* Segmentation: anti-aliasing the full-resolution mask skips runs far from its edge
Select Subject, Object Selection and Select Sky finish with antialias_u8
over the whole full-resolution mask, which read nine neighbours through a
bounds-checked closure for every pixel: 171-185 ms of an 800 ms first
Select Subject on 24 MP (ubuntu-latest, 4 cores). A run of 64 pixels whose
3x3 neighbourhood holds one value has no edge pixel, so it is only compared
now; other runs take the per-pixel rule as before. The output is byte for
byte the same, checked against the per-pixel rule over widths around the run
length, single rows and columns, discs, noise and stripes. Part of #211.
* Segmentation: the mask's anti-aliasing and its test share one per-pixel rule
The 3x3 rule moves into edge_value, which antialias_u8 calls for spans that
are not uniform and the test calls for every pixel, so the test checks only
that skipping never misses an edge pixel instead of keeping a second copy of
the rule. A small case pins the rule's values.
* perf(seam): drop the original-column map from the shrink proxy, same pixels (-15 % at 24 MP)
Content-Aware Scale's proxy seam search tracked each panel column's original
column in an idx map, one usize per proxy pixel, shifted by every seam. The
original column is x plus the number of columns already taken at or left of it,
so a per-row list of the taken columns answers the same question without a
buffer the size of the energy map.
The recorded P28 budget number predates the proxies: the 2026-10-05 baseline
commit carves at full height and recomputes energy per seam, and measures
266 s for the 24 MP case on the machine this was timed on, against 2.6 s now.
Documented in docs/seam-carving-performance.md; enforce stays off until a
budgeted-class run can vouch for the whole row.
* review: generate exactly ch samples per pixel in bench_seam, reject ch > 4
The fixture flattened four samples per pixel and then truncated to w*h*ch, so a
three-channel run grouped samples across pixel boundaries and any ch above four
built a buffer shorter than the rows carve_width indexes.
Generate ch samples inside the per-pixel closure and refuse ch outside 1..=4.
The three-channel row is dropped from the perf note rather than re-measured
here: it moved inside the timer noise either way, and the other agents' builds
on this machine make a fresh number worthless. The four-channel hashes are
unchanged, which is the check that this edit touched only the fixture.
* perf(seam): find the removed column by binary search, not a linear scan
Recording a seam recovered each removed original column by scanning the row's
list of already-taken columns, so a deep shrink paid a comparison per removed
column on every row and every seam: quadratic in the seam count, and worse than
the index map this PR replaced once a shrink got deep enough.
`o - i` never decreases along the row's sorted list, so that count is a binary
search, and the original column belongs where the search stopped; the rows are
independent and update in parallel, like the pixel shift. What is left per
removed column is one insert.
Output is unchanged: same columns, same order, same hashes (10 % and 50 % width
cases below), and `counted_removed_columns_match_the_tracked_map` now runs a
tracking and a counting carver side by side to one column and compares every
row's list after every seam.
docs/seam-carving-performance.md gains those cases, states that the ~104x
figure is the two example runs rather than the baseline row, and names
perf/baseline.json as the source of the 954 s.
---------
Co-authored-by: ebosido444-ai <ebosido444-ai@users.noreply.github.com>
* Speed up Crystallize and Pointillize with cell-space rasterization
* Measure pixelate commands on unlocked layers and Background separately
* Record pixel-exact 24 MP filter and command benchmarks
* Fix magnetic lasso gradient polarity cost
* Add magnetic lasso gradient polarity regressions
* style: cargo fmt
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* 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.
* fix(algo): avoid copying duplicate EXIF text payloads
Keep the first valid text value for each EXIF tag, preventing duplicate entries from repeatedly decoding and copying large attacker-controlled strings.
🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
* test(exif): count text decodes so the duplicate-tag test observes skipped payloads
---------
Co-authored-by: Codebuff <noreply@codebuff.com>
refine_mask_with builds its result densely over the selection's bounds
(grown by the spread, plus Refine Edge Brush additions). After a thin
edit across a huge canvas, or with a whole-canvas selection there, that
is billions of pixels. matting::refine_area names that area and
select.refineEdge refuses one over MAX_REFINE_AREA (2^30 px) with an
error before allocating, leaving the selection unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FszacAJDjznVimaxQktoJT
The Brush edit, the Lasso edit and EdgeBrush::new allocated a dense buffer
over their bounding box, which the swept-area cap does not bound: a 1 px
diagonal across a 100,000 px canvas sweeps ~141k px but asked for ~40 GB.
- matting::brush_surface (replaces brush_region) and EdgeBrush paint only
the tiles a stroke touches (painted_tiles), capped at MAX_PAINTED_TILES.
- selection::polygon_surface rasterises only the tiles the Lasso's edges
cross, byte-identical to selection::polygon; interior tiles share one.
- selection::combine_sparse adds/subtracts visiting only the new shape's
tiles, keeping the rest of the selection shared.
- select.refineEdge refuses edits and refineStrokes that would cross too
many tiles before painting anything, leaving the selection unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FszacAJDjznVimaxQktoJT
The tilt-shift reference tests and tiled_clear_and_fill_match_one_pass_over_the_area cross pixel
formats or channel counts with independent geometry and fill settings. Each run takes a quarter of
the combinations through Sweep, a different quarter each UTC day.
precheck_agrees_with_execute built all five document states for every call and kept one; it now
builds only the state the call needs.
swept_blocks_match_the_pixel_search skips the cases where the colour mode reaches extrude only
through a channel count and ink setting an earlier mode already covered. It and
damaged_layer_data_opens_flattened run a quarter of their cases per run, a different quarter each
UTC day (photocraft-testkit's Sweep). PHOTOCRAFT_FULL_SWEEP=1 runs every case, and
PHOTOCRAFT_SWEEP_PART=<n> replays the part a failing run printed.
Review of #8783:
- matting::brush `segments` no longer collects its pieces: they are made
as they are painted, and each segment is first cut to the painted area
grown by the brush. A 1 px stroke zigzagging across 2 million px
(within MAX_SWEPT_AREA) split into ~134 million pieces collected up
front (~2 GB) in brush_region and EdgeBrush::new; now ~5000 pieces.
- select.refineEdge `edits`: the Brush edits' swept area in all is
bounded by MAX_SWEPT_AREA, as refineStrokes' is.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
select.refineEdge takes `edits`: the Select and Mask workspace's Quick
Selection, Brush, Lasso, Object Selection, Select Subject, Clear Selection
and Invert, applied in order to the starting selection (or layer mask)
before the edge is refined. Cancel discards them, OK is one history step,
and agents drive them over MCP and the control channel.
- engine/select_mask_edits.rs: parse (bounded, every bad field an Err)
and a pure apply/apply_one the workspace preview can share
- algo matting::brush_region: soft round brush coverage (hardness)
- ATN export reports steps with edits instead of dropping them
- "Edits" label in the 21 catalogs
Part of #7069 and #8148 (the workspace UI follows in a second PR).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
round_i32 added 0.5 in f32, which rounds up at the binade crossing:
0.49999997 (0.5 - 2^-25) + 0.5 becomes 1.0, so round_i32 gave 1 (and -1 for
the negative) where f32::round gives 0. The sum is now taken in f64, where
it is exact for every f32 the i32 cast does not saturate; still no roundf call.
rounding_matches_the_standard_library now covers +-0.49999997, the ties and
8388607.5, and a new test compares every f32 bit pattern in [-4096, 4096]
against f32::round (0.6 s). Both fail on the previous head. The code comment
and docs/oil-paint-performance.md no longer claim the f32 sum is exact.
Co-authored-by: ebosido444-ai <316167330+ebosido444-ai@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Main merged #5940, #6472, #7838 and #7922 directly. Kept from the batch:
the snapshot/History-follow identity at the history limit (#5940,
c7ef3676a; main still had the 3-tuple HistorySpot) and the strokeless
shape at a 0 px width step (#6472, 962539607; main still added the
stroke). Main already carries the other two fix-ups (34c9d0ea4, dff8bfd0e).
Overlaps with main resolved:
- #5842 vs main's #5568: one path, apply_color, tool-gated as #5842
has it, with #5568's float fill, kept fill opacity and lock skip;
shape_fill_ui's duplicate foreground_changed/gesture dropped; #5568's
tests run with a shape tool.
- #6230 vs main's #5180: one Type tool path click (path_target), no
second type.attachToPath; vertical type does not start type on a path,
as main has it; main's path_at/path_cursor and canvas arm dropped.
- #5452 moved the shape swatch to shape_paint_ui.rs: main's hex field
(#6201) ported there.
- type_extra_cmds::with_text takes main's dpi form, pub(crate) for
type_area_cmds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every tests/*.rs file is its own executable that links the whole crate
graph. Move each crate's test files into tests/it/ as modules of one
tests/it/main.rs; helper modules under tests/common move with them.
These keep their own binary (own process):
- engine panic_hunt: a process-wide panic hook counts every panic.
- engine text_gamma: changes the process-wide text gamma.
- engine smart_filter_stages, mask_cache_purge: assert on global caches.
- codecs pnm_peak_memory: reads the process's peak RSS.
- photocraft mac_menu_appkit: harness = false.
docs/gaps.md: header and change-log row once each (the PR's row on top).
docs/ui-parity.md: header and change-log row once each; the Selection row keeps main's #6549 sentence and adds the PR's #6548 sentence.
crates/ui-egui/src/puppet_ui.rs: tests added by both sides in the same place; kept both whole (#6754's pin-deletion tests, then #6786's drag-preview tests).
The line integral convolution takes 21 bilinear samples and 20 tangent lookups per
output pixel, and every one of those computed its coordinate with `f32::floor` or
`f32::round`. On the baseline x86-64 target both lower to library calls (`jmp floorf`
and `jmp roundf` in `rustc --emit asm`), so the filter spent most of its time in 82
call-and-return pairs per pixel.
`sample()` now truncates the coordinate it already clamped to the window, and the
tangent and bristle lookups use `round_i32`/`round_index`, which produce the same
integer for every position a tile can produce.
The output is bit-identical: six 600x400 fixtures (default, short, long, unlit, RGB
without alpha, CMYK) compare sample for sample with `cmp` and a per-channel maximum
delta of 0.0. 24 MP Oil Paint is 1.88x faster on the minimum of five alternating pairs,
measured on a contended 4-core machine; docs/oil-paint-performance.md has the numbers,
that caveat and the alternatives that were measured and rejected.
With --reps > 1 the console line printed the last sample and called it the
median. Sort the samples and print the median (mean of the two middle values
for an even count); the JSON report is unchanged.
The four new scenarios also made docs/scorecard.md stale, which failed
'test (ubuntu-latest)': cargo xtask scorecard now lists P64-P67.
bench_filters2 timings for the heavy algo filters were invisible to the perf
harness: the example was never listed in perf/budgets.toml, so no budget, no
baseline and no regression check covered Oil Paint, Relight, Reduce Noise,
Smart Blur, the blur gallery, Crystallize or Color Halftone.
The example now takes --reps, --quick and --json (the report shape the harness
reads, hand-written so the crate does not grow a serde dependency), and four
scenarios track the worst rows at a 1 s target: Oil Paint 8.8 s, Smart Blur
r12 11.6 s, Relight 3.7 s, Reduce Noise 2.8 s on a 24 MP layer.
Conflicts, both sides kept: main's .act swatches next to the batch's .asl
styles (Open filters, preset extensions), main's Render smart filters next
to the batch's Distort ones (53 mapped), parity doc status lines.
docs/scorecard.md regenerated.
Follow-ups the merge needed: the Distort smart filter readers pass their
enum type to main's stricter enum_name (the type their writers use), and
the CMYK effect tests set main's new StrokeFx overprint and Bevel
gloss_anti_alias.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Every new Camera Raw preview ran three per-pixel passes over the 447k-pixel
proxy on one thread: the RGB histogram (2.7 ms), the clipping warning
overlay (1.5-2 ms) and the selection mask blend (0.7 ms). They now run with
rayon (plain loops on wasm32). Histogram blocks add with saturating integer
sums, so the counts equal one pass.
camera_raw_bench, Core Ultra X7 358H: interactive scope update (P36) p95
11.0 -> 8.1 ms (budget 10), exposure proxy update (P31) p50 6.1 -> 4.1 ms.
Part of #211.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CD2ipSyvx1tYwnHJZyDs59