DPX at 10 or 12 bits and Cineon (always 10) are planned as U16, so
DepthReduced, which compares sample types, never fired: a 16-bit document
saved at 10 bits lost six bits per sample silently, and a float one was
told it would be reduced to 16-bit.
FidelityWarning::BitsReduced { from, bits } names the stored depth. Float
samples always get it; U16 ones only when some sample is not already a
code value at that depth, so a frame opened from a 10- or 12-bit file
saves back without a warning. U8 widens losslessly and stays silent.
* security(paint): cap distance-spaced steps per segment and procedural tip primitives
* test(paint): count steps through a Cell so the walker closure can be shared
* paint: keep the spacing carry-over when a segment ends at the step cap; lower the tip work budget
* fix(ui-egui): a click on the bottom layer's Smart Filters selects them again (#8764)
The empty space below the last Layers panel row deselects every layer on a
click (#8100). It started under the last layer's own row, so it covered that
layer's effect and Smart Filter rows and, registered after them, took their
clicks: on the bottom smart object (a new document's Layer 0), clicking a
filter deselected everything instead of selecting it.
The click area now starts under everything drawn for the last layer. The
drop zone below it (#1721) is unchanged.
* Contributors: credit drunkeninja42 as Kushagra Sharma
* perf(gpu): reuse filled-shape effect maps during translation
Hash filled-shape outlines in local coordinates so whole-pixel movement can use the existing effect-map translation reuse path. Keep its region-size and exact coverage checks, and invalidate the key when the outline deforms.
Add a shape-key regression and a synchronized styled-shape group benchmark with fresh-render parity verification. On RTX 4070 Laptop, the 1600x1200 fixture improves from 61.336 to 33.120 ms median; a 24 MP full-canvas fixture improves from 223.018 to 184.289 ms. Warm frames rebuild zero effect programs, and the quick fixture matches fresh rendering exactly.
* fix(gpu): stabilize fractional shape effect fingerprints
Quantize only origin-normalized fingerprint coordinates to a 1/65536 pixel grid and canonicalize signed zero. Preserve the render path and exact coverage guard.
Cover fractional anchors and handles, positive and negative integer translations and handle deformation. Exercise fractional origins in the benchmark and document its entry points. All 40 GPU tests, Clippy and WASM checks pass; reused output matches a fresh compositor exactly.
---------
Co-authored-by: kotulsky <7805437+kotulsky@users.noreply.github.com>
Review of #8994: with Canvas Border set to Line, the border stayed on
the document's rectangle while a size-changing preview (a 90° rotation)
drew over its own area. The canvas closure now returns the edge it drew
(the preview's area on the CPU canvas), and the border outlines it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Anchors selected on one saved path stayed selected after the panel picked
another, so the arrow keys still nudged the old path (or whichever path
took its index after a delete). They now check the target is still one of
the paths Direct Selection edits and clear the selection if not.
* packaging(macos): retry stapling when Apple's ticket lookup drops
The 0.15.0 release's macOS job failed after both notarizations were
accepted: stapling the DMG lost its connection to Apple's CloudKit
ticket service (error 68). Retry stapling up to five times with a
growing pause before failing the job.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* packaging(macos): retry stapler validation too
The re-run of the 0.15.0 macOS job stapled the DMG, then lost the
CloudKit connection in `stapler validate` (error 68 again). Retry
staple + validate as a pair in package.sh, and retry the validation
verify.sh does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A filter preview that changes the canvas's dimensions (Image > Image
Rotation) kept the unpreviewed document on the CPU canvas (a mirrored
or pixel-aspect view, no usable GPU, a GPU fallback), since drawing its
texture over the document's bounds stretched it. The CPU canvas now
draws it over its own area centred on the document's, as the GPU canvas
centres it (preview_bounds, CpuTextures::bounds), with the drop shadow,
checkerboard and rotated-view mapping on that area too. Such previews
redraw whole and have no detail texture (both work in the document's
pixels). Turning Preview off or cancelling restores the document.
Fixes#8989.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CodeRabbit noted that the wheel tests this PR adds assumed a notch is 40
points (5 px at 800 %). That is egui's native line_scroll_speed; on wasm32
egui uses 8. The tests now take it from egui instead:
- wheel_nav tests: configured_notch(), a context set up like the app's;
- canvas tests in input_tests: notch_pan_distance(), the harness's
line_scroll_speed over NOTCH_PAN_ZOOM.
Distances that aren't notches (the 600-point page, the 53-point pixel step)
stay as they are.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The saved path picked in the Paths panel was drawn as the target path
(#6550), but none of the path tools looked at it: Path Selection only
moved a vector mask, a shape path or the work path, and Direct Selection
and the anchor tools only hit those too. A PSD's "Path 1" could be seen
but not selected or changed.
Direct Selection gets a PathRef::Saved target (the path's index, sent to
the engine by name) after the layer path and before the work path, and
Path Selection and Free Transform Path target the selected saved path the
same way. The engine's path.* commands already took a saved path's name.
Brings in 72 commits from main (bde13687). Two conflicts:
- apps/photocraft-web/Cargo.toml: main added the web-sys RequestMode feature;
the list is main's plus this branch's MediaQueryList.
- crates/ui-egui/src/prefs_ui.rs: main deleted the translation-coverage test
preference_labels_are_translated (c3aa3f9e, missing translations are now
reported by xtask ci); this branch had only adjusted it, so it goes.
docs/scorecard.md is regenerated; it differs from main only in UI-217-10.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review of #8978:
- A preview whose dimensions change (Image > Rotation) would have been
stretched over the document's bounds by the CPU canvas, so it draws
only previews the document's size, reduced by their proxy factor
(preview_fits); others keep today's behaviour there.
- "Reduced" (no detail texture, whole redraws) is the proxy factor
k > 1, not a size difference, so a full-resolution preview keeps the
detail texture of a large document.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With Preview on, the canvas showed the document unfiltered whenever it
drew through the CPU canvas: a mirrored or pixel-aspect-corrected view,
a machine without a usable GPU, or a GPU fallback. Only the GPU canvas
branch and the channel views asked for the filter preview. The CPU
canvas now draws its result too (filter_channel_preview); a zoomed-out
proxy covers the document at its own size, so its texture is redrawn
whole and has no detail texture.
Fixes#8972.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* security(compose): bound dense working sets and repeated group evaluation
Smart-filter mixing and stroke fitting work tile by tile instead of over the
union of sparse surfaces; render_layer admits its rect like render; hidden
layers no longer disable the proxy and uniform-area shortcuts; Blend If
groups nested deeper than two are judged by their result, so nesting costs
polynomial instead of exponential work; effect builds past an absolute byte
cap are skipped.
* CodeRabbit #8949: exact tile skips in filter_mask, allocated-tile scan in fit_stroke, contrasting backdrop in the nested-group test
- filter_mask mix/unmix: bare tiles (no tiles on either side) skip for
non-alpha formats too when the two surfaces' default pixels are equal;
differing defaults still get processed, so the clone's default pixel is
never left wrong under a non-unit mask
- shape_split fit_stroke: scan the union of the three surfaces' allocated
tiles (exact skip when all three defaults are transparent) instead of
every tile of a 300k-canvas document; non-transparent defaults fall
back to the full scan
- tests: the nested-groups assertion now uses a black backdrop so the
asserted white can only come from rendering through the group chain
fixes CodeRabbit review comments (major/minor) on filter_mask.rs:27,29,
shape_split.rs:126, tests.rs:2354
* compose: skip a BTreeSet in fit_stroke — a sorted Vec compiles under the wasm toolchain (E0599 on ::drain there)
* security(algo): bound working sets sized by caller-controlled extents
Zero-width inpaint grids, content-aware fill mosaics of turned thin images,
Liquify fields over huge canvases, tiny-scale reduction weights, warp tile
bookkeeping and per-tile source footprints, and windowed filters whose halo
follows an unclamped radius or a stacked Filter Gallery reach all asked for
memory sized by file or parameter input. Each is now bounded; halo
arithmetic saturates instead of overflowing.
* security(algo): budget variants by dimensions, size-check footprints first, widen the tile-halo maths
CodeRabbit follow-up:
- fill_cow settles the mosaic budget on variant dimensions only, so a
variant's image and mask buffers are not built when the mosaic would
exceed MAX_MOSAIC_PIXELS (complete falls back to the image itself).
- warp validates the footprint's pixel area against MAX_FOOTPRINT_PIXELS
before src.has_tiles_in, so an oversized footprint returns without the
tile-grid scan.
- auto_tile widens the input-derived halo arithmetic to i64 and clamps
before the cast, so an i32::MAX radius cannot overflow in debug builds
and the 2048 tile cap stands.
* fix(algo): drop the duplicate mosaic_size call left behind in fill_cow
The variant-budget rework kept both the new dimensions-based budget and
the old variants-based one; the old call no longer compiled after
mosaic_size started taking dimensions. Keep the comment that explains
the fallback.
Delete the coverage tests, the per-language parity checks and the tests
of one language's exact wording. Catalog integrity, placeholder and
plural-form checks stay, as do the engine tests (locale resolution,
plural rules, fallback, language scoping, drawing in a language); those
compare against tr() or the plural rule rather than catalog wording.
Docs describe the report instead of a failing test.
The report prints, per language, the source keys its catalog lacks (one
line for a language with none). `cargo xtask ci` runs it as its last step
and never fails on it: a missing translation shows in English. A report
that can't run prints a warning.
* packaging: use [[ for bash conditional tests (SonarCloud S7688)
SonarCloud flags every bare [ conditional in bash scripts under packaging/ as a reliability smell (rule S7688). Convert the flagged [ ... ] tests to [[ ... ]]: safer word-splitting and no glob surprises, same semantics for these string and file tests.
* packaging/web: parse size overrides as base 10 under [[
A waiting stroke (Remove after each stroke off) was tinted with the brush it
was drawn with but removed with the brush at ↩, so changing the brush in
between made the tint and the removal disagree.
- `paint.remove`: a `strokes` entry may be `{"points", "size"?, "hardness"?,
"spacing"?}`, which override the call's for that stroke; a plain point
list still takes the call's.
- The UI keeps each waiting stroke as such an entry with the brush it was
drawn with; its tint (`stroke_brush` of it) and its removal both use it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The chip drew eight concentric circles from `hardness` alone, with an
ad-hoc ramp, so it was always round and didn't show the engine's falloff.
It now draws `brush_preview::tip_texture`, as the Brush Preset picker's
thumbnails do: `dab_coverage` for a round tip's hardness, its angle and
roundness, a sampled tip's bitmap. `panels::brush_tip` is gone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CodeRabbit follow-up:
- data_end counts the scan-data bytes it crosses, and decode's
truncated-input bound is measured on those, not on bytes.len(): APP
padding padded before the scan used to charge the budget without
describing one more pixel.
- a regression test builds a truncated JPEG with ~256 KiB of complete
COMMENT padding whose patched 20000x20000 frame was admissible under
the file-length bound, and verifies padding alone cannot admit it
(while the same file at its real size still recovers).
- Limits::max_alloc docs state the animated WebP charging, next to the
PNG text/XMP sentence.