composite_area already honours lock transparency, but paint_gradient and
bucket_fill(_src) always passed false, and paint.bucket / paint.gradient
dropped the flag target_surface returns. Thread it through, as Edit > Fill
and the brushes already do.
Co-authored-by: prekabreki <prekabreki@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The single-layer path removed the layer with no guard, leaving a document
with no layers and no active layer. It now refuses with the same error as
the multi-select path; the edit is discarded, so nothing changes and no
history step is recorded.
Co-authored-by: prekabreki <prekabreki@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Clear, Cut and Image > Adjustments refuse a pixel-locked layer (#1103)
clear_area, destructive_adjust and auto_adjust took the layer's surface
directly. They now go through commands::paint_surface, which refuses a
pixel- or fully-locked layer with the same message painting uses.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* edit.cut: refuse a pixel-locked layer before filling the clipboard
copy() ran before clear_selected noticed the lock, so a refused Cut still
replaced the clipboard. Check the lock first (check_pixels_unlocked, shared
with paint_surface) and assert the clipboard is unchanged in the test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: prekabreki <prekabreki@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* SVG opens as shape layers and places as a vector smart object (FILE-215-9, part of #215)
crates/io/src/svg.rs: usvg parses (already in the tree through egui_extras); paths become
ShapeLayer knots with the transform chain baked in, quadratics raised to cubics and every
subpath after the first joining the component so the fill rule spans the path as SVG does;
groups keep opacity and blend mode (transform-only and id-only wrapper groups vanish);
text is shaped into outline shapes with the craft fonts plus the system's; gradients become
the layer's gradient fill (stops and angle, with a warning). Clip paths, masks, filters,
pattern paints and images are rasterised by resvg into pixel layers over their bounding box
rather than dropped. Limits: 16384 px canvas, 2000 elements, 24 M px per raster, bounded
recursion; malformed input is IoError::Svg.
smart_cmds::render rasterises an SVG source at the placement's axis scale (1/16 steps,
cached per scale) and divides the placement by it, affine and projective alike; place_bytes
renders that cache up front. svg/svgz join the openable lists; book page; scorecard row done.
Co-Authored-By: Claude Mythos 5.1 <noreply@anthropic.com>
* SVG import: never read local files from <image href>
usvg's default image resolver reads any absolute or relative path in an
<image href>, so opening an untrusted SVG could embed local files
(/etc/..., ~/.ssh/...) into the document. Parse with a resolver that
only accepts data: URLs (resources_dir None, string hrefs ignored).
Embedded sub-SVGs already get no <image> resolution from usvg. The font
database (and its system-font scan) is now only loaded when the drawing
may contain text. Regression test: a PNG on disk referenced by absolute
path, file:// URL and relative path is not loaded; a data: URL still is.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: leon <claude.udder842@passmail.net>
Co-authored-by: Claude Mythos 5.1 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* Filters refuse a pixel-locked layer (#1102)
run_filter never consulted layer locks. It now refuses a pixel- or
fully-locked layer with the same message painting uses. Channel, Quick
Mask and mask targets are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Filters keep alpha under lock transparency; render filters and Fill Path refuse a pixel-locked layer (#1102)
- run_filter puts the layer's alpha back in the tiles a filter changed when
transparency is locked (the same tile walk as restrict_to_color).
- Flame / Tree / Picture Frame take the surface through paint_surface.
- path.fill refuses a pixel- or fully-locked layer.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: prekabreki <prekabreki@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* fix(engine): leave empty selection empty for subtract and intersect
* test(engine): cover empty selection subtract and intersect modes
* test(engine): read selection coverage through Surface API
* test(engine): cover selection subtraction history and format regressions
* fix(algo): Path Blur clamps per-path speeds (#708)
filter.blurGallery.pathBlur clamps the scalar speed to 0..500, but
each entry of the documented paths array was deserialized without any
clamp. The halo takes the maximum speed and computes
Radius(v.max(0.0).ceil() as i32 + 1): a large finite speed saturates
the float-to-int cast to i32::MAX and the + 1 overflows, so the command
dies in a caught panic (and wraps to a negative halo with overflow
checks off).
Clamp each entry speed to the same 0..500 the scalar path uses, right
where the params are built.
* Path Blur test: drop an unused before-sample (a flat background can't change under blur)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Fenty <bfenty@users.noreply.github.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
filter.cameraRaw deserialized its params verbatim, so an out-of-range
noiseLuminance reached the denoise stage and became a guided-filter
radius of round((1 + 3 * n/100) * pixelScale). A value of 1e9 turned
six box-mean passes into unbounded work: the command never finished.
noiseColor had the same defect through its blur sigma.
raw_params now clamps the four noise fields to the documented 0..100,
matching how the sibling lens commands clamp their ranges at the same
boundary. The in-range path is untouched.
Co-authored-by: Brandon Fenty <bfenty@users.noreply.github.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
* Camera Raw: read and write settings as a live PSD Smart Filter (part of #215)
- Map Camera Raw 18.4 (process 5/6) filter settings to and from PSD: custom white
balance, Light, Presence, parametric and four point curves, HSL, Color Grading,
sharpening, noise, grain and numeric post-crop vignette controls.
- Keep the original Photoshop descriptor; edits patch only changed fields. Unknown
settings or versions stay an opaque filter, preserved byte for byte.
- PSD stores integer slider levels: export rounds continuous PhotoCraft values
(hue 360 -> 0) instead of dropping the filter.
- Tests: field mapping, patching, opaque fallback, malformed input, rounding, and
8/16/32-bit PSD -> edit -> .pcraft -> PSD with mask, filter order and undo/redo.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Camera Raw PSD: use as_chunks for curve pairs (clippy 1.99)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Camera Raw PSD: no product names in Rust comments, strings or test names
Only the descriptor class key itself keeps its literal name.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
columns/rows came straight from params with no cap. layout_lines loops
once per requested line and each push_unique scans the growing guide
vector, so a count like 200000000 (well inside u32) never returned -
blocking the app, CLI, MCP and the TCP control channel.
Reject counts above 1000 with an actionable BadParams naming the
offending parameter. Photoshop's own dialog caps at 32; 1000 is a
generous ceiling whose worst case finishes instantly, so legitimate
grids are untouched.
Adds regression tests: the issue repro (200M columns, and rows)
returns an actionable error immediately, and bounded layouts at the
cap still produce their guides.
Co-authored-by: Brandon Fenty <bfenty@users.noreply.github.com>
The field documents "1 = neutral; lower = more contrast", but the tone
map used gamma directly as the powf exponent, so 0.5 lightened
midtones and 2.0 darkened them - the slider ran opposite to its own
contract. The crate's other two gamma controls (hdr.rs, proof_sim.rs)
already use the reciprocal exponent.
Apply 1.0 / gamma like they do. Neutral (1.0) is unchanged; every
other value now moves midtones the documented way.
Adds a regression test on a flat 50% grey: gamma 0.5 darkens the
midtone, gamma 2.0 lightens it (reversed on main: 0.71 / 0.25).
Co-authored-by: Brandon Fenty <bfenty@users.noreply.github.com>
uses_colours() derived from the params notation, but Colored Pencil
(paper tint from the background) and Neon Glow (foreground mixed over
background) read the colours in their pixel code without listing them.
The single-filter commands therefore skipped colour injection and both
filters always ran with the default white/black regardless of the
session colours, while filter.filterGallery honoured them.
Add the two filters to uses_colours() at the source, so the command
injection, the gallery dialog note, and smart-filter re-renders all
agree.
Adds regression tests: filter.gallery.coloredPencil and
filter.gallery.neonGlow output must change with the session
background (both identical on main).
Co-authored-by: Brandon Fenty <bfenty@users.noreply.github.com>
gauss_box_radii followed Kovesi 2010 exactly, and the rounded triple
degenerates below sigma ~0.577: wl floors to 1 and all three radii
become 0, so every box pass early-returned and gauss_blur_n handed the
buffer back untouched for any sigma in [0.3, 0.577). Callers with
documented small sigmas (Camera Raw sharpenRadius 0.5 and colour noise,
HDR Toning radius 1, Shadows/Highlights mask blur, Oil Paint
cleanliness) silently did nothing in that band.
Clamp the degenerate case to one radius-1 pass - the smallest real
blur - which only engages where the ideal box widths are already at
their floor, leaving every sigma the paper handles well unchanged.
Adds the command-level regression test from the issue: cameraRaw
sharpenRadius 0.5 must change pixels (control: radius 3.0).
Co-authored-by: Brandon Fenty <bfenty@users.noreply.github.com>
* Add rendering policy with legacy preference migration and safe adapter selection
* Add rendering mode preferences and GPU fallback recovery warning
* Complete CPU fallback recovery and validate rendering mode UI
* Simplify GPU recovery and reuse preference persistence
* Rendering modes: rebase onto the preferences merge/retry save path; drop docs screenshots
save_preferences now writes through write_prefs (merging other instances' changes) and
only marks the revision saved when the write succeeds, so a failed recovery choice is
still retried. Removes a leftover diff3 marker in es.tsv, the docs/images/rendering-*.png
screenshots and their ATTRIBUTION row (no binary assets in the repo), and regenerates
docs/scorecard.md for the new performance.renderingMode preference.
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>
blend_text_gamma_setting switched the compositor's process-wide text gamma
(psblend::set_text_gamma) to 1.0 and back while other engine unit tests ran
in parallel threads. turns_at_every_depth_and_mode flattens a document with
a type layer before a turn and again after edit.undo; when the toggle landed
between the two flattens, anti-aliased glyph edges blended under different
gammas and the bit-exact comparison failed (seen once on Windows CI).
Move the gamma test to crates/engine/tests/text_gamma.rs (its own process),
document the hazard on set_text_gamma, and make the undo assertion report
the turn, depth, mode, differing pixel count, max difference and first pixel.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Right-click opens the Brush Preset picker (or erases), eraser-on-Background tests, README comma
- #75: the right mouse button never paints with the tool. With a painting tool it opens the
Brush Preset picker at the pointer (Photoshop). New preference Preferences › Tools ›
Right-click with painting tools (tools.rightClickWithPaintingTools: brushPicker | erase,
serde default brushPicker): with erase, a right drag with the Brush erases with the current
brush (one paint.stroke, one undo step, pen pressure/tilt as usual). ui.pointer accepts
"button": "secondary". New module ui-egui/src/paint_mouse.rs; picker body shared with the
options-bar chip.
- #76: the engine already paints the background colour when erasing a transparency-locked
layer (the Background); added tests at 8/16/32-bit in RGB/CMYK/Gray, with a selection, on a
normal layer (erases to transparency), on a transparency-locked layer, and through the UI.
- paint.stroke/pencil refuse coordinates beyond ±1e6 (an f64::MAX point ran out of memory).
- #73: test that the canvas hands the live stroke over to the committed document with no
frame lacking the stroke, refreshing only the stroke's rect.
- #69: README comma.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Brush smoothing follows the options bar; the live stroke draws the catch-up tail (#73)
- The options-bar Smoothing % (pro and classic bars, Brush and Eraser) edits the session
brush's smoothing, which the live stroke and the committed paint.stroke both use; the
canvas no longer forces 0.3. The session brush starts at Photoshop's 10 %.
- With smoothing the brush lags the pointer and catches up when the stroke ends. LiveStroke
now draws that catch-up tail on every step (StrokeRenderer::tail_preview renders only the
tiles the tail touches; the previous step's tail is restored first), so the preview at the
last move equals the commit and nothing pops in on release. At 0 % there is no lag.
- Path › Stroke Path ignores smoothing (it strokes the exact geometry).
- Tests: paint tail_preview = finished stroke; canvas: 0 % preview = commit, 50 % preview =
commit with the end reaching the pointer, smoothing changes the stroke, commit equals
paint.stroke with that smoothing, no stale tails after sharp turns.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Smoothing is a per-tool option that brush presets keep
- Picking a brush preset (tools.setBrush / paint.* "preset", .abr import with select) keeps the
current smoothing amount and mode instead of the preset's (BrushSettings::picked_over),
as Photoshop's options bar does. An explicit smoothing alongside the preset still applies.
- The Brush and the Eraser each remember their smoothing (UiState smoothing_tool /
tool_smoothing, synced on tool switch before the options bar, the canvas and each stroke);
a tool seen for the first time starts at 10 %.
- Tests: engine presets keep 10 % / a user value with pulled-string mode; UI per-tool
smoothing across Brush/Eraser/other tools and on the first stroke after a switch.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
CI's `cargo fmt --all -- --check` has failed on every run because the
repo had no rustfmt config and was never formatted. Use the same settings
as EffectCraft/FilmCraft (max_width 160, use_small_heuristics Max), which
also keeps the churn small. Formatting only; no code changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
warp_gray cleared the old + warped region as one dense buffer spanning their
union, so moving a selection/mask/channel far away (e.g. dx=1e5) allocated a
~1e5x1e5 buffer and hung the app. Write the two regions separately instead —
Surface is tile-sparse, so a distant region costs only its own tiles. Found by
the panic_hunt fuzz (now also catches hangs via a per-command thread timeout).
Regression test covers a far transform.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Rule 9 (no panic on input). Fixes found via fuzzing:
- edit.paste: a short/non-array "center" (e.g. [5], []) indexed a[1] and
panicked; now requires len>=2, else falls back to the canvas centre.
- edit.transform.again: a quad row with <2 columns indexed [1] and panicked;
now rejected with the other shape checks.
Tests: paste_with_malformed_center_does_not_panic (fast regression) plus
tests/panic_hunt.rs, an opt-in (#[ignore], run --ignored) fuzz that dispatches
every command with adversarial params and asserts none panic.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>