; + if l == 0 { + while r >= 11 { + let k = r.min(138); + out.push((18, (k - 11) as u8)); + r -= k; + } + if r >= 3 { + out.push((17, (r - 3) as u8)); + r = 0; + } + } else { + out.push((l, 0)); + r -= 1; + while r >= 3 { + let k = r.min(6); + out.push((16, (k - 3) as u8)); + r -= k; + } + } + out.extend(std::iter::repeat_n((l, 0), r)); + i += run; + } + out +} + +#[cfg(test)] +mod tests { + use super::*; + + /// A minimal inflater (RFC 1951), only to check the encoder; returns None on any error. + pub(crate) fn inflate(z: &[u8]) -> Option> { + let (cmf, flg) = (*z.first()?, *z.get(1)?); + if cmf & 0x0F != 8 || (u16::from(cmf) << 8 | u16::from(flg)) % 31 != 0 || flg & 0x20 != 0 { + return None; + } + let mut pos = 16_usize; // bit position + let bit = |pos: &mut usize| -> Option { + let b = (*z.get(*pos / 8)? >> (*pos % 8)) & 1; + *pos += 1; + Some(u32::from(b)) + }; + let bits = |pos: &mut usize, n: u32| -> Option { + let mut v = 0; + for k in 0..n { + v |= bit(pos)? << k; + } + Some(v) + }; + // decode one symbol with canonical lengths (slow but simple) + let decode = |pos: &mut usize, lens: &[u8]| -> Option { + let (mut code, mut first, mut index) = (0_i32, 0_i32, 0_i32); + let mut count = [0_i32; 16]; + for &l in lens { + count[l as usize] += 1; + } + let mut sorted: Vec = (0..lens.len()).filter(|&s| lens[s] > 0).collect(); + sorted.sort_by_key(|&s| (lens[s], s)); + for &c in count.iter().skip(1) { + code |= bit(pos)? as i32; + if code - c < first { + return sorted.get((index + code - first) as usize).copied(); + } + index += c; + first = (first + c) << 1; + code <<= 1; + } + None + }; + let mut out = Vec::new(); + loop { + let last = bit(&mut pos)?; + match bits(&mut pos, 2)? { + 0 => { + pos = pos.div_ceil(8) * 8; + let at = pos / 8; + let len = u16::from_le_bytes([*z.get(at)?, *z.get(at + 1)?]); + let nlen = u16::from_le_bytes([*z.get(at + 2)?, *z.get(at + 3)?]); + if len != !nlen { + return None; + } + out.extend_from_slice(z.get(at + 4..at + 4 + len as usize)?); + pos = (at + 4 + len as usize) * 8; + } + t @ (1 | 2) => { + let (ll, dl) = if t == 1 { + let (a, b) = fixed_lengths(); + (a.to_vec(), b.to_vec()) + } else { + let hlit = bits(&mut pos, 5)? as usize + 257; + let hdist = bits(&mut pos, 5)? as usize + 1; + let hclen = bits(&mut pos, 4)? as usize + 4; + let mut cl = [0_u8; 19]; + for &s in &CL_ORDER[..hclen] { + cl[s] = bits(&mut pos, 3)? as u8; + } + let mut lens = Vec::new(); + while lens.len() < hlit + hdist { + match decode(&mut pos, &cl)? { + s @ 0..=15 => lens.push(s as u8), + 16 => { + let p = *lens.last()?; + lens.extend(std::iter::repeat_n(p, 3 + bits(&mut pos, 2)? as usize)); + } + 17 => lens.extend(std::iter::repeat_n(0, 3 + bits(&mut pos, 3)? as usize)), + _ => lens.extend(std::iter::repeat_n(0, 11 + bits(&mut pos, 7)? as usize)), + } + } + if lens.len() != hlit + hdist { + return None; + } + (lens[..hlit].to_vec(), lens[hlit..].to_vec()) + }; + loop { + let s = decode(&mut pos, &ll)?; + if s < 256 { + out.push(s as u8); + } else if s == 256 { + break; + } else { + let k = s - 257; + let len = *LEN_BASE.get(k)? as usize + bits(&mut pos, u32::from(LEN_EXTRA[k]))? as usize; + let d = decode(&mut pos, &dl)?; + let dist = *DIST_BASE.get(d)? as usize + bits(&mut pos, u32::from(DIST_EXTRA[d]))? as usize; + let from = out.len().checked_sub(dist)?; + for j in 0..len { + out.push(out[from + j]); + } + } + } + } + _ => return None, + } + if last == 1 { + break; + } + } + let at = pos.div_ceil(8); + let sum = u32::from_be_bytes(z.get(at..at + 4)?.try_into().ok()?); + (sum == adler32(&out) && at + 4 == z.len()).then_some(out) + } + + /// A deterministic pseudo-random byte stream. + fn noise(n: usize, seed: u32) -> Vec { + let mut s = seed | 1; + (0..n) + .map(|_| { + s ^= s << 13; + s ^= s >> 17; + s ^= s << 5; + s as u8 + }) + .collect() + } + + #[test] + fn round_trips() { + let mut cases: Vec> = vec![vec![], vec![7], vec![1, 2], vec![0; 1], vec![0; 300_000], b"abcabcabcabcabcabd".to_vec()]; + cases.push(noise(100_000, 9)); // incompressible: stored blocks + cases.push(noise(70_000, 3).iter().map(|b| b % 4).collect()); // few symbols: dynamic + cases.push((0..200_000_u32).map(|i| ((i / 7) % 251) as u8).collect()); // long repeats + // text-like: a small vocabulary repeated at varying distances + let words: [&[u8]; 6] = [b"red ", b"green ", b"blue ", b"alpha ", b"half ", b"linear "]; + cases.push(noise(40_000, 5).iter().flat_map(|b| words[*b as usize % 6].iter().copied()).collect()); + // skewed frequencies: Huffman lengths beyond 15 before limiting + let mut fib = Vec::new(); + let (mut a, mut b) = (1_usize, 1_usize); + for s in 0..25_u8 { + fib.extend(std::iter::repeat_n(s.wrapping_mul(37), a.min(200_000))); + (a, b) = (b, a + b); + } + let r = noise(fib.len() * 4, 11); + for i in (1..fib.len()).rev() { + let j = u32::from_le_bytes(r[i * 4..i * 4 + 4].try_into().unwrap()) as usize % (i + 1); + fib.swap(i, j); + } + cases.push(fib); + for (k, data) in cases.iter().enumerate() { + let z = zlib(data, &mut Tables::default()); + assert_eq!(inflate(&z).as_deref(), Some(data.as_slice()), "case {k}"); + // worst case: stored overhead only + assert!(z.len() <= data.len() + (data.len().div_ceil(BLOCK_TOKENS) + 1) * 5 + 6, "case {k}: {} > {}", z.len(), data.len()); + } + // repetitive data actually compresses + assert!(zlib(&cases[4], &mut Tables::default()).len() < 2_000); + assert!(zlib(&cases[8], &mut Tables::default()).len() < 20_000); + } + + #[test] + fn length_limits_hold() { + // Fibonacci frequencies give the deepest possible Huffman tree + let mut f = vec![0_u32; 286]; + let (mut a, mut b) = (1_u32, 1_u32); + for x in f.iter_mut().take(30) { + *x = a; + (a, b) = (b, a.saturating_add(b)); + } + let l = lengths(&f, 15); + assert!(l.iter().all(|&d| d <= 15)); + // a complete code: Kraft sum exactly 1 + let kraft: f64 = l.iter().filter(|&&d| d > 0).map(|&d| 0.5_f64.powi(i32::from(d))).sum(); + assert!((kraft - 1.0).abs() < 1e-12); + // one used symbol still gets a two-symbol code + let l = lengths(&[0, 5, 0], 7); + assert_eq!(l.iter().filter(|&&d| d == 1).count(), 2); + } + + #[test] + fn adler_matches_rfc_example() { + // the classic test vector for "Wikipedia" + assert_eq!(adler32(b"Wikipedia"), 0x11E6_0398); + assert_eq!(adler32(&[]), 1); + } + + /// The match finder before #4659: tables of its own per input, every chain candidate compared + /// from its first byte. + struct OldMatcher<'a> { + data: &'a [u8], + head: Vec, + prev: Vec, + bits: u32, + } + + impl<'a> OldMatcher<'a> { + /// Tables of its own, sized to `data`. + fn new(data: &'a [u8]) -> Self { + let span = data.len().clamp(256, WINDOW).next_power_of_two(); + let bits = span.trailing_zeros().clamp(8, 15); + OldMatcher { data, head: vec![0; 1 << bits], prev: vec![0; span], bits } + } + + /// The hash of the three bytes at `p`, None near the end. + fn hash(&self, p: usize) -> Option { + let b = self.data.get(p..p + MIN_MATCH)?; + let v = (u32::from(b[0]) << 16) | (u32::from(b[1]) << 8) | u32::from(b[2]); + Some((v.wrapping_mul(0x9E37_79B1) >> (32 - self.bits)) as usize) + } + + /// Record position `p` (stored as `p + 1`; 0 = empty). + fn insert(&mut self, p: usize) { + let Some(h) = self.hash(p) else { return }; + let mask = self.prev.len() - 1; + self.prev[p & mask] = self.head[h]; + self.head[h] = p as u32 + 1; + } + + /// The longest earlier match at `p`: (length, distance), length 0 if none. + fn longest(&self, p: usize) -> (usize, usize) { + let Some(h) = self.hash(p) else { return (0, 0) }; + let max = (self.data.len() - p).min(MAX_MATCH); + let here = &self.data[p..p + max]; + let mask = self.prev.len() - 1; + let (mut best, mut best_dist) = (0, 0); + let mut cand = self.head[h] as usize; + for _ in 0..MAX_CHAIN { + let Some(c) = cand.checked_sub(1) else { break }; + if c >= p || p - c > WINDOW { + break; + } + let l = self.data[c..].iter().zip(here).take_while(|(a, b)| a == b).count(); + if l > best { + (best, best_dist) = (l, p - c); + if l == max { + break; + } + } + let next = self.prev[c & mask] as usize; + if next == 0 || next > c { + break; + } + cand = next; + } + (best, best_dist) + } + } + + /// Inputs shaped like OpenEXR chunks (predicted half-float scan lines, single and 16-line) + /// and the awkward cases: empty, tiny, incompressible, long runs, few symbols. + fn chunk_cases() -> Vec> { + let line: Vec = (0..23_040_u32).map(|i| (128 + ((i / 3) % 7) as i32 - 3 + (i % 5 == 0) as i32) as u8).collect(); + let mut cases = vec![line.clone(), line.iter().rev().copied().collect(), vec![], vec![9], b"abcabcabcabd".to_vec()]; + cases.push(line.iter().cycle().take(16 * 23_040).copied().collect()); // ZIP: 16 lines, longer than the window + cases.push(noise(23_040, 4)); + cases.push(noise(70_000, 3).iter().map(|b| b % 4).collect()); + cases.push(vec![0; 300_000]); + cases.push(line.clone()); + cases.push(noise(300, 7)); + cases + } + + /// The matcher's greedy parse of `data`: what `deflate` codes. + fn parse (usize, usize), G: FnMut(usize)>(data: &[u8], mut longest: F, mut insert: G) -> Vec<(usize, usize)> { + let (mut out, mut i) = (Vec::new(), 0); + while i < data.len() { + let (len, dist) = longest(i); + let step = if len >= MIN_MATCH { len } else { 1 }; + out.push(if len >= MIN_MATCH { (len, dist) } else { (0, 0) }); + for p in i..i + step { + insert(p); + } + i += step; + } + out + } + + /// Shared tables and the candidate skip find exactly the matches the previous matcher found. + #[test] + fn shared_tables_find_the_matches_the_old_per_chunk_tables_found() { + let mut tables = Tables::default(); + for (k, data) in chunk_cases().iter().enumerate() { + let old = std::cell::RefCell::new(OldMatcher::new(data)); + let want = parse(data, |p| old.borrow().longest(p), |p| old.borrow_mut().insert(p)); + let new = std::cell::RefCell::new(Matcher::new(data, &mut tables).expect("fits")); + let got = parse(data, |p| new.borrow().longest(p), |p| new.borrow_mut().insert(p)); + assert!(got == want, "chunk {k}: the parse differs"); + } + } + + /// Shared tables compress to the bytes fresh ones give, also when positions run out and they start over. + #[test] + fn shared_tables_give_the_bytes_of_fresh_ones() { + let mut shared = Tables::default(); + for (k, data) in chunk_cases().iter().enumerate() { + let z = zlib(data, &mut shared); + assert!(z == zlib(data, &mut Tables::default()), "chunk {k}"); + assert_eq!(inflate(&z).as_deref(), Some(data.as_slice()), "chunk {k}"); + } + // positions about to run past u32: the tables start over and the bytes stay the same + for gap in [0_u32, 1, 2, 23_040, 23_041, 1 << 20] { + shared.base = u32::MAX - gap; + for (k, data) in chunk_cases().iter().enumerate() { + assert!(zlib(data, &mut shared) == zlib(data, &mut Tables::default()), "gap {gap}, chunk {k}"); + } + } + } + + /// Later chunks use the same tables instead of allocating new ones. + #[test] + fn the_tables_are_reused_across_chunks() { + let mut tables = Tables::default(); + let line = &chunk_cases()[0]; + zlib(line, &mut tables); + let (head, prev) = (tables.head.as_ptr(), tables.prev.as_ptr()); + for _ in 0..4 { + zlib(line, &mut tables); + assert!(tables.head.as_ptr() == head && tables.prev.as_ptr() == prev, "a chunk allocated new tables"); + } + assert_eq!(tables.base as usize, 5 * line.len()); + } +} diff --git a/crates/fcproj-format/Cargo.toml b/crates/fcproj-format/Cargo.toml new file mode 100644 index 000000000..aa210187d --- /dev/null +++ b/crates/fcproj-format/Cargo.toml @@ -0,0 +1,16 @@ +[package] +name = "craft-fcproj-format" +description = "FilmCraft project envelope and schema migrations" +version.workspace = true +edition.workspace = true +license.workspace = true +rust-version.workspace = true +repository.workspace = true + +[lints] +workspace = true + +[dependencies] +serde = { workspace = true } +serde_json = { workspace = true } +thiserror = { workspace = true } diff --git a/crates/fcproj-format/src/lib.rs b/crates/fcproj-format/src/lib.rs new file mode 100644 index 000000000..98e50cf59 --- /dev/null +++ b/crates/fcproj-format/src/lib.rs @@ -0,0 +1,297 @@ +//! FilmCraft `.fcproj` JSON envelope, version detection and historical migrations. +//! Project and editor-view types are supplied by the caller. +#![cfg_attr(not(test), deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable))] +use serde::{Deserialize, Serialize}; +use serde_json::Value; +/// The `format` tag written in every envelope. +pub const FORMAT_ID: &str = "filmcraft.project"; + +/// A migration upgrades a whole document from schema `n` to `n + 1`. +pub type Migration = fn(Value) -> Result; + +/// `MIGRATIONS[i]` upgrades schema `i + 1` to `i + 2`. Append one function per schema bump; never +/// edit a shipped one. +pub const MIGRATIONS: &[Migration] = &[ + v1_to_v2, v2_to_v3, v3_to_v4, v4_to_v5, v5_to_v6, v6_to_v7, v7_to_v8, v8_to_v9, v9_to_v10, v10_to_v11, v11_to_v12, v12_to_v13, v13_to_v14, + v14_to_v15, v15_to_v16, v16_to_v17, +]; + +/// The schema version this build writes (and the newest it reads). +pub const SCHEMA_VERSION: u32 = MIGRATIONS.len() as u32 + 1; + +/// Oldest schema this build can still upgrade. +pub const OLDEST_SCHEMA: u32 = 1; + +#[derive(Debug, thiserror::Error)] +pub enum FormatError { + #[error("not a FilmCraft project: {0}")] + NotAProject(String), + #[error( + "this project was saved by a newer version of FilmCraft (project schema v{found}); this build reads up to v{supported}. Update FilmCraft to open it." + )] + TooNew { found: u32, supported: u32 }, + #[error("project schema v{found} is too old for this build (oldest supported: v{oldest})")] + TooOld { found: u32, oldest: u32 }, + #[error("upgrading project from schema v{from} to v{}: {msg}", from + 1)] + Migration { from: u32, msg: String }, + #[error("project file is damaged: {0}")] + Corrupt(String), +} + +/// Serialized current-schema envelope. +#[derive(Serialize)] +pub struct EnvelopeOut<'a, P, V> { + pub format: &'static str, + pub schema_version: u32, + pub generator: String, + pub project: &'a P, + #[serde(skip_serializing_if = "Option::is_none")] + pub view: Option<&'a V>, +} +/// Parsed envelope. View parsing remains optional to preserve forward compatibility. +#[derive(Deserialize)] +pub struct EnvelopeIn

{ + #[serde(default)] + pub generator: Option, + pub project: P, + #[serde(default)] + pub view: Option, +} +/// Serialize the current envelope, with optional editor view and caller-provided generator. +pub fn encode(project: &P, view: Option<&V>, generator: String, pretty: bool) -> Vec { + let envelope = EnvelopeOut { format: FORMAT_ID, schema_version: SCHEMA_VERSION, generator, project, view }; + let result = if pretty { serde_json::to_vec_pretty(&envelope) } else { serde_json::to_vec(&envelope) }; + result.unwrap_or_default() +} + +/// Decode an envelope, migrating older documents while retaining the current-version fast path. +pub fn decode(bytes: &[u8]) -> Result<(EnvelopeIn

, u32), FormatError> { + let found = detect_bytes(bytes)?; + if found == SCHEMA_VERSION { + let env = serde_json::from_slice(bytes).map_err(|e| FormatError::Corrupt(e.to_string()))?; + return Ok((env, found)); + } + let doc: Value = serde_json::from_slice(bytes).map_err(|e| FormatError::Corrupt(e.to_string()))?; + let doc = migrate_with(MIGRATIONS, doc, found)?; + let env = serde_json::from_value(doc).map_err(|e| FormatError::Corrupt(format!("after upgrading from schema v{found}: {e}")))?; + Ok((env, found)) +} + +/// Probe and validate the schema directly from bytes without building a JSON tree. +pub fn detect_bytes(bytes: &[u8]) -> Result { + let probe: Probe = serde_json::from_slice(bytes) + .map_err(|e| if e.is_syntax() || e.is_eof() { FormatError::Corrupt(e.to_string()) } else { FormatError::NotAProject(e.to_string()) })?; + let found = probe_version(&probe)?; + check_supported(found)?; + Ok(found) +} +/// Just enough of a document to know which schema it is. +#[derive(Deserialize)] +struct Probe { + #[serde(default)] + format: Option, + #[serde(default)] + schema_version: Option, + /// Bare (pre-envelope) project: its `version` field. + #[serde(default)] + version: Option, + #[serde(default)] + items: Option, + #[serde(default)] + root: Option, +} + +/// Detect the schema version of a parsed document. +pub fn detect_version(doc: &Value) -> Result { + let probe: Probe = serde_json::from_value(doc.clone()).map_err(|e| FormatError::NotAProject(e.to_string()))?; + probe_version(&probe) +} + +fn probe_version(p: &Probe) -> Result { + if let Some(v) = p.schema_version { + if let Some(f) = &p.format + && f != FORMAT_ID + { + return Err(FormatError::NotAProject(format!("format is `{f}`, expected `{FORMAT_ID}`"))); + } + return Ok(v); + } + // Before the envelope (v1) a file was the bare Project object with `version`. + // Its `version` was 1, or 2 on builds that briefly bumped it when caption tracks were added + // (same shape; the new field loads with its default). Anything higher is from the future. + if p.items.is_some() && p.root.is_some() { + return Ok(match p.version.unwrap_or(1) { + 0..=2 => 1, + v => v, + }); + } + Err(FormatError::NotAProject("no `schema_version` and not a legacy project".into())) +} + +pub fn check_supported(found: u32) -> Result<(), FormatError> { + if found > SCHEMA_VERSION { + return Err(FormatError::TooNew { found, supported: SCHEMA_VERSION }); + } + if found < OLDEST_SCHEMA { + return Err(FormatError::TooOld { found, oldest: OLDEST_SCHEMA }); + } + Ok(()) +} + +/// Run `table` from schema `from` up to `table.len() + 1`, stamping `schema_version` after each step. +pub fn migrate_with(table: &[Migration], mut doc: Value, from: u32) -> Result { + let target = table.len() as u32 + 1; + if from > target { + return Err(FormatError::TooNew { found: from, supported: target }); + } + if from == 0 { + return Err(FormatError::TooOld { found: 0, oldest: 1 }); + } + for v in from..target { + let step = table[(v - 1) as usize]; + doc = step(doc).map_err(|msg| FormatError::Migration { from: v, msg })?; + match doc.as_object_mut() { + Some(o) => { + o.insert("schema_version".into(), Value::from(v + 1)); + } + None => return Err(FormatError::Migration { from: v, msg: "migration produced a non-object document".into() }), + } + } + Ok(doc) +} + +/// v1 → v2: wrap the bare project in the envelope and drop the per-project `version` field. +fn v1_to_v2(doc: Value) -> Result { + let Value::Object(mut project) = doc else { return Err("expected a JSON object".into()) }; + project.remove("version"); + Ok(serde_json::json!({ + "format": FORMAT_ID, + "schema_version": 2, + "generator": "FilmCraft (schema v1)", + "project": Value::Object(project), + })) +} + +/// v2 → v3: graphic clips added; existing data needs no change. +fn v2_to_v3(doc: Value) -> Result { + Ok(doc) +} + +/// v3 → v4: Essential Sound fields added; existing data needs no change. +fn v3_to_v4(doc: Value) -> Result { + Ok(doc) +} + +/// v4 → v5: colour-management fields added; existing data needs no change. +fn v4_to_v5(doc: Value) -> Result { + Ok(doc) +} + +/// v5 → v6: media identity and ingest settings added; existing data needs no change. +fn v5_to_v6(doc: Value) -> Result { + Ok(doc) +} + +/// v6 → v7: effect masks (Bézier mask paths as keyframable parameters); existing data needs no change. +fn v6_to_v7(doc: Value) -> Result { + Ok(doc) +} + +/// v7 → v8: multi-camera source sequences / clips and merged clips added; existing data needs no +/// change. +fn v7_to_v8(doc: Value) -> Result { + Ok(doc) +} + +/// v8 → v9: per-media-item transcripts (`project.transcripts`) added; existing data needs no change. +fn v8_to_v9(doc: Value) -> Result { + Ok(doc) +} + +/// v9 → v10: clip time interpolation, Hold Filters, Field Options and audio source channels; the +/// Modify ▸ Audio Channels map; subclips' Restrict Trims flag. Existing data needs no change. +fn v9_to_v10(doc: Value) -> Result { + Ok(doc) +} + +/// v10 → v11: search bins (`project.search_bins`), Flash Cue markers, Project Settings safe areas, +/// capture format and scratch disks. Existing data needs no change. +fn v10_to_v11(doc: Value) -> Result { + Ok(doc) +} + +/// v11 → v12: graphics templates, rolls / crawls, responsive design and per-character text styles +/// on graphic clips, and source graphics. Existing data needs no change. +fn v11_to_v12(doc: Value) -> Result { + Ok(doc) +} + +/// v12 → v13: container audio streams and per-clip stream selection. `MediaInfo` reads the old +/// `audio` field as stream zero; `TrackItem::audio_stream` defaults to zero. +fn v12_to_v13(doc: Value) -> Result { + Ok(doc) +} + +/// v13 → v14: narrations (Text to Speech). Existing data needs no change. +fn v13_to_v14(doc: Value) -> Result { + Ok(doc) +} + +/// v14 → v15: project notes. Existing data needs no change. +fn v14_to_v15(doc: Value) -> Result { + Ok(doc) +} + +/// v15 → v16: existing colors remain presets; authors and the named custom palette default empty. +fn v15_to_v16(doc: Value) -> Result { + Ok(doc) +} + +/// v16 → v17: marker URLs default empty. Older builds must not silently drop saved URLs. +fn v16_to_v17(doc: Value) -> Result { + Ok(doc) +} + +#[cfg(test)] +mod tests { + use super::*; + use serde_json::json; + + #[test] + fn legacy_and_current_envelopes_preserve_values() { + let bare = json!({"version": 2, "items": {}, "root": {}, "extra": [1, 2]}); + assert_eq!(detect_version(&bare).unwrap(), 1); + let migrated = migrate_with(MIGRATIONS, bare, 1).unwrap(); + assert_eq!(migrated["schema_version"], 17); + assert_eq!(migrated["project"]["extra"], json!([1, 2])); + assert_eq!(migrated["generator"], "FilmCraft (schema v1)"); + assert_eq!(detect_bytes(serde_json::to_string(&migrated).unwrap().as_bytes()).unwrap(), 17); + let parsed: EnvelopeIn = serde_json::from_value(migrated).unwrap(); + assert!(parsed.view.is_none()); + assert_eq!(parsed.project["extra"], json!([1, 2])); + } + #[test] + fn typed_encode_and_decode_current_and_legacy() { + let project = json!({"items": {}, "root": {}, "name": "Cut"}); + let view = json!({"active": 3}); + for pretty in [false, true] { + let bytes = encode(&project, Some(&view), "FilmCraft test".into(), pretty); + let (file, version): (EnvelopeIn, _) = decode(&bytes).unwrap(); + assert_eq!(version, SCHEMA_VERSION); + assert_eq!(file.project, project); + assert_eq!(file.view, Some(view.clone())); + } + let old = br#"{"version":1,"items":{},"root":{},"name":"Cut"}"#; + let (file, version): (EnvelopeIn, _) = decode(old).unwrap(); + assert_eq!(version, 1); + assert_eq!(file.project["name"], "Cut"); + } + #[test] + fn rejects_wrong_future_and_damaged_envelopes() { + assert!(matches!(detect_bytes(br#"{"format":"other","schema_version":1}"#), Err(FormatError::NotAProject(_)))); + assert!(matches!(detect_bytes(br#"{"schema_version":18}"#), Err(FormatError::TooNew { found: 18, supported: 17 }))); + assert!(matches!(detect_bytes(br#"{"schema_version":0}"#), Err(FormatError::TooOld { found: 0, oldest: 1 }))); + assert!(matches!(detect_bytes(b"{"), Err(FormatError::Corrupt(_)))); + assert!(matches!(migrate_with(MIGRATIONS, json!([]), 1), Err(FormatError::Migration { from: 1, .. }))); + } +} diff --git a/crates/fig/src/archive.rs b/crates/fig/src/archive.rs index e408ddbe7..e083aa1dc 100644 --- a/crates/fig/src/archive.rs +++ b/crates/fig/src/archive.rs @@ -12,8 +12,10 @@ use crate::{Error, corrupt}; const MAX_META: u64 = 1 << 20; /// Largest file [`write_archive`] writes (the canvas and the images together). const MAX_FILE: u64 = 2 * 1024 * 1024 * 1024; -/// Most entries an archive may declare; checked before the archive is indexed. -pub const MAX_ENTRIES: u64 = 1_000_000; +/// Most entries an archive may declare; checked before the archive is indexed. The archive holds one +/// entry per image, so 100 000 leaves room for far more than a `.fig` file carries, and a header +/// count can no longer make the index alone cost a hundred megabytes. +pub const MAX_ENTRIES: u64 = 100_000; /// What one open may read from `images/`. #[derive(Clone, Copy, Debug, PartialEq, Eq)] diff --git a/crates/figcraft-format/Cargo.toml b/crates/figcraft-format/Cargo.toml new file mode 100644 index 000000000..009e40c01 --- /dev/null +++ b/crates/figcraft-format/Cargo.toml @@ -0,0 +1,18 @@ +[package] +name = "craft-figcraft-format" +description = "The native FigCraft document JSON and archive contract." +version.workspace = true +edition.workspace = true +license.workspace = true +rust-version.workspace = true +repository.workspace = true + +[lints] +workspace = true + +[dependencies] +serde = { workspace = true } +serde_json = { workspace = true } +craft-zip = { workspace = true } +zip = { version = "9", default-features = false, features = ["deflate-flate2"] } +thiserror = { workspace = true } diff --git a/crates/figcraft-format/src/lib.rs b/crates/figcraft-format/src/lib.rs new file mode 100644 index 000000000..7475c5e21 --- /dev/null +++ b/crates/figcraft-format/src/lib.rs @@ -0,0 +1,849 @@ +#![deny(clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::unimplemented, clippy::todo, clippy::unreachable)] +//! The neutral JSON header and version contract for FigCraft's native project format. +//! App-specific document fields remain a generic flattened payload. +pub mod library; +use serde::de::{IgnoredAny, MapAccess, Visitor}; +use serde::{Deserialize, Deserializer, Serialize}; +use serde_json::Value; + +pub const EXTENSION: &str = "figcraft"; +pub const FORMAT: &str = "figcraft"; +pub const VERSION: u32 = 1; + +fn is_false(value: &bool) -> bool { + !value +} + +/// One native version-history index entry. Invalid optional values decode to the same defaults as +/// the historical lenient reader; the app applies document-specific validity and issue reporting. +#[derive(Clone, Debug, Default, PartialEq, Serialize)] +pub struct VersionEntry { + pub id: String, + pub name: String, + pub description: String, + pub time: u64, + #[serde(skip_serializing_if = "is_false")] + pub auto: bool, + pub images: Vec, +} + +impl<'de> Deserialize<'de> for VersionEntry { + fn deserialize>(deserializer: D) -> Result { + let value = Value::deserialize(deserializer)?; + let string = |key: &str| value.get(key).and_then(Value::as_str).unwrap_or_default().to_owned(); + let images = value.get("images").and_then(Value::as_array).into_iter().flatten().filter_map(Value::as_str).map(str::to_owned).collect(); + Ok(Self { + id: string("id"), + name: string("name"), + description: string("description"), + time: value.get("time").and_then(Value::as_u64).unwrap_or(0), + auto: value.get("auto").and_then(Value::as_bool).unwrap_or(false), + images, + }) + } +} + +/// Native version-history index, oldest first. +#[derive(Clone, Debug, Default, PartialEq, Serialize)] +pub struct VersionIndex { + pub versions: Vec, +} + +/// Encode the native version-history index using the FigCraft wire schema. +pub fn write_version_index(index: &VersionIndex) -> Result, serde_json::Error> { + serde_json::to_vec(index) +} + +/// Decode the native version-history index using the format's lenient optional-field defaults. +pub fn read_version_index(bytes: &[u8]) -> Result { + serde_json::from_slice(bytes) +} + +impl<'de> Deserialize<'de> for VersionIndex { + fn deserialize>(deserializer: D) -> Result { + let value = Value::deserialize(deserializer)?; + let versions = value + .get("versions") + .and_then(Value::as_array) + .map(|entries| entries.iter().cloned().map(serde_json::from_value).collect::, _>>()) + .transpose() + .map_err(serde::de::Error::custom)? + .unwrap_or_default(); + Ok(Self { versions }) + } +} + +/// State stored on a native branch entry. +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, Serialize, Deserialize)] +#[serde(rename_all = "camelCase")] +pub enum BranchState { + #[default] + Active, + Merged, + Archived, +} + +/// One native branch index entry. +#[derive(Clone, Debug, Default, PartialEq, Serialize)] +#[serde(rename_all = "camelCase")] +pub struct BranchEntry { + pub id: String, + pub name: String, + pub time: u64, + pub client: u32, + pub state: BranchState, + pub base_images: Vec, + pub head_images: Vec, +} + +impl<'de> Deserialize<'de> for BranchEntry { + fn deserialize>(deserializer: D) -> Result { + let value = Value::deserialize(deserializer)?; + let string = |key: &str| value.get(key).and_then(Value::as_str).unwrap_or_default().to_owned(); + let images = + |key: &str| value.get(key).and_then(Value::as_array).into_iter().flatten().filter_map(Value::as_str).map(str::to_owned).collect(); + let state = value.get("state").cloned().and_then(|v| serde_json::from_value(v).ok()).unwrap_or_default(); + let client = value.get("client").and_then(Value::as_u64).and_then(|v| u32::try_from(v).ok()).unwrap_or(0); + Ok(Self { + id: string("id"), + name: string("name"), + time: value.get("time").and_then(Value::as_u64).unwrap_or(0), + client, + state, + base_images: images("baseImages"), + head_images: images("headImages"), + }) + } +} + +/// Native branch index. +#[derive(Clone, Debug, Default, PartialEq, Serialize)] +pub struct BranchIndex { + pub branches: Vec, +} + +/// Encode the native branch index using the FigCraft wire schema. +pub fn write_branch_index(index: &BranchIndex) -> Result, serde_json::Error> { + serde_json::to_vec(index) +} + +/// Decode the native branch index using the format's lenient optional-field defaults. +pub fn read_branch_index(bytes: &[u8]) -> Result { + serde_json::from_slice(bytes) +} + +impl<'de> Deserialize<'de> for BranchIndex { + fn deserialize>(deserializer: D) -> Result { + let value = Value::deserialize(deserializer)?; + let branches = value + .get("branches") + .and_then(Value::as_array) + .map(|entries| entries.iter().cloned().map(serde_json::from_value).collect::, _>>()) + .transpose() + .map_err(serde::de::Error::custom)? + .unwrap_or_default(); + Ok(Self { branches }) + } +} + +#[derive(Clone, Debug, Default, PartialEq, Eq)] +pub struct Header { + pub version: Option, + pub images: Option>, +} + +pub fn read_header(text: &str) -> Result { + serde_json::from_str(text) +} + +impl<'de> Deserialize<'de> for Header { + fn deserialize>(d: D) -> Result { + struct V; + impl<'de> Visitor<'de> for V { + type Value = Header; + fn expecting(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { + f.write_str("a FigCraft document object") + } + fn visit_map>(self, mut m: A) -> Result { + let mut h = Header::default(); + while let Some(k) = m.next_key::()? { + match k.as_str() { + "version" => h.version = m.next_value::()?.as_u64(), + "imageIndex" => { + h.images = m + .next_value::()? + .as_array() + .map(|a| a.iter().filter_map(|v| v.get("hash").and_then(Value::as_str).map(str::to_owned)).collect()) + } + _ => { + m.next_value::()?; + } + } + } + Ok(h) + } + } + d.deserialize_map(V) + } +} + +#[derive(Clone, Copy, Debug, PartialEq, Eq, thiserror::Error)] +#[error("this file was saved by a newer FigCraft (format version {0})")] +pub struct NewerVersion(pub u32); + +/// Failure while reading the native document envelope or its app-supplied payload type. +#[derive(Debug, thiserror::Error)] +pub enum ReadDocumentError { + #[error(transparent)] + Json(#[from] serde_json::Error), + #[error(transparent)] + NewerVersion(#[from] NewerVersion), +} + +/// Read a document payload after checking the native header and version first. +/// +/// The payload type is supplied by the consumer, keeping app model conversion outside this crate. +pub fn read_document_json(text: &str) -> Result<(D, Header), ReadDocumentError> { + let header = read_header(text)?; + check_version(&header)?; + let document = serde_json::from_str(text)?; + Ok((document, header)) +} + +/// The native document fields needed to render a recent-file preview. +#[derive(Clone, Debug, Default, PartialEq, Deserialize)] +pub struct PreviewDocument { + #[serde(default)] + pub pages: Vec, + #[serde(default)] + pub nodes: std::collections::BTreeMap, + #[serde(default)] + pub kind: PreviewKind, +} + +/// Preview-relevant fields from one native document node. +#[derive(Clone, Debug, Default, PartialEq, Deserialize)] +pub struct PreviewNode { + #[serde(default)] + pub background: Option, +} + +/// The supported native document kinds as stored on disk. +#[derive(Clone, Copy, Debug, Default, PartialEq, Eq, Deserialize)] +#[serde(rename_all = "camelCase")] +pub enum PreviewKind { + #[default] + Design, + Board, +} + +/// Decode the preview-relevant subset of the native FigCraft document schema. +pub fn read_preview_document(bytes: &[u8]) -> Result { + serde_json::from_slice(bytes) +} + +/// Largest `thumbnail.png` returned for a recent-file preview. +pub const MAX_PREVIEW_THUMBNAIL_BYTES: u64 = 8 * 1024 * 1024; +/// Largest `document.json` read to build a recent-file preview. +pub const MAX_PREVIEW_DOCUMENT_BYTES: u64 = 64 * 1024 * 1024; + +/// Native `.figcraft` members needed by the recent-file preview. +#[derive(Clone, Debug, Default, PartialEq)] +pub struct PreviewArchive { + /// Whether the archive has a `document.json` entry, even if it is damaged or oversized. + /// Consumers use this to distinguish a malformed native archive from other ZIP formats. + pub has_document: bool, + /// The thumbnail bytes, absent when missing, damaged, or above [`MAX_PREVIEW_THUMBNAIL_BYTES`]. + pub thumbnail_png: Option>, + /// The first-page fields, absent when `document.json` is missing, damaged, or oversized. + pub document: Option, +} + +/// Read only the thumbnail member for a native archive-list thumbnail. +/// +/// This avoids unpacking the document when a caller needs only the image. Missing, damaged, +/// and oversized entries return None; an unreadable archive is an error. +pub fn read_preview_thumbnail(reader: R) -> Result>, FormatError> { + let mut archive = SeekArchive::open(reader, u64::MAX).map_err(|error| FormatError::Corrupt(error.to_string()))?; + Ok(archive.read("thumbnail.png", MAX_PREVIEW_THUMBNAIL_BYTES).ok().flatten()) +} + +/// Read the bounded native preview members of a `.figcraft` archive. +/// +/// An unreadable archive is an error; a missing, damaged, or oversized member is represented by +/// `None`, so the caller can still show the other part of the preview. The thumbnail and document +/// retain the separate limits used by the app's recent-file preview. +pub fn read_preview_archive(reader: R) -> Result { + let mut archive = SeekArchive::open(reader, u64::MAX).map_err(|error| FormatError::Corrupt(error.to_string()))?; + Ok(read_preview_archive_from(&mut archive)) +} + +/// Read bounded native preview members from an already-open shared archive reader. +/// +/// This lets callers preserve fallback handling for other ZIP formats without implementing ZIP +/// member access themselves. +pub fn read_preview_archive_from(archive: &mut impl NativeArchive) -> PreviewArchive { + let has_document = archive.contains("document.json"); + let thumbnail_png = archive.read_member("thumbnail.png", MAX_PREVIEW_THUMBNAIL_BYTES).ok().flatten(); + let document = if has_document { + archive.read_member("document.json", MAX_PREVIEW_DOCUMENT_BYTES).ok().flatten().and_then(|bytes| read_preview_document(&bytes).ok()) + } else { + None + }; + PreviewArchive { has_document, thumbnail_png, document } +} + +pub fn check_version(h: &Header) -> Result<(), NewerVersion> { + if let Some(v) = h.version + && v > u64::from(VERSION) + { + return Err(NewerVersion(u32::try_from(v).unwrap_or(u32::MAX))); + } + Ok(()) +} + +#[derive(Clone, Debug, PartialEq, Eq, Serialize)] +pub struct ImageIndexEntry { + pub hash: String, + pub width: u32, + pub height: u32, +} + +pub fn write_document_json(doc: &D, images: &[ImageIndexEntry], w: W) -> Result<(), serde_json::Error> { + #[derive(Serialize)] + struct Flat<'a, D: ?Sized> { + format: &'static str, + version: u32, + #[serde(rename = "imageIndex")] + images: &'a [ImageIndexEntry], + #[serde(flatten)] + document: &'a D, + } + Flat { format: FORMAT, version: VERSION, images, document: doc }.serialize(&mut serde_json::Serializer::new(w)) +} + +#[cfg(test)] +mod tests { + use super::*; + + #[test] + fn preview_archive_reads_native_members_and_preserves_document_presence() { + let document = br#"{"pages":[7],"nodes":{"7":{"background":{"r":0.25}}},"kind":"board"}"#; + let bytes = write_archive([ + Member { name: "thumbnail.png", bytes: b"png bytes", compression: Compression::Stored }, + Member { name: "document.json", bytes: document, compression: Compression::Deflated }, + ]) + .unwrap(); + assert_eq!(read_preview_thumbnail(std::io::Cursor::new(bytes.clone())).unwrap().as_deref(), Some(b"png bytes".as_slice())); + let mut archive = SeekArchive::open(std::io::Cursor::new(bytes.clone()), u64::MAX).unwrap(); + assert!(read_preview_archive_from(&mut archive).has_document); + let preview = read_preview_archive(std::io::Cursor::new(bytes)).unwrap(); + assert!(preview.has_document); + assert_eq!(preview.thumbnail_png.as_deref(), Some(b"png bytes".as_slice())); + let document = preview.document.unwrap(); + assert_eq!(document.pages, [7]); + assert_eq!(document.kind, PreviewKind::Board); + assert_eq!(document.nodes["7"].background.as_ref().unwrap()["r"], 0.25); + + // A malformed native document still identifies this as a FigCraft archive, allowing the + // app to retain its existing fallback instead of interpreting it as another ZIP format. + let bytes = write_archive([Member { name: "document.json", bytes: b"{", compression: Compression::Stored }]).unwrap(); + let preview = read_preview_archive(std::io::Cursor::new(bytes)).unwrap(); + assert!(preview.has_document); + assert!(preview.document.is_none()); + assert!(preview.thumbnail_png.is_none()); + assert!(read_preview_archive(std::io::Cursor::new(b"not a zip".as_slice())).is_err()); + } + + #[test] + fn preview_archive_keeps_thumbnail_and_document_size_limits_independent() { + let too_large_thumbnail = vec![0; MAX_PREVIEW_THUMBNAIL_BYTES as usize + 1]; + let bytes = write_archive([ + Member { name: "thumbnail.png", bytes: &too_large_thumbnail, compression: Compression::Deflated }, + Member { name: "document.json", bytes: br#"{"pages":[3]}"#, compression: Compression::Stored }, + ]) + .unwrap(); + let preview = read_preview_archive(std::io::Cursor::new(bytes)).unwrap(); + assert!(preview.thumbnail_png.is_none()); + assert_eq!(preview.document.unwrap().pages, [3]); + + let too_large_document = vec![b' '; MAX_PREVIEW_DOCUMENT_BYTES as usize + 1]; + let bytes = write_archive([Member { name: "document.json", bytes: &too_large_document, compression: Compression::Deflated }]).unwrap(); + let preview = read_preview_archive(std::io::Cursor::new(bytes)).unwrap(); + assert!(preview.has_document); + assert!(preview.document.is_none()); + } + + #[test] + fn header_skips_app_fields() { + let h = read_header(r#"{"version":1,"imageIndex":[{"hash":"abc"}],"nested":{"unused":true}}"#).unwrap(); + assert_eq!(h.version, Some(1)); + assert_eq!(h.images, Some(vec!["abc".into()])); + } + + #[test] + fn version_index_preserves_lenient_defaults_and_writes_compact_auto_flag() { + let index = read_version_index(br#"{"extra":true,"versions":[{"id":"v1","time":7,"auto":false,"images":["a",3]},{}]}"#).unwrap(); + assert_eq!(index.versions.len(), 2); + assert_eq!(index.versions[0].images, vec!["a"]); + assert_eq!(index.versions[1], VersionEntry::default()); + let encoded: Value = serde_json::from_slice(&write_version_index(&index).unwrap()).unwrap(); + assert!(encoded["versions"][0].get("auto").is_none()); + assert_eq!(encoded["versions"][0]["time"], 7); + } + + #[test] + fn branch_index_uses_native_camel_case_and_tolerates_invalid_optional_fields() { + let index = + read_branch_index(br#"{"branches":[{"id":"b1","client":4294967296,"state":"unknown","baseImages":["a",false],"headImages":["b"]},{}]}"#) + .unwrap(); + let first = &index.branches[0]; + assert_eq!(first.client, 0); + assert_eq!(first.state, BranchState::Active); + assert_eq!(first.base_images, vec!["a"]); + assert_eq!(first.head_images, vec!["b"]); + let encoded = serde_json::to_value(&index).unwrap(); + assert_eq!(encoded["branches"][0]["baseImages"][0], "a"); + assert!(encoded["branches"][0].get("base_images").is_none()); + } + + #[test] + fn document_limit_keeps_compressible_files_and_caps_expansion() { + const MB: u64 = 1024 * 1024; + assert_eq!(document_limit(0, 0), MAX_MEMBER); + assert_eq!(document_limit(MB, 2 * MB), MAX_MEMBER); + assert_eq!(document_limit(u64::MAX, MB), MAX_MEMBER); + assert_eq!(document_limit(MAX_DOCUMENT, MAX_DOCUMENT), MAX_DOCUMENT); + assert_eq!(document_limit(244_362_923, 260 * MB), 244_362_923 * DOCUMENT_RATIO); + } + + #[test] + fn read_budget_enforces_member_and_archive_totals() { + let bytes = write_archive([ + Member { name: "one", bytes: b"12345678", compression: Compression::Stored }, + Member { name: "two", bytes: b"abcdefgh", compression: Compression::Stored }, + ]) + .unwrap(); + let mut archive = Archive::open(&bytes, ArchiveLimits { max_entries: 4, max_total_bytes: 128 }).unwrap(); + let mut budget = ReadBudget::new(10); + assert_eq!(budget.read(&mut archive, "one", 16).unwrap(), b"12345678"); + assert!(budget.exceeds_remaining(&mut archive, "two")); + assert!(budget.exhausted_message().contains("contents together pass 10 bytes")); + assert!(budget.read(&mut archive, "two", 16).unwrap_err().to_string().contains("contents together pass 10 bytes")); + + let mut member_budget = ReadBudget::new(100); + assert!(member_budget.read(&mut archive, "two", 4).unwrap_err().to_string().contains("two is too large")); + assert_eq!(member_budget.read(&mut archive, "two", 16).unwrap(), b"abcdefgh"); + } + #[test] + fn newer_version_is_reported_before_document_decode() { + assert_eq!(check_version(&Header { version: Some(2), images: None }), Err(NewerVersion(2))); + assert_eq!(check_version(&Header { version: Some(u64::MAX), images: None }), Err(NewerVersion(u32::MAX))); + + #[derive(Deserialize)] + struct Payload { + document: u32, + } + let text = r#"{"version":2,"document":{"wrong":"type"}}"#; + assert!(matches!(read_document_json::(text), Err(ReadDocumentError::NewerVersion(NewerVersion(2))))); + let (payload, header) = read_document_json::(r#"{"version":1,"imageIndex":[{"hash":"a"}],"document":5}"#).unwrap(); + assert_eq!(payload.document, 5); + assert_eq!(header.images, Some(vec!["a".into()])); + } + #[test] + fn writer_flattens_document_fields() { + let doc = serde_json::json!({"name":"example"}); + let mut bytes = Vec::new(); + write_document_json(&doc, &[], &mut bytes).unwrap(); + let v: Value = serde_json::from_slice(&bytes).unwrap(); + assert_eq!(v["format"], FORMAT); + assert_eq!(v["version"], VERSION); + assert_eq!(v["name"], "example"); + } + + #[test] + fn preview_document_reads_native_page_background_and_kind() { + let doc = read_preview_document(br#"{"pages":[7],"nodes":{"7":{"background":{"r":0.1}}},"kind":"board","unrelated":true}"#).unwrap(); + assert_eq!(doc.pages, vec![7]); + assert_eq!(doc.nodes["7"].background.as_ref().and_then(|v| v.get("r")).and_then(Value::as_f64), Some(0.1)); + assert_eq!(doc.kind, PreviewKind::Board); + assert_eq!(read_preview_document(b"{}").unwrap(), PreviewDocument::default()); + assert!(read_preview_document(br#"{"kind":"unknown"}"#).is_err()); + } +} + +/// Limits for the native ZIP package reader. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct ArchiveLimits { + pub max_entries: u64, + pub max_total_bytes: u64, +} + +/// A bounded reader for a native `.figcraft` ZIP package. +pub struct Archive<'a> { + inner: craft_zip::Archive<'a>, +} + +impl<'a> Archive<'a> { + pub fn open(bytes: &'a [u8], limits: ArchiveLimits) -> Result { + let inner = craft_zip::Archive::open(bytes, craft_zip::Limits { max_entries: limits.max_entries, max_total_bytes: limits.max_total_bytes })?; + Ok(Self { inner }) + } + + pub fn names(&self) -> impl Iterator> { + self.inner.names() + } + + pub fn has_duplicate_names(&self) -> bool { + self.inner.has_duplicate_names() + } + + pub fn contains(&self, name: &str) -> bool { + self.inner.contains(name) + } + + pub fn declared_size(&mut self, name: &str) -> Option { + self.inner.declared_size(name) + } + + pub fn compressed_size(&mut self, name: &str) -> Option { + self.inner.compressed_size(name) + } + + pub fn read(&mut self, name: &str, max_bytes: u64) -> Result>, craft_zip::Error> { + self.inner.read(name, max_bytes) + } + + pub fn budget_used(&self) -> u64 { + self.inner.budget().used() + } +} + +/// One member to write into a native ZIP package. +#[derive(Clone, Copy, Debug)] +pub struct Member<'a> { + pub name: &'a str, + pub bytes: &'a [u8], + pub compression: craft_zip::Compression, +} + +/// Write a deterministic native ZIP package using the shared ZIP writer. +pub fn write_archive<'a>(members: impl IntoIterator>) -> Result, craft_zip::Error> { + let mut writer = craft_zip::Writer::new(); + for member in members { + writer.add(member.name, member.bytes, member.compression)?; + } + writer.finish() +} + +#[cfg(test)] +mod archive_tests { + use super::*; + + #[test] + fn bounded_archive_reader_and_writer_round_trip_members() { + let bytes = + write_archive([Member { name: "document.json", bytes: br#"{"version":1}"#, compression: craft_zip::Compression::Deflated }]).unwrap(); + let mut archive = Archive::open(&bytes, ArchiveLimits { max_entries: 4, max_total_bytes: 128 }).unwrap(); + assert!(archive.contains("document.json")); + assert_eq!(archive.read("document.json", 32).unwrap(), Some(br#"{"version":1}"#.to_vec())); + assert_eq!(archive.budget_used(), 13); + } + + #[test] + fn archive_reader_enforces_entry_and_total_limits() { + let bytes = write_archive([Member { name: "document.json", bytes: b"123456789", compression: craft_zip::Compression::Deflated }]).unwrap(); + let mut small_entry = Archive::open(&bytes, ArchiveLimits { max_entries: 4, max_total_bytes: 100 }).unwrap(); + assert!(small_entry.read("document.json", 8).is_err()); + let mut small_total = Archive::open(&bytes, ArchiveLimits { max_entries: 4, max_total_bytes: 8 }).unwrap(); + assert!(small_total.read("document.json", 100).is_err()); + } +} + +/// A native archive writer that keeps the streaming document serializer out of the consumer. +pub struct ArchiveWriter { + inner: craft_zip::Writer, +} + +impl ArchiveWriter { + pub fn new() -> Self { + Self { inner: craft_zip::Writer::new() } + } + + pub fn add(&mut self, name: &str, bytes: &[u8], compression: craft_zip::Compression) -> Result<(), craft_zip::Error> { + self.inner.add(name, bytes, compression) + } + + pub fn add_document_json(&mut self, name: &str, document: &D, images: &[ImageIndexEntry]) -> Result<(), ArchiveWriteError> { + self.inner + .add_with(name, craft_zip::Compression::Level(3), |w| write_document_json(document, images, w).map_err(std::io::Error::other)) + .map_err(ArchiveWriteError::Archive) + } + + pub fn finish(self) -> Result, craft_zip::Error> { + self.inner.finish() + } +} + +impl Default for ArchiveWriter { + fn default() -> Self { + Self::new() + } +} + +#[derive(Debug, thiserror::Error)] +pub enum ArchiveWriteError { + #[error("could not serialize native document JSON: {0}")] + Json(#[from] serde_json::Error), + #[error("could not write native archive: {0}")] + Archive(#[from] craft_zip::Error), +} + +#[cfg(test)] +mod streaming_writer_tests { + use super::*; + + #[test] + fn native_writer_streams_flat_document_into_archive() { + let document = serde_json::json!({"name":"streamed"}); + let mut writer = ArchiveWriter::new(); + writer.add_document_json("document.json", &document, &[]).unwrap(); + let bytes = writer.finish().unwrap(); + let mut archive = Archive::open(&bytes, ArchiveLimits { max_entries: 4, max_total_bytes: 4096 }).unwrap(); + let json = archive.read("document.json", 1024).unwrap().unwrap(); + let value: Value = serde_json::from_slice(&json).unwrap(); + assert_eq!(value["name"], "streamed"); + assert_eq!(value["format"], FORMAT); + } +} + +/// ZIP compression modes for native archive members. +pub use craft_zip::Compression; + +/// A native ZIP reader for seekable sources that should not be copied into memory first. +pub struct SeekArchive { + inner: zip::ZipArchive, +} + +impl SeekArchive { + pub fn open(reader: R, max_entries: u64) -> Result { + let inner = zip::ZipArchive::new(reader).map_err(|e| craft_zip::Error::NotZip(e.to_string()))?; + let count = inner.len() as u64; + if count > max_entries { + return Err(craft_zip::Error::TooManyEntries { declared: count, limit: max_entries }); + } + Ok(Self { inner }) + } + + pub fn contains(&self, name: &str) -> bool { + self.inner.index_for_name(name).is_some() + } + + pub fn names(&self) -> Vec { + self.inner.file_names().filter_map(Result::ok).map(|n| n.into_owned()).collect() + } + + pub fn declared_size(&mut self, name: &str) -> Option { + self.inner.index_for_name(name).and_then(|i| self.inner.by_index_raw(i).ok().map(|f| f.size())) + } + + pub fn read(&mut self, name: &str, max_bytes: u64) -> Result>, craft_zip::Error> { + let Some(index) = self.inner.index_for_name(name) else { return Ok(None) }; + let mut file = self.inner.by_index(index).map_err(|e| craft_zip::Error::Entry { name: name.to_owned(), msg: e.to_string() })?; + if file.size() > max_bytes { + return Err(craft_zip::Error::EntryTooLarge { name: name.to_owned(), limit: max_bytes }); + } + let mut bytes = Vec::new(); + let mut limited = std::io::Read::take(&mut file, max_bytes.saturating_add(1)); + let read = std::io::Read::read_to_end(&mut limited, &mut bytes); + if let Err(error) = read { + return Err(craft_zip::Error::Entry { name: name.to_owned(), msg: error.to_string() }); + } + if bytes.len() as u64 > max_bytes { + return Err(craft_zip::Error::EntryTooLarge { name: name.to_owned(), limit: max_bytes }); + } + Ok(Some(bytes)) + } +} + +/// Shared access contract implemented by in-memory and seekable native archive readers. +pub trait NativeArchive { + fn contains(&self, name: &str) -> bool; + fn names(&self) -> Vec; + fn declared_size(&mut self, name: &str) -> Option; + fn compressed_size(&mut self, name: &str) -> Option; + fn read_member(&mut self, name: &str, max_bytes: u64) -> Result>, craft_zip::Error>; +} + +impl NativeArchive for Archive<'_> { + fn contains(&self, name: &str) -> bool { + Archive::contains(self, name) + } + fn names(&self) -> Vec { + Archive::names(self).map(|n| n.into_owned()).collect() + } + fn declared_size(&mut self, name: &str) -> Option { + Archive::declared_size(self, name) + } + fn compressed_size(&mut self, name: &str) -> Option { + Archive::compressed_size(self, name) + } + fn read_member(&mut self, name: &str, max_bytes: u64) -> Result>, craft_zip::Error> { + Archive::read(self, name, max_bytes) + } +} + +impl NativeArchive for SeekArchive { + fn contains(&self, name: &str) -> bool { + SeekArchive::contains(self, name) + } + fn names(&self) -> Vec { + SeekArchive::names(self) + } + fn declared_size(&mut self, name: &str) -> Option { + SeekArchive::declared_size(self, name) + } + fn compressed_size(&mut self, name: &str) -> Option { + self.inner.index_for_name(name).and_then(|i| self.inner.by_index_raw(i).ok().map(|f| f.compressed_size())) + } + fn read_member(&mut self, name: &str, max_bytes: u64) -> Result>, craft_zip::Error> { + SeekArchive::read(self, name, max_bytes) + } +} + +#[cfg(test)] +mod seek_archive_tests { + use super::*; + + #[test] + fn seekable_reader_matches_bounded_in_memory_reader() { + let bytes = write_archive([Member { name: "document.json", bytes: b"payload", compression: Compression::Stored }]).unwrap(); + let mut reader = SeekArchive::open(std::io::Cursor::new(bytes), 4).unwrap(); + assert!(reader.contains("document.json")); + assert_eq!(reader.read_member("document.json", 16).unwrap(), Some(b"payload".to_vec())); + assert_eq!(reader.declared_size("document.json"), Some(7)); + assert!(reader.read_member("document.json", 3).is_err()); + } +} + +/// Largest individual ZIP member read by the native format. +pub const MAX_MEMBER: u64 = 512 * 1024 * 1024; +/// The document expansion ratio that its compressed size may claim. +pub const DOCUMENT_RATIO: u64 = 16; +#[cfg(not(target_pointer_width = "32"))] +pub const MAX_DOCUMENT: u64 = 4 * 1024 * 1024 * 1024; +#[cfg(target_pointer_width = "32")] +pub const MAX_DOCUMENT: u64 = 1024 * 1024 * 1024; + +/// Upper bound for the expanded document member based on both the archive and member policy. +/// The floor preserves support for large but highly compressible documents; the ceiling prevents +/// a tiny archive from claiming an unbounded expansion. +pub fn document_limit(packed: u64, archive_bytes: u64) -> u64 { + packed.min(archive_bytes).saturating_mul(DOCUMENT_RATIO).clamp(MAX_MEMBER, MAX_DOCUMENT) +} +/// Most version snapshots kept on save and read on open. +pub const MAX_VERSIONS: usize = 100; +/// Maximum version and branch name lengths, in Unicode scalar values. +pub const MAX_VERSION_NAME: usize = 200; +pub const MAX_VERSION_DESCRIPTION: usize = 2000; +/// Maximum bytes of an index and all version snapshots read. +pub const MAX_VERSION_INDEX: u64 = 16 * 1024 * 1024; +pub const MAX_VERSION_BYTES: u64 = 1024 * 1024 * 1024; +/// Maximum images read and maximum combined image bytes. +#[derive(Clone, Copy, Debug, PartialEq, Eq)] +pub struct ImageBudget { + pub images: usize, + pub bytes: u64, +} +pub const IMAGE_BUDGET: ImageBudget = ImageBudget { images: 20_000, bytes: 4 * 1024 * 1024 * 1024 }; +/// Maximum branch histories kept. +pub const MAX_BRANCHES: usize = 100; +/// The branch index member. +pub const BRANCH_INDEX: &str = "branches/index.json"; +/// Maximum uncompressed bytes read from one native archive in total. +#[cfg(not(target_pointer_width = "32"))] +pub const MAX_UNPACKED: u64 = 6 * 1024 * 1024 * 1024; +#[cfg(target_pointer_width = "32")] +pub const MAX_UNPACKED: u64 = 2 * 1024 * 1024 * 1024; + +/// Native-format errors shared with the consumer's document mapping. +#[derive(Debug, PartialEq, Eq)] +pub enum FormatError { + Io(String), + Corrupt(String), + Newer(u32), +} + +impl std::fmt::Display for FormatError { + fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { + match self { + Self::Io(e) => write!(f, "{e}"), + Self::Corrupt(e) => write!(f, "not a valid FigCraft file: {e}"), + Self::Newer(v) => write!(f, "this file was saved by a newer FigCraft (format version {v})"), + } + } +} + +impl std::error::Error for FormatError {} + +/// Tracks the total bytes unpacked from one native archive across its document, images, library, +/// history, and branch members. +#[derive(Clone, Debug)] +pub struct ReadBudget { + left: u64, + total: u64, +} + +impl ReadBudget { + pub fn new(total: u64) -> Self { + Self { left: total, total } + } + + fn passed(&self) -> String { + match self.total / (1024 * 1024) { + 0 => format!("too large: the file's contents together pass {} bytes", self.total), + mb => format!("too large: the file's contents together pass {mb} MB"), + } + } + + /// Whether the member's declared uncompressed size exceeds what remains in this budget. + pub fn exceeds_remaining(&self, archive: &mut impl NativeArchive, name: &str) -> bool { + archive.declared_size(name).is_some_and(|size| size > self.left) + } + + /// Human-readable reason for leaving out a member after the archive-wide limit is spent. + pub fn exhausted_message(&self) -> String { + self.passed() + } + + /// Read one native member within its own size limit and the remaining archive-wide budget. + /// A failed read that consumed a member is charged at its declared size. + pub fn read(&mut self, archive: &mut impl NativeArchive, name: &str, limit: u64) -> Result, FormatError> { + let declared = archive.declared_size(name); + if declared.is_some_and(|size| size > self.left && size <= limit) { + return Err(FormatError::Corrupt(format!("{name}: {}", self.passed()))); + } + let cap = limit.min(self.left); + let size = declared.unwrap_or(0); + let result = match archive.read_member(name, cap) { + Ok(Some(bytes)) => Ok(bytes), + Ok(None) => Err(FormatError::Corrupt(format!("{name}: not found"))), + Err(error) if error.is_limit() => Err(FormatError::Corrupt(format!("{name} is too large"))), + Err(error) => Err(FormatError::Corrupt(format!("{name}: {error}"))), + }; + if size <= cap { + self.left = self.left.saturating_sub(result.as_ref().map_or(size, |bytes| bytes.len() as u64)); + } + result + } +} + +/// Whether `id` can name a version or branch: 1–64 ASCII letters, digits, `-` and `_`. +pub fn is_valid_version_id(id: &str) -> bool { + (1..=64).contains(&id.len()) && id.bytes().all(|b| b.is_ascii_alphanumeric() || b == b'-' || b == b'_') +} + +/// Clip text to at most `max` Unicode scalar values. +pub fn clip(s: &str, max: usize) -> String { + s.chars().take(max).collect() +} diff --git a/crates/figcraft-format/src/library.rs b/crates/figcraft-format/src/library.rs new file mode 100644 index 000000000..7cf6b5b2c --- /dev/null +++ b/crates/figcraft-format/src/library.rs @@ -0,0 +1,205 @@ +//! Wire schema and member limits for published libraries embedded in a native project archive. +use serde::{Deserialize, Serialize}; +use std::collections::{BTreeMap, BTreeSet}; + +pub const MANIFEST: &str = "library/manifest.json"; +pub const DOCUMENT: &str = "library/document.json"; +pub const IMAGES_PREFIX: &str = "library/images/"; +pub const MAX_MANIFEST_BYTES: u64 = 16 * 1024 * 1024; +pub const MAX_IMAGE_BYTES: u64 = 1024 * 1024 * 1024; +pub const MAX_ID_CHARS: usize = 64; +pub const MAX_NAME_CHARS: usize = 200; +pub const MAX_DESCRIPTION_CHARS: usize = 2000; + +/// Native library manifest, independent of the consumer's document model. +#[derive(Clone, Debug, Default, PartialEq, Serialize, Deserialize)] +#[serde(default)] +pub struct Manifest { + pub id: String, + pub name: String, + pub version: u64, + pub description: String, + pub time: u64, + pub components: Vec, + pub styles: Vec