egui placed an upward list's bottom by the size it had the frame before, so
it moved up by ~46 pt as it settled and a click two frames after opening
missed the row (main's intermediate_ui_scale_choices test). Its top is now
placed from the rows' measured height plus the popup frame's margin.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
egui scrolls only by `TouchPhase::Move` wheel events; a trackpad's Start, End or Cancel
event with a delta no longer pans the canvas.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
wasmi 2.0's strict limits divide the code section's size by its function
count before validating it, so a module whose code section declares 0
functions but more than 1000 bytes (crafted, truncated inside the section,
or a duplicate code section) panicked in Plugin::load. Check every code
section for that one case before handing the bytes to wasmi and return
Error::Module; everything else is still left to wasmi.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Drags a Free Transform edge with real pointer events in the offscreen
app and checks the W/H readout in pixels, centimeters and percent, and
that the old percent readout is gone for the other units. Set
PHOTOCRAFT_TRANSFORM_READOUT_EVIDENCE_DIR to save the frames as PNGs.
Also note on Readout::Scale why the size is the scaled width and height
rather than the skewed edge lengths.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The tests built the expected explorer.exe / reg.exe path from SystemRoot with unwrap(),
so they panicked when it was unset and ignored the C:\Windows fallback for an unset or
relative value. The rule is now a small pure function windows_root(value) in each crate;
the tests use it and check the absolute, unset, relative, rooted-without-drive and empty
cases without touching the process environment.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
While a handle is dragged, the W/H readout beside the pointer showed
the scale in percent whatever the Units & Rulers unit. It now shows the
box's transformed size in the ruler unit at the document's resolution,
as Photoshop does, through the marquee's `size_readout`. With the
rulers in percent it keeps showing the scale of the original box,
since a percent of the document size means nothing for the box.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A character the text's font lacks made usvg's fallback search ask every
face in the database whether it has the character, and for a face loaded
from a file (every system font) that reads the whole font file. Once per
text element: 20 elements with U+10FF00 took 3 s with 273 installed faces.
The fallback selector is now our own copy of usvg's loop (same order,
used-face exclusion and style filter), and the per-face check answers
from a per-process cache of each file face's cmap table, picked as
ttf-parser's Face::parse picks it (last record of a linear walk, so an
unsorted table directory gives the same table) and looked up the same
way. Bounded at 64 MiB; a file that doesn't load isn't remembered.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Reveal in Explorer and the CJK locale lookup started `explorer` and `reg`
by bare name, and Windows looks in the app's own folder first: an
explorer.exe or reg.exe next to a portable PhotoCraft ran instead.
Both now start from %SystemRoot%.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The dialog now starts at the quality and Format Options it was last answered
with, also after a restart, as Photoshop's does. The answer is kept in the
remembered dialog settings (prefs.dialogs) when OK is pressed; Cancel keeps
nothing, and a stored value that no longer fits is ignored.
An opened JPEG still starts at its own quality (File > Save keeps it, #4032),
with the remembered format; leaving that quality as it is does not replace the
quality remembered for other documents.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Localization keys such as $$$/Presets/Shapes/Tiger=Tiger were kept raw, so the Shapes panel showed the key. File → Open now reports the group the importer stored, including a suffix when that name was already taken.
commit_shortcut skipped the highlighted row and fell back to the first
match, so with the current family highlighted it applied a different
family than the menu showed. Both paths now share accept_row (exact name,
then the highlighted row, then the first). Also remove the blank line
that split the ui-parity revision-history table.
Refs #5426. Follow-up to #5499.
Cause:
While #5499 parallelized local_blur and the per-dab mix in DabTool::Focus,
local_sharpen in crates/algo/src/retouch.rs remained single-threaded.
In addition, protect_detail redundantly evaluated coordinate clamping
and strided memory lookups across 8 neighbors inside the per-channel
inner loop for every pixel in the dab. On large brush sizes (e.g. 1000–2500 px),
strokes took seconds and stalled the application.
Change:
- Process local_sharpen window rows in parallel with Rayon across threads,
with wasm32 serial fallback matching local_blur.
- Hoist protect_detail 3x3 neighborhood offset calculations outside the
channel loop.
- Fast-path interior pixels (x > 0 && x + 1 < w && y > 0 && y + 1 < h)
with direct stride offsets, skipping boundary clamping entirely for >99.9%
of pixels in large brush dabs.
- Add retouch_sharpen_exact integration test verifying bit-for-bit parity
against the frozen scalar implementation across layouts, edges, and channels.
Review (CodeRabbit on #9618): deserialising the whole performance section strictly meant one
wrongly typed value (say "cacheLevels": "four") reset Graphics Adapter to automatic at startup,
while the app, which loads the file through read_lenient (#6611), kept the user's adapter.
read_display_server had the same flaw.
prefs::read_section exposes the engine's lenient reader for one section; gpu_startup's
read_performance uses it for both the graphics adapter and the Linux display server. Tests: a
valid adapter and display server survive a malformed sibling value.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>