* TIFF: open YCbCr images as RGB (#10316)
Scanners often write JPEG-compressed YCbCr TIFFs with 2x2 chroma
subsampling. The tiff crate decodes and upsamples them but returns raw
YCbCr samples, which the decoder rejected with "colour type YCbCr(8)".
The new tiff/ycbcr.rs converts them to RGB after decoding: JPEG data
uses the JFIF full-range BT.601 conversion (as libtiff does), other
compressions honour ReferenceBlackWhite and YCbCrCoefficients with the
TIFF defaults as fallback. 8 and 16-bit. YCbCr with extra samples, or
in other sample formats, fails with a clear unsupported error.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Contributors: add Yuki-init's display name
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* TIFF YCbCr: test that YCbCrCoefficients set the luma weights
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* TIFF: YCbCr tags with the wrong value count use the defaults
tag_f64s now returns None unless the tag holds exactly n values, so a
YCbCrCoefficients or ReferenceBlackWhite tag with extra values falls
back to the TIFF defaults instead of being truncated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Lrgj5pnShYtsSnrsZq5iRn
* TIFF YCbCr: tests for tiled JPEG with ICC, big-endian 16-bit and planar
Covers the layouts the review asked for: a tiled 4:2:0 JPEG file with an
embedded profile (checked against an independent JPEG decoder), a
big-endian 16-bit file, and planar data in both byte orders.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Lrgj5pnShYtsSnrsZq5iRn
* TIFF YCbCr: cite the formula sources and pin the green-channel choice
Documents where the conversion comes from: TIFF 6.0 Section 21
(YCbCrCoefficients, pp. 90-91) and libtiff's tif_color.c
(TIFFYCbCrToRGBInit, TIFFYCbCrtoRGB). Red and blue follow the spec;
green follows libtiff and adds Y unscaled, which matches the spec only
when the luma weights sum to 1. The module doc records this as a
deliberate compatibility choice, and a new test pins it: with weights
0.25/0.25/0.25, mid gray stays (128, 128, 128).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Lrgj5pnShYtsSnrsZq5iRn
* Codecs README: YCbCr TIFFs with three 8- or 16-bit samples decode to RGB
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Lrgj5pnShYtsSnrsZq5iRn
* Format tiff_ycbcr tests with rustfmt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(tiff): compare grouped YCbCr decoding and update support notes
* style(tiff): align grouped decode call with test formatting
* style(tiff): apply repository rustfmt to grouped YCbCr regression
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: FredApps <153173854+FredApps@users.noreply.github.com>
Co-authored-by: FredApps <FredApps@users.noreply.github.com>
* Format: take the .pcraft envelope from craft-pcraft-format
The manifest envelope, TileRef, Hash, FORMAT_VERSION, EXTENSION, the
entry names, tile_path and blob_path, is_valid_hash, the manifest
nesting and item checks and the migration runner now come from
craft-libs' craft-pcraft-format. photocraft-format re-exports them, so
callers do not change. Manifest is craft_pcraft_format::Manifest<DocM>.
Saved bundles keep the same bytes. A save still checks only the nesting
and the item count of its manifest, not its size, and the manifest
nesting limit is still computed from photocraft_doc::MAX_GROUP_DEPTH.
Tests: a save, load and save again gives the same bytes in every color
mode and depth, with and without previews; MAX_GROUP_DEPTH equals
craft_pcraft_format::MAX_GROUP_DEPTH.
* Format: pin craft-pcraft-format from storytold/craft-libs
* Format: read manifests with craft-pcraft-format's read_manifest_with
Pins craft-pcraft-format to dca870a0, the head of craft-libs #407.
Loading reads the manifest with read_manifest_with, which runs the
limit check, the version check, the migration steps and the envelope
read in one path; the layers are still decoded one at a time after the
rest of the document. A save builds the manifest with Manifest::new and
tile references with TileRef::new, and the bundle's directory names and
the .zst suffix come from the crate's constants. Saved bytes and error
messages are unchanged.
* Format: pin craft-pcraft-format to craft-libs #407's head 04dd1bd
04dd1bd changes only the crate's docs and a test, so photocraft-format
builds the same code.
* Format: pin craft-pcraft-format to craft-libs main at b8f7d93
craft-libs #407 merged as b8f7d93. Its read_manifest also refuses a
thumbnail other than thumb.png and a composite other than
composite/preview.png, the only names photocraft-format writes.
The flag added in the previous commit cannot be shown to do anything: with
it removed, a press on the grip inside the *active* document still drags
the border and still paints nothing. egui gives the press to the grip,
which is laid out after the canvases, so the canvas never sees it — the
reported case does not reproduce. Rather than keep a guard no test can
tell apart, it goes, and the comment says why the order is enough.
What stays is the test it prompted, which is the one the review asked for:
a grab 2 pt inside the active document's edge, with the Brush selected,
drags the border, adds no history entry, and `arranged_cells` reports both
cells meeting at the new position.
Also documents `arranged_cells` and `arrange`, which the docstring check
had nothing to read on.
Three review findings:
- A grip overlaps the active document's edge, where the press started a
brush stroke as well as the drag. The frame now records that the press
began on a grip (`arrange_grip_press`) and the canvases ignore it, the
way the inactive-document path already did.
- `drag_border` capped the gap at half the even spacing, so on a 300 pt
axis a 2-up cell stopped at 75 pt although 80 would have fitted. The
cap is gone; the limit is MIN_CELL itself.
- With the neighbours closer together than two minimum cells, the clamp
wrote a value from the wrong side of them and put the list out of
order. That case now leaves the border, and the list, as they were.
The tests cover the short axis and both directions of the no-room case,
and each fails with its own change reverted.
* perf: reuse large buffers so a pinned glibc trim threshold costs no speed (#10479)
With MALLOC_TRIM_THRESHOLD_ set (#8743), glibc maps every allocation over
128 KiB fresh, so tiles, compositor bands and filter scratch were page-faulted
in on every call: viewport compose 3.6x slower, Gaussian Blur up to 1.8x,
.pcraft save 2.4x.
- photocraft_raster::scratch: a process-wide, capped pool of large Vec
buffers (u8/u16/f32/[f32; 4]). A miss drops the oldest buffers worth its
size down to a 256 MiB floor, so peak memory stays near glibc's defaults.
The engine releases the pool after any command that closes a document,
Headless::close does too, and an idle thread releases it after 2 s unused.
- Tiles, compose Buffer/masks/bands, algo Image and the blur paths, engine
adjustment/filter scratch, GPU tile upload and io import buffers go through
the pool.
- .pcraft save reuses one zstd FrameCompressor per worker (byte-identical).
- interactive_bench reports minor page faults per run.
36 MP, trim threshold set vs glibc defaults: viewport 5.6 vs 6.1 ms (was
22.7), Blur r4 445 vs 437 ms (was 852), Curves CPU proxy 35.0 vs 35.7 ms
(was 50), flatten 146 vs 147 ms (was 237), save .pcraft 251 vs 230 ms (was
545). RSS after closing a 5400x7200 document: 21 MB with trim set.
* fix(linux): restart once with glibc's trim threshold set so closed documents return memory (#8743)
With glibc's default malloc, the memory of a closed document stays with the
process: freeing a large buffer raises glibc's dynamic mmap threshold, so
later tile buffers land in heap pages glibc keeps. A session that opened and
closed a 5400x7200 document held ~450 MiB afterwards.
photocraft_engine::malloc_tuning::apply(), called first in both binaries'
main, re-executes the binary once with MALLOC_TRIM_THRESHOLD_=131072 on Linux
with glibc: same PID, arguments, argv[0] and environment, in safe Rust
(CommandExt::exec). A malloc variable or glibc.malloc.* tunable the user or a
package already set is left alone, so setting MALLOC_TRIM_THRESHOLD_ opts
out. No-op on macOS, Windows, musl and wasm.
Depends on the scratch-buffer pool (#10479), without which the pinned
threshold slows filters, previews, flatten and .pcraft save by 26-209%.
* fix: corpus tests build again; return remaining pooled buffers (#10479 review)
- Corpus-gated io tests and the Photoshop-oracle test moved `.px` out of a
`Buffer`, which now implements `Drop` (E0509 with `--features corpus`):
use `into_px()`.
- io: masks, the merged image, extra channels, smart-filter masks,
smart-object pixels and patterns give their `interleave` buffers back to
the scratch pool after writing them.
- compose: recycle the mask values of a pass-through group, `effect_shape`
and both `split_parts` callers.
- Styled Free Transform preview copies each band out instead of moving it,
so the band returns to the pool.
- interactive_bench: the fault count covers each whole run, untimed setup
included; say so.
* fix: corpus tests build again; return remaining pooled buffers (#10479 review)
- Corpus-gated io tests and the Photoshop-oracle test moved `.px` out of a
`Buffer`, which now implements `Drop` (E0509 with `--features corpus`):
use `into_px()`.
- io: masks, the merged image, extra channels, smart-filter masks,
smart-object pixels and patterns give their `interleave` buffers back to
the scratch pool after writing them.
- compose: recycle the mask values of a pass-through group, `effect_shape`
and both `split_parts` callers.
- Styled Free Transform preview copies each band out instead of moving it,
so the band returns to the pool.
- interactive_bench: the fault count covers each whole run, untimed setup
included; say so.
* fix: recycle the masks of adjustment-atop and shape layers (#10479 review)
* fix: recycle the masks of adjustment-atop and shape layers (#10479 review)
* fix: recycle effect hiding masks (#10479 review)
* fix: recycle effect hiding masks (#10479 review)
* fix: recycle knockout, CMYK and GPU shape masks; release the pool when Split/Merge Channels close a document (#10479 review)
* fix: recycle knockout, CMYK and GPU shape masks; release the pool when Split/Merge Channels close a document (#10479 review)
* fix: release the pool when a closed document's cancelled workers exit; idle-release switch for tests; don't pool PSD planes (#10479 review)
* fix: release the pool when a closed document's cancelled workers exit; idle-release switch for tests; don't pool PSD planes (#10479 review)
* fix: release the pool for workers cancelled before their document closes; take the pool lock for budget and idle-release changes (#10479 review)
* fix: release the pool for workers cancelled before their document closes; take the pool lock for budget and idle-release changes (#10479 review)
* fix: idle release rechecks activity under the lock; try_take_exact releases the pool and retries before failing (#10479 review)
* fix: idle release rechecks activity under the lock; try_take_exact releases the pool and retries before failing (#10479 review)
* test: cover a successful try_take_exact miss; name the failed-reservation test for what it checks (#10479 review)
* test: cover a successful try_take_exact miss; name the failed-reservation test for what it checks (#10479 review)
* fix: pool the final GPU effect shape buffer and return it after use (#10479 review)
* fix: pool the final GPU effect shape buffer and return it after use (#10479 review)
* docs: describe scratch pool lifecycle regression invariants
* test: verify polling releases cancelled document scratch buffers
* perf: recycle remaining transform selection and PSD scratch buffers
* fix: free PSD gather bands before allocating channel planes
* test: assert exact scratch take reuses pooled storage
* perf: reuse large buffers so a pinned glibc trim threshold costs no speed (#10479)
With MALLOC_TRIM_THRESHOLD_ set (#8743), glibc maps every allocation over
128 KiB fresh, so tiles, compositor bands and filter scratch were page-faulted
in on every call: viewport compose 3.6x slower, Gaussian Blur up to 1.8x,
.pcraft save 2.4x.
- photocraft_raster::scratch: a process-wide, capped pool of large Vec
buffers (u8/u16/f32/[f32; 4]). A miss drops the oldest buffers worth its
size down to a 256 MiB floor, so peak memory stays near glibc's defaults.
The engine releases the pool after any command that closes a document,
Headless::close does too, and an idle thread releases it after 2 s unused.
- Tiles, compose Buffer/masks/bands, algo Image and the blur paths, engine
adjustment/filter scratch, GPU tile upload and io import buffers go through
the pool.
- .pcraft save reuses one zstd FrameCompressor per worker (byte-identical).
- interactive_bench reports minor page faults per run.
36 MP, trim threshold set vs glibc defaults: viewport 5.6 vs 6.1 ms (was
22.7), Blur r4 445 vs 437 ms (was 852), Curves CPU proxy 35.0 vs 35.7 ms
(was 50), flatten 146 vs 147 ms (was 237), save .pcraft 251 vs 230 ms (was
545). RSS after closing a 5400x7200 document: 21 MB with trim set.
* fix: corpus tests build again; return remaining pooled buffers (#10479 review)
- Corpus-gated io tests and the Photoshop-oracle test moved `.px` out of a
`Buffer`, which now implements `Drop` (E0509 with `--features corpus`):
use `into_px()`.
- io: masks, the merged image, extra channels, smart-filter masks,
smart-object pixels and patterns give their `interleave` buffers back to
the scratch pool after writing them.
- compose: recycle the mask values of a pass-through group, `effect_shape`
and both `split_parts` callers.
- Styled Free Transform preview copies each band out instead of moving it,
so the band returns to the pool.
- interactive_bench: the fault count covers each whole run, untimed setup
included; say so.
* fix: recycle the masks of adjustment-atop and shape layers (#10479 review)
* fix: recycle effect hiding masks (#10479 review)
* fix: recycle knockout, CMYK and GPU shape masks; release the pool when Split/Merge Channels close a document (#10479 review)
* fix: release the pool when a closed document's cancelled workers exit; idle-release switch for tests; don't pool PSD planes (#10479 review)
* fix: release the pool for workers cancelled before their document closes; take the pool lock for budget and idle-release changes (#10479 review)
* fix: idle release rechecks activity under the lock; try_take_exact releases the pool and retries before failing (#10479 review)
* test: cover a successful try_take_exact miss; name the failed-reservation test for what it checks (#10479 review)
* fix: pool the final GPU effect shape buffer and return it after use (#10479 review)
* docs: describe scratch pool lifecycle regression invariants
* test: verify polling releases cancelled document scratch buffers
* perf: recycle remaining transform selection and PSD scratch buffers
* fix: free PSD gather bands before allocating channel planes
* test: assert exact scratch take reuses pooled storage
The author rebased #8327 onto a newer main; its code is the same change as the reviewed head
c60cbcb99 (range-diff: Cargo.lock context and the people.toml position only). The batch keeps
its reviewed and fixed-up version of every file, so the tree is unchanged; merging the new head
lets GitHub mark #8327 as merged with the batch.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CodeRabbit on batch 11: the rest of a set paused at a step dialog was kept by name, so an action
deleted and made again under the same name while the dialog was open played under the original
Play. Actions have no id, so the rest is now kept as the actions themselves and continues only
while each one is unchanged; moving them in the list still works.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- #8714: a Camera Raw Filter step imports only with a process version
the PSD mapping has verified (PrVN/PrVe 5/184549376 or 6/251920384)
and a CrVe text; other steps stay kept steps that save back unchanged.
- #7642: the patternName fallback matches names only, so a pattern
whose id equals that text no longer wins. A step that names its
pattern only by patternName exports that name.
- #8327: the iso-2022-jp label is refused (its names are 7-bit escape
sequences that would pass through as ASCII).
- #10417: the Levels export gate also refuses endpoints of the wrong
shape (both ends text, a text gamma) and channels that aren't objects.
- #10459: a set paused at a step dialog keeps the rest of the set by
action name, so actions.move while the dialog is open neither plays a
foreign action nor skips one; "Played N steps" covers the whole Play
(steps before the pause and the dialog's own step); the suggested .atn
name is cut to 255 bytes at a character boundary; actions.list returns
every set, empty ones too; set_names dedups with a HashSet.
Co-authored-by: FinnNk <1424666+FinnNk@users.noreply.github.com>
Co-authored-by: Silvertree2010 <133535268+Silvertree2010@users.noreply.github.com>
Co-authored-by: regexer <3689390+regexer@users.noreply.github.com>
Co-authored-by: marshiyar <57055687+marshiyar@users.noreply.github.com>
Co-authored-by: hectorviov <1501713+hectorviov@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Stream PSD user and real masks into tile rows
* Correct streamed-mask fixture metadata and test lint issues
* Budget PSD pattern expansion and bound multichannel conversion
---------
Co-authored-by: FredApps <FredApps@users.noreply.github.com>
#8327 adds `AtnFile::encoding`. Two tests from other PRs in the batch
build an AtnFile literal without it (#8714 fix-up's Camera Raw save-back
test, #8858's Vibrance string-class test), so the engine's lib tests
didn't compile on the batch. They now fill the rest with
`..Default::default()`, as the other literals in the file do.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The previous fix-up counted a letter glued to an ASCII letter as common
only when it was Latin. That turned ordinary windows-1251 names such as
"CMYKканал" or Ukrainian "Копiя" (Latin i) into a windows-1252 (or GBK)
reading of the whole set, with no warning, where the PR read them right.
Back to the PR's scoring for single-byte pages; the ¥ removal stays.
windows-1252 "Fläche m²" / "Preis £5" reading as windows-1251 is the
PR's own known limitation, now noted in the code. A Russian set plus
"CMYKканал" is pinned as windows-1251.
Co-authored-by: regexer <3689390+regexer@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mac Roman 0xA5 is the bullet, windows-1252 0xA5 is ¥: with ¥ counted in
the windows-1252 re-weigh, Mac "Ebene • Kopie" read as "Ebene ¥ Kopie".
Drop ¥; the other kept symbols are no letter or typographic mark in Mac
Roman.
windows-1251 has І і д where windows-1252 has ² ³ ä, so "Fläche m²" read
as "Flдche mІ" (already so in the PR head; main showed it right). In a
single-byte page, a letter glued to an ASCII letter now counts as common
only when it is a Latin letter. Real Cyrillic names ("Новый слой",
"Копия слоя 2", "Канал RGB") still read as windows-1251. "Preis £5"
(Ј between a space and a digit) still reads as windows-1251; documented.
Co-authored-by: regexer <3689390+regexer@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- A set whose action paused at a step dialog inside an action it calls handed the rest of
the set to the called action's dialog, which resumes alone: after OK the set's next
action ran and the caller's own remaining steps never did. The rest of a set now rides
only on a pause of the target action itself; otherwise the set stops there.
- Set folders could only be collapsed with the pointer on the triangle: a keyboard or
accessibility activation always just selected. Without a pointer it now selects and
toggles, as before, and the row reports whether it is selected (not whether it is expanded).
Co-authored-by: hectorviov <1501713+hectorviov@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Three gaps in the set selection the fix-up added:
- Play on a set whose action paused at a step dialog dropped the rest of the set after OK.
The pause now carries the remaining actions (also across a second dialog of the same
action), and OK plays them under the same gate. Cancel still stops the whole set.
- A collapse or expand selected the set and cleared the action, so Play then ran a whole
other set. As on action rows, the triangle's column now only collapses or expands, and
the rest of the folder row selects the set.
- Selecting a step (or an action through Load, Insert Stop, a drag or recording) kept an old
selected set, which came back after Delete. Every action selection now clears it.
Co-authored-by: hectorviov <1501713+hectorviov@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conflict: main moved crates/engine/src/atn_cmds.rs into crates/engine/src/cmd/atn_cmds/; the refusal is ported into steps_make.rs set_to's layer.setAdjustment arm unchanged. The tests followed the rename.
`resolve_or_named` resolved the recorded `Idnt` with `resolve`, which also
matches by name in each collection. So a document pattern named like the
id beat the library pattern with that id, and a foreign id equal to some
pattern's name (`Idnt "Red"`, `Nm "Blue"`) played "Red" and never tried
the recorded name. The id is now looked up by id only (document, then
library), then the name decides. A library pattern found by id whose
pixels can't be read reports that instead of falling back, main's
paintable check still runs, and a pattern installed neither way is named
in the error (`no pattern "My Wood" (id "...")`) instead of only its UUID.
Co-authored-by: Silvertree2010 <133535268+Silvertree2010@users.noreply.github.com>
Co-authored-by: FredApps <153173854+FredApps@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A click on a set folder row (also how a set collapses or expands) now clears the selected
action, so Play fell back to the first action of the whole list, often from another set,
and Batch and Create Droplet started on it too. Example: select Resize (set Web), collapse
Retouch, press Play: Retouch's first action ran instead.
With a set selected, Play now plays every action of that set in order, stopping at the
first that fails or pauses for a dialog (an empty set plays nothing), as the reference
application plays a selected set. Batch and Create Droplet start on the selected set's
first action. A click on an action row clears the selected set, so a later Delete doesn't
bring the old set back as the target of New Action or Play.
Co-authored-by: hectorviov <1501713+hectorviov@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Vibrance's adjustment class is now Photoshop's string id `vibrance`, so Set and Make steps
save back with the class Photoshop runs instead of the `Vbrn` code it refuses (#8383).
`Vbrn` moves to LEGACY_CLASSES, so actions saved before still load. A recorded Vibrance
with non-zero temperature or tint stays a kept step and saves back unchanged.
Co-authored-by: jturnbach <70334515+jturnbach@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A loaded Camera Raw Filter step became an import-only filter.cameraRaw, so
Save Actions dropped it (skipUnsupported) or failed, where before it was
written back byte for byte. Keys outside the table (e.g. PerV = 20, a
profile name) were ignored, so the step replayed a different look without
being reported.
The step now keeps its recorded descriptor in a hidden param and saves back
as that descriptor while its params are still what it imports as; edited or
PhotoCraft-recorded steps are still refused. A step with any unknown key, or
with no settings, stays a kept Photoshop step, as the PSD mapping does.
Co-authored-by: FinnNk <1424666+FinnNk@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The previous fix-up put main's atn fuzz target back. It compares the
parsed actions with assert_eq!, so any valid file holding a NaN double
fails the fuzzer falsely (NaN != NaN); the PR's own contract test
nan_payloads_signed_zero_and_infinities_survive_without_false_fuzz_failures
covers exactly that. The target now calls the shared check again, at its
new path under tests/it/common.
Co-authored-by: denstu473 <63150090+denstu473@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The contract tests no longer compiled on main (AtnItem::dictionary is now
Vec<u8>, so it is compared with b"Future event"), sat outside main's
tests/it layout, and the new atn.yml added six legs (two macOS) to every
PR against ci.yml's policy. Move the tests into tests/it so the existing
CI runs them, drop atn.yml and docs/atn-testing.md, keep main's ci.yml and
fuzz target, and note the committed fixtures in ATTRIBUTION.md.
Co-authored-by: denstu473 <63150090+denstu473@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>