* refactor(cdr): depend on shared craft-cdr from craft-libs (#1665)
Replaces the local crates/cdr (vectorcraft-cdr) crate with craft-cdr
from storytold/craft-libs#2697 (storytold/craft-libs#778), keeping
VectorCraft's document/artboard/layer/shape mapping in crates/engine.
Co-authored-by: dotCooCoo <RobertLeeLW@gmail.com>
* build(cdr): pin craft-cdr to craft-libs main (75d179f8df)
* build(cdr): pin craft-cdr to craft-libs main (9b28b56bc7)
* docs(cdr): point CDR/CMX provenance and test paths to craft-cdr
---------
Co-authored-by: dotCooCoo <RobertLeeLW@gmail.com>
Plugin::load now validates each module with wasmparser and a fixed list
of WebAssembly features before wasmi reads it: mutable globals, sign
extension, saturating conversions, multi-value, bulk memory, reference
types, tail calls, extended constant expressions and floats. A module
that uses any other feature is refused, so a wasmi upgrade, a change to
its defaults or one of its crate features turned on elsewhere in the
build does not change what a plug-in may use. wasmi's Config is set to
the same list for every setter it has in this build.
The list is the set wasmi accepted before, without multi-memory. A
second memory was already refused by the one-memory limit; its error
now says "multiple memories". No module that loaded before is refused.
wasmparser 0.228.0 is the version wasmi 2.0.0 already uses.
A new test loads a plug-in that uses each listed feature and checks
that modules using SIMD, relaxed SIMD, memory64, a second memory,
custom page sizes, wide arithmetic, threads, exceptions, legacy
exceptions, function references or garbage collection are refused with
an error that names the feature.
docs/plugins.md lists the whole set; it named only four of the allowed
features before.
The mesh sentence in docs/mcp.md no longer lists the .ai writers among
the exports that draw a mesh. Both docs now say the fallback applies to
an image nested 8 deep, name the containers each writer counts (ordinary
groups and layers do not count), and say an image nested deeper is not
written. The Perspective Distort line cites the two nesting tests.
The mesh paragraph in docs/mcp.md named DXF, WMF and HP-GL as the only
exports that do not draw a distorted image as a mesh. The editing data
of a .ai file writes an image inside 8 or more nested opacity masks,
symbols, live objects and pattern tiles as the parallelogram nearest its
corners, and a legacy .ai (version 7) does so inside 8 or more nested
symbols and live objects, each with a warning. The paragraph and the
Perspective Distort line in docs/target-app-parity.md now say so.
Part of #1210.
- `cargo xtask corpus --scripts` fetches 397 scripts and 1 included file
from 19 MIT and Apache-2.0 repositories at pinned commits into
corpus/scripts, each repository's license file beside its scripts.
xtask/script-corpus.txt lists every file with its size and SHA-256; a
download replaces the previous copy only when both match. One more
candidate is listed as excluded because it is over the 1 MiB limit for
one script. The fetch caps 64 MiB of scripts and 128 MiB of transfer,
refuses `..`, absolute paths, reserved names, case-fold collisions and
links, and holds a lock file while it runs.
- `cargo xtask corpus` takes at most one of --download, --affinity and
--scripts, and refuses other arguments with the usage text.
- `cargo xtask scripts --pass-rate [--report <path>]` runs each script in
a process of its own (the ignored `row` test in
crates/engine/tests/script_corpus.rs): a new document, 5,000,000
instructions, no dialog answers, file reads only in the script's
repository folder and a new folder of its own, file writes only in that
folder, stopped after 10 seconds. It writes a JSON report with the pass
rate overall, per repository and for scripts without dialogs or
windows, how many parse, and each script's outcome, with local folders
removed from messages. Script failures never fail the command, and
without the corpus it prints a skip line and exits 0.
Since #8289 a saved .ai's editing copy keeps live the blends the .ai
reader opens with their settings and as they are drawn. The three File
formats rows in docs/target-app-parity.md still listed every live
object other than four graph types as read back as art. They now name
those blends as kept live too. Blends written as their steps stay in
the gap, and file-format-parity.md lists them.
* Text styles panel: add item right-click context menu (Options, Duplicate, Delete, Redefine)
* fix(text-styles): require text target for clear overrides and update parity docs
* Properties panel: add contextual Quick Actions for text (Create Outlines, Retype, Find and Replace)
* Properties panel: address review comments on Retype reveal idempotency, test assertions and parity docs
- Forward secondary clicks on name label links for stroke/fill and effect rows
- Enable Reduce to Basic Appearance when any selected subject needs reduction (including compound path children)
- Add headless regression tests for label context clicks and multi-selection reduction
- Update M14 roadmap entry and UI parity evidence, and clean conflict markers in ui-parity.md
* Adopt the shared AGENTS.md (RFC-0072) and test rules (RFC-0071)
* Drop a restated rule from This repository
* Remove the cleanroom xtask command, which does not exist
* AGENTS.md: correct the repository facts
The desktop app's platform modules allow unsafe; list them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: FredApps <FredApps@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: FredApps <153173854+FredApps@users.noreply.github.com>
The monitor list for floating panels now carries each monitor's work
area: rcWork from GetMonitorInfo on Windows, and visibleFrame on macOS,
turned into the top-left origin and physical pixels of the monitor's
rect. The work area is converted to points with the same scale and zoom
as the rect, and craft-ui uses it to decide whether a reopened panel's
title bar can be grabbed and where to move it. A work area with no width
or height is left out.
Restoring the main window now keeps it inside its monitor's work area
too, so a window saved at the top edge of a monitor with a taskbar along
its top opens below the taskbar, and a window taller or wider than the
work area is made to fit it, with room for a system title bar and
border when the window has them.
On Windows the panel list stays empty when the monitors have different
scale factors, as before, and a build without the wgpu feature reads no
work areas. Linux and the BSDs keep the whole monitor rect.