Commit Graph
22167 Commits
Author SHA1 Message Date
Behdad Esfahbod 340aec5f77 [subset] Check instanced GPOS value ranges
Adding a variation delta directly to HBINT16 silently wraps values
outside the signed 16-bit range. Use the serializer's checked assignment
with a wide sum, as other layout coordinates already do.

Cover all four value fields, both signs, valid boundaries, partial
instancing, and both axis orders with a reproducible two-axis font.
All 255 configured tests pass; the new test fails with the old library.

Assisted-by: Codex
2026-10-10 22:16:15 -06:00
Behdad Esfahbod 3f60c824e6 [math] Preserve and apply shared GDEF variation indices
MATH devices were evaluated against an empty variation store, and the
subsetter dropped them without folding deltas into the base values.
MATH users were also omitted when pruning GDEF's shared store.

Evaluate MATH VariationIndex devices with GDEF's store. Collect indices
from retained constants, glyph information, and assemblies, then fold
pinned deltas and remap residual devices during subsetting. Preserve
ordinary ppem device corrections and reject unrepresentable values.

Add a reproducible two-axis fixture covering every MathValueRecord
location, axis orders, full and partial instances, and glyph pruning.
All 255 configured tests pass; reduced-feature syntax checks pass.

Assisted-by: Codex
2026-10-10 21:54:15 -06:00
Behdad Esfahbod 5eb8323371 [ot] Evaluate constant item variation regions at default coordinates
A zero-peak region ignores every axis and can contribute a nonzero
variation delta even when coordinates are absent or zero. Default
shortcuts skipped these deltas in HVAR/VVAR advances, vertical origins,
GPOS value records, anchors, and generic VariationDevice access.

Cache whether metrics and GDEF stores have regions active at the default
when their accelerators load. Preserve existing fast paths for ordinary
stores and evaluate constant-region stores at default coordinates.

Add a generated CFF2 font exercising constant and ordinary regions,
empty and short coordinate arrays, and returning to default after
non-default shaping. Skera and FontTools full instances give the same
metrics and positioning. All 254 configured Meson tests pass; core files
also compile with variation, vertical, and layout support disabled.

Assisted-by: Codex
2026-10-10 20:41:56 -06:00
Behdad Esfahbod d89fa94461 [subset] Preserve distinct feature parameters and variations (#6350)
Matching lookup lists alone cannot establish that same-tag features are
duplicates. Different feature parameters carry distinct UI metadata, and
alternate features can have different behavior in each language system.

Compare complete supported parameter data before merging, and preserve
indices referenced by retained feature variations. Fully pinned fonts can
still merge features after resolving their alternates.

Add GSUB/GPOS regressions with differing cv01 character arrays, size
values, ss01 names, and ss02 alternatives, plus identical ss03 controls.
The full Meson suite passes, as do reduced-feature compilation checks.

Assisted-by: Codex
2026-10-10 17:07:29 -06:00
Behdad Esfahbod d45fa42c5a [vector] Restore reset defaults and clear PDF draw resources (#6355)
Vector reset omitted draw colors, paint background, SVG id prefixes,
and custom palette overrides. Restore these defaults while preserving
configuration across clear.

The PDF draw context also kept its opacity resource dictionary and
counter after clear, including automatic clearing after render.
Clear both with the accumulated output so later documents do not carry
resources from earlier renders.

Add regressions for SVG/PDF reset, configuration preservation, custom
palette overrides, and fresh-context PDF equivalence after explicit
clear, reset, and repeated rendering. All three regression groups fail
against the original implementation.

Tested with meson test -C build: 288 passed, 1 skipped, 0 failed.
The vector suite and the issue reproducer pass. The regression also
compiles and runs as C99 with -Wall -Wextra -Werror.

Fixes: https://github.com/harfbuzz/harfbuzz/issues/6342
Assisted-by: OpenAI Codex
2026-10-10 16:59:13 -06:00
Behdad Esfahbod 199cae8b28 [buffer] Document message callback lifetime across reset (#6354)
Buffer reset has retained message callbacks since their introduction,
although its documentation promised a freshly created buffer state.
Clarify that callbacks, their associated data, and attached user data
survive reset. Document explicit callback removal and destruction.

Add a regression test for callback-directed lookup skipping after
clear/reset, restoration of ligatures after removal, and exactly-once
destruction on replacement, removal, and final buffer destruction.

Tested with meson test -C build: 287 passed, 1 skipped, 0 failed.
The API test also compiles without buffer messages or OT shaping.

Fixes: https://github.com/harfbuzz/harfbuzz/issues/6341
Assisted-by: OpenAI Codex
2026-10-10 16:58:18 -06:00
Behdad Esfahbod 86c92a4735 [buffer] Clarify what clear_contents preserves (#6353)
The documentation listed only two retained settings, although
clear_contents also preserves flags, cluster level, invisible and
not-found glyphs, the message callback, and attached user data.
Document these properties and the random-state reset without changing
existing behavior.

Tested with meson test -C build: 287 passed, 1 skipped, 0 failed.

Fixes: https://github.com/harfbuzz/harfbuzz/issues/6340
Assisted-by: OpenAI Codex
2026-10-10 16:57:40 -06:00
Behdad Esfahbod 818eea711c [algs] Preserve float precision when rounding (#6351)
Adding 0.5 in float can round up before floor sees the value. This
turns the predecessor of 0.5 into 1, and changes sufficiently large
odd integral floats by one. The latter also shifts exactly representable
16.16 axis bounds when the subsetter writes fvar.

Add 0.5 in double to preserve the input's fractional bits while retaining
the existing tie direction toward positive infinity. Exercise half ties,
their neighboring floats, and large positive and negative integers.

Validation: the new regression fails before the fix; the full Meson suite
passes all 282 configured tests after it.

Assisted-by: Codex
2026-10-10 12:08:50 -06:00
ebraminio 8aa60f12a9 [number] Deduplicate the logic around fraction (#6352)
* [number] Don't mutate input pointer on failure

Two last tests in test_parse_double used to fail but now don't.

* [number] Deduplicate the logic around fraction

Forming a new abstraction, _hb_fraction, just to consume the fraction part.
2026-10-10 12:05:00 -06:00
Behdad Esfahbod fd0c6c9a69 [subset] Preserve parameters and names in alternate features (#6349)
Alternate feature serialization lacks the source feature tag, so it
cannot identify and retain cvXX, ssXX, or size parameters. Name-ID
closure also visits only the original base feature, losing labels
referenced by retained alternates or fully instantiated substitutions.

Carry source tags for retained features after deduplication. Collect
name IDs from the current base feature, retained variation records,
and original features used by catch-all records.

Add ordinary, full, partial, and range regressions for all three
parameter formats in GSUB and GPOS. Ten of twelve regression subtests
fail before the fix. The complete Meson suite passes: 284 tests.
The changed layout units also compile with HB_NO_VAR.

Assisted-by: Codex
2026-10-10 04:12:54 -06:00
Behdad Esfahbod 1bf17d7295 [subset] Allow instancing axes unrelated to VARC (#6265)
Replace the blanket VARC instancing rejection with validation of the
axes referenced by retained components, variation regions, and nested
conditions. Remap those indices when unrelated axes are removed, while
keeping passthrough VARC incompatible with instancing.

Preserve component-internal variation data during avar2 instancing by
avoiding culling based on font-level reachable ranges. Add drawing
comparisons and rejection tests with reproducible fixtures, including
hidden component axes and non-default unrelated-axis pins.

Serialize sparse-region axis references using the revised VARC format.
Regenerate the fixtures without changing their logical VARC contents or
other table bytes, apart from the head checksum adjustment.

Validation: meson test -C build (283 passed); subset suite (130 passed);
C++11 compile checks with HB_NO_VAR and HB_NO_VAR_COMPOSITES.

Assisted-by: Codex
2026-10-10 04:04:32 -06:00
Behdad Esfahbod a2ccf95182 [subset] Preserve IFTB offsets when downgrading CFF2 (#6348)
Pass the requested minimum CharStrings offset size to the CFF1 INDEX
serializer, matching the CFF2 path. The conversion previously emitted
short offsets even with HB_SUBSET_FLAGS_IFTB_REQUIREMENTS enabled.

Add a Cantarell downgrade regression, registered with the existing
IFTB tests only when experimental API support is enabled. The original
subsetter fails it with and without preprocessing.

Full Meson suites pass with experimental API enabled (282 tests) and
disabled (263 tests, one skipped).

Assisted-by: Codex
2026-10-10 01:46:09 -06:00
Behdad Esfahbod 117264edde [subset] Preserve fractional CFF displacements during specialization (#6347)
Check the real value when identifying zero operands. Converting to an
integer truncates nonzero displacements smaller than one, allowing
h/v specialization to remove them and change glyph geometry.

Add move and line regressions covering positive and negative fractions
on either axis, alongside genuine zero operands. The regression fails
before the fix. The full Meson suite passes: 277 passed, one skipped.

Assisted-by: Codex
2026-10-10 01:23:34 -06:00
Behdad Esfahbod 3396dcb35b [subset] Keep original IUP contour points across serialization retries (#6345)
IUP adds rebased constant tuples to the plan's cached contour points.
When serialization needs a larger buffer, another gvar serialization
adds the same constants again. The final optimizer then chooses points
against a different outline than the one stored in glyf.

Copy the cached points before IUP instancing so every serialization
attempt starts from the original contour coordinates. Check allocation
failure through the usual vector error path.

Add a shifted-default Roboto Flex regression that fails on the original
subsetter. The entire configured Meson suite passes (278 tests, one
skipped).

Assisted-by: Codex
2026-10-10 01:13:14 -06:00
Behdad Esfahbod be6a74f303 [subset] Preserve line directions when specializing CFF charstrings (#6346)
hlineto and vlineto alternate directions for successive arguments.
Combining commands by their operator alone changes geometry: two
vertical segments become a vertical segment followed by a horizontal
one. Merging a mixed run also needs its argument-count parity.

Require odd parity before combining opposite starting directions and
even parity before combining identical h/v operators.

Test all 1,020 orientation sequences through eight segments at stack
limits 4 and 48, reconstructing every point after specialization. The
original specializer fails that test. Add a shifted Source Serif
instance and update its existing downgrade reference. The entire
configured Meson suite passes (277 tests, one skipped).

Assisted-by: Codex
2026-10-10 01:12:39 -06:00
Behdad Esfahbod 1e6ea99af7 [subset] Preserve deltas when a contour exceeds the IUP set limit (#6344)
The IUP optimizer returns success for contours larger than its 512-bit
sets can represent, leaving every optimized reference false. The
resulting partial instance loses the contour's variation data.

Keep all explicit references when skipping optimization at that limit.
This preserves the existing bounded search without changing geometry.

Add a 600-point contour regression. The original subsetter fails both
normal and preprocessed checks. The entire configured Meson suite
passes (278 tests, one skipped).

Assisted-by: Codex
2026-10-10 00:36:51 -06:00
Behdad Esfahbod 372119f1df [subset] Clamp fully instantiated COLRv1 alpha before narrowing (#6343)
Variation deltas can take solid and gradient stop alpha outside [0, 1].
Narrowing those values to F2DOT14 before clamping wraps alpha 2 to -2
and alpha -3 to 1, reversing the intended opacity.

Clamp full instances in float before storing alpha. Keep partial bases
unclamped because surviving axes can bring them back into range.

Add synthetic solid and gradient regressions for overflow, underflow,
and a partial base of 1.5 reduced to 0.5 by the retained axis. The old
subsetter fails four full-instance checks; the entire configured Meson
suite passes (278 tests, one skipped).

Assisted-by: Codex
2026-10-10 00:30:06 -06:00
Behdad Esfahbod c82300aefb Merge pull request #6339 from ebraminio/hb-number-copy-free
[number] Make integer parsings copy free
2026-10-09 14:27:49 -06:00
Behdad Esfahbod 78c25c84b4 Merge pull request #6338 from ebraminio/hb-number-ragel-free
[number] Make hb-number-parser ragel free
2026-10-09 14:27:38 -06:00
Behdad Esfahbod 56282e5b38 Merge pull request #6337 from ebraminio/hb-number
[number] Harden hb-number-parser implementation
2026-10-09 14:26:57 -06:00
Ebrahim 97513326fc [number][test] Add some more tests for the expected behavior 2026-10-09 16:17:08 +03:30
Ebrahim f8ba9744b1 [number] Make hb_parse_int and hb_parse_uint copy free
There is one known difference before and after the change that comes
from 32bit/64bit machines difference on long size which is documented
in the tests.
2026-10-09 15:49:50 +03:30
Ebrahim 30f2eae3c7 [number] Only alter value pointer if parsing was successful
Previously it was changing value's pointer even if the parsing
was declarad a failure, this fixes it by changing the value
only if the parsing was successful.
2026-10-09 15:02:35 +03:30
Ebrahim 1622d7cb09 [test][number] Expand the testing
This only adds new tests to the current implementation but changes nothing.
2026-10-09 15:02:31 +03:30
Ebrahim 9cf8b485c7 [subset] Avoid use of undefined behavior in calling hb_parse_double 2026-10-09 06:04:13 +03:30
Ebrahim 864683d182 [number] Make the number parser free from ragel 2026-10-09 04:52:28 +03:30
Ebrahim 3150953d5b ci: Fix 32-bit Windows build bot by accepting tolerance 2026-10-09 04:50:53 +03:30
Ebrahim 2855792914 [test] Minor reorg, separate parse int tests from parse double 2026-10-09 04:12:45 +03:30
Ebrahim 5d441040db [test] Use isinf and signbit from std::
Once the project was avoiding STL and everything related at all cost
but now use of std:: provided things exists on the project despite
still no runtime dependency to STL (which is fantastic) and
since this in tests and even this doesn't pull STL, use of std::
should be okay to resolve the issue some of the CI bots have with
accessing signbit and isinf (unlike my local machine which is ok).
2026-10-09 03:53:25 +03:30
Ebrahim d210429af2 [test][number] Import cmath to have access to isinf in different compilers
Apparently fails on some CI bots
2026-10-09 03:07:37 +03:30
Ebrahim 4ba6eca464 [number] When exponential overflow let's just return 0
Following what strtod does also.
2026-10-09 02:33:55 +03:30
Ebrahim f64ef34b6e [number] Add three more already passing tests 2026-10-09 02:29:07 +03:30
Ebrahim e9aebd3a23 [number] Make sure we match what a libc strtod does is small numbers 2026-10-09 02:03:49 +03:30
Ebrahim a7b63551cb [number] Reject all whitespace 2026-10-09 01:35:16 +03:30
Ebrahim 3b06f4957e [number] Add some more already passing tests 2026-10-09 01:22:30 +03:30
Ebrahim 8ec1db922b [number] Handle rejection of invalid input better
Previously it was reporting the far it could parse on such cases but
with this change it's able to revert to what it couldn't parse.
2026-10-09 01:03:07 +03:30
Ebrahim cc5fef8df1 [number] Round instead of truncating long fractional parts
Previously we considered 0.9999999999999999 as 0.9999999999999990
and this rounds when it's not possible to record more frac.
2026-10-08 23:32:29 +03:30
Ebrahim 6134871eee [number] Consider 1e512 as infinity and 1e-512 as zero 2026-10-08 23:32:24 +03:30
Behdad EsfahbodandClaude Opus 5.5 4834503edd [map] Fix use-after-free when setting a key/value from the map itself (#6334)
m.set (99, m.get (1)) passes a reference into the map's own items
array.  If the insertion needs to grow the table, alloc() frees that
array before the value is read, so set_with_hash() copies from freed
memory.  Same for the key, eg. m.set (m.get (1), v).

When growth is needed and the key or value lies within our storage,
copy them out first and insert the copies.  The copies are
default-constructed and then assigned, matching how items are filled,
since some key types (eg. OT::OffsetTo) are not copy-constructible.

Added test-map cases for value and key aliasing, including a
non-trivial value type.  Reproduced with ASan before the fix; test-map
is clean under ASan+UBSan after it, and meson test passes.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-07 13:06:48 -06:00
Behdad Esfahbod d997502000 [meson] Guard tests for disabled optional libraries (#6335)
* [meson] Skip vector tests when vector is disabled

The vector budget test was built unconditionally even when vector was
explicitly disabled. Its empty dependency supplied neither include paths
nor libraries, breaking otherwise valid reduced-feature builds.

Skip the vector test directory when the library is disabled. Add a Linux
CI build and test pass with vector disabled to cover this configuration.

Reproduced the missing hb.h failure on Linux before the fix. The full
suite with vector, raster, and GPU disabled passed 229 tests with one
skip; all three vector tests passed with vector enabled.

Fixes: https://github.com/harfbuzz/harfbuzz/issues/6277
Assisted-by: OpenAI Codex <codex@openai.com>

* [ci] Remove extra vector-disabled build pass

Keep the fix focused on guarding the vector test directory without adding
another build and test pass to the Linux workflow.

CI-only change; the previously validated Meson guard is unchanged.

Assisted-by: OpenAI Codex <codex@openai.com>

* [meson] Guard dependency fuzzers when subset is disabled

With experimental APIs enabled, dependency fuzzers referenced
libharfbuzz_subset even when the subset library was disabled, causing
Meson setup to fail with an unknown-variable error.

Require subset support when building the dependency fuzzers and when
registering their tests.

Full tests with experimental APIs enabled: 92 passed with subset
disabled, and 247 passed with subset enabled. Both dependency fuzzers
and their 13 tests remain registered in the enabled configuration.

Assisted-by: OpenAI Codex <codex@openai.com>
2026-10-07 12:55:53 -06:00
Behdad Esfahbod bf0828e9f4 [set] Restore sorted-array additions to inverted sets (#6333)
Adding a sorted array to an inverted set delegates to the underlying deletion path, but set_sorted_array() always added bits. Excluded members stayed excluded, and existing included members could become excluded.

Use the requested bit value, as set_array() already does. The regression covers duplicates, excluded and included members, and both existing and absent pages.

Assisted-by: OpenAI Codex
2026-10-07 12:19:48 -06:00
Behdad Esfahbod 0529c75459 [vector] Preserve contents on self-copy assignment (#6331)
Copy assignment resets the destination before reading the source. When
they are the same vector, this also clears the source. In particular,
hb_set_set(set, set) loses the set's stored pages.

Return immediately for self assignment. Cover trivial and nontrivial
vector elements and public set assignment in both inversion modes.

Validation: both regressions fail before the fix; meson test -C build
passes (261 passed, 1 skipped).

Assisted-by: OpenAI Codex
2026-10-07 12:16:05 -06:00
Behdad Esfahbod 3957e16e6c [set] Skip empty pages during reverse iteration (#6332)
Deleting the last member of a page leaves an empty page in the sparse set. `hb_bit_page_t::get_max()` returned zero for that page, so reverse iteration could report a value that was never in the set. For example, add and delete 512, then iterate backward from 1024: HarfBuzz reported 512 from an empty set.

Return INVALID for an empty page, matching get_min(), so previous() skips it. Cover reverse scalar/range iteration with retained empty pages, including inverted sets.

Assisted-by: OpenAI Codex
2026-10-07 12:09:14 -06:00
Behdad Esfahbod c22ae24405 [set] Reject unequal populations when comparing mixed inversion (#6329)
The mixed-inversion equality path compares zipped iterators, which stop
at the shorter set. An empty set therefore compares equal to a full
inverted set, as does any matching finite prefix.

Check populations before comparing members and add regression coverage
for empty sets and matching prefixes in both argument orders.

Validation: meson test -C build (261 passed, 1 skipped).

Assisted-by: OpenAI Codex
2026-10-07 12:06:58 -06:00
Behdad Esfahbod 4981dcc115 [set] Seek batch iteration correctly across missing pages (#6330)
When the cursor's page is absent, bfind selects the next stored page.
Using the cursor's offset in that different page can skip its members,
or skip the entire page when the cursor ends on a page boundary.

Apply the cursor offset only when the selected page matches its major
index. Cover batch and scalar iterator agreement across missing pages
for both ordinary and inverted sets.

Validation: the new regression fails before the fix; meson test -C build
passes (261 passed, 1 skipped).

Assisted-by: OpenAI Codex
2026-10-07 12:05:32 -06:00
Behdad EsfahbodandClaude Opus 5.5 4dd9449555 [machinery] Cache failed face accelerators as a tagged Null pointer (#6327)
* [machinery] Don't cache the Null object in face accelerator loaders

When libharfbuzz and libharfbuzz-subset (or raster, vector, gpu,
cairo) are separate shared libraries, each has its own hidden
_hb_NullPool.  If an accelerator's allocation failed, the lazy loader
stored the calling library's Null object in the face.  When the face
was later destroyed from the other library, do_destroy () did not
recognize that Null object and ran the destructor on read-only
memory.

hb_face_lazy_loader_t's create () only fails on allocation failure,
so stop storing get_null () there and retry on next access instead.
A stored instance is now always a real object.  Other loaders, whose
get_null () is a shared static default, keep caching it.  The
success path and get_relaxed () are unchanged.

Since a failed load no longer leaves anything behind for
get_relaxed (), the subsetter, which runs closure_lookups () on its
own GSUB/GPOS blob and recurses through the face accelerator with
get_relaxed (), now forces the accelerator to load through the public
hb_ot_layout_has_substitution/positioning () and fails the plan if
it is missing.  Calling get () from libharfbuzz-subset is not an
option since the accelerator constructor uses a hidden symbol.

Testing: the six hb-subset-fuzzer crashes against a shared-library
ASan build (write to _hb_NullPool in hb_ot_face_t::fini ()) no
longer reproduce.  A deterministic regression test is not practical
since it needs a specific allocation to fail in a process using two
of the shared libraries.  meson test: 261 passed, 1 skipped.

Fixes: https://github.com/harfbuzz/harfbuzz/issues/6324

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* [machinery] Cache failed face accelerators as a tagged Null pointer

The previous commit stopped caching the Null object when a face
accelerator failed to allocate.  That made repeated accesses to the
same accelerator inconsistent under OOM: a failed get () returned
Null and a later one could retry and succeed.  CIFuzz found the
subset driver taking its table from one access and its blob from the
next, then subsetting with the Null CFF1 accelerator (null fdSelect).
Many places access an accelerator more than once, so fixing callers
one at a time is not viable.

Revert that, and the subsetter guard it needed, and instead cache the
failure as get_null () with its low bit set.  get_stored () and
get_stored_relaxed () mask the bit off, so every access keeps seeing
the same object, and do_destroy () skips tagged pointers by the bit
rather than by comparing against this library's Null address.  That
comparison is what failed when libharfbuzz and libharfbuzz-subset
(or raster, vector, gpu, cairo) are separate shared libraries, each
with its own _hb_NullPool.

Only hb_face_lazy_loader_t opts in; other loaders compile as before.
A loader instance must stay one pointer in size, so the flag is
carried in the pointer rather than a separate member.  The tagged
pointer only ever refers to the static Null object, never to heap
memory.

Testing: the CIFuzz testcase no longer reproduces (static and shared
ASan builds), nor do the six shared-build hb-subset-fuzzer crashes.
meson test: 261 passed, 1 skipped.

Fixes: https://github.com/harfbuzz/harfbuzz/issues/6324

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-06 20:19:24 -06:00
Behdad EsfahbodandClaude Opus 5.5 bab434adb5 [metrics] Remove commented-out GET_METRIC_X macro (#6328)
ff9183045f commented out the unused GET_METRIC_X in
_hb_ot_metrics_get_position_common () to silence -Wunused-macros,
but the trailing backslashes on the // lines then triggered gcc's
-Wcomment.  Delete the dead macro and its matching #undef instead.

meson test: 261 passed, 1 skipped; no warnings from hb-ot-metrics.cc
with -Wunused-macros.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-06 19:58:12 -06:00
Behdad Esfahbod a52249ff62 Merge pull request #6326 from harfbuzz/fix/cff2-fdselect-varc
[subset] Bound CFF2 FDSelect lookups by the CharStrings count
2026-10-06 19:34:36 -06:00
Behdad Esfahbod dcb6e077fb Merge pull request #6313 from harfbuzz/fix-layout-condition-precision
[subset] Fix condition precision and instancing edge cases
2026-10-06 14:03:12 -06:00
Behdad Esfahbod 9af58580c6 [subset] Merge main into the condition precision branch
Resolve condition-evaluation conflicts while preserving the layout
instancing fixes and main's double-precision and static VARC support.

Tests: full Meson suite (279 passed), and C++11 syntax checks with
HB_NO_VAR, HB_NO_SUBSET_LAYOUT, and HB_OPTIMIZE_SIZE.

Assisted-by: OpenAI Codex <codex@openai.com>
2026-10-05 18:51:15 -06:00