- The free-memory check ran for every flat TIFF on Linux with a 512 MiB
reserve, so small TIFFs that open on main were refused on a small VM or
in a container: it now runs only for sizes the whole-image decode
refused (over 2^30 px or its byte budget).
- In a cgroup, memory.current counts page cache the kernel reclaims
before memory.max is hit; its inactive_file part now counts as free.
A `docker --memory=2g` batch whose cache filled up refused every TIFF.
- A single-strip TIFF is one group, the whole image, and every converted
piece of it was held at once: conversion now runs in rounds of at most
32 MiB of source rows per core.
- TiffBands::open reported no ExtraSamplesDropped/OverLimit warnings on
32-bit targets (wasm); both decoders now share prepared_warnings, and
Truncated goes before MorePages by position.
- Drops the ar.tsv rows of "Fix CI after merging upstream": main added the
same two keys (#10173), so the merged catalogue had duplicates and
i18n::tests::bundled_catalogs_are_consistent failed.
- Doc comments: the swap cap is every cgroup's above too (the ancestor
walk FredApps asked for); a cut-off strip decodes as zeros.
Co-authored-by: Khawar Zamman Wani <28915582+wanikhawar@users.noreply.github.com>
Co-authored-by: FredApps <153173854+FredApps@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
background_save_outcomes_report_failure_then_recovery failed on main's Windows
integration job: the real (second) autosave write flushes every tile to disk and
took over 30 s on a busy runner (normally 2-3 s; a concurrent autosave test ran
past 60 s too). The bound only exists to turn a worker that never reports into a
failure instead of a hang, so it is now 5 minutes, still under the 10 minutes
after which nextest's ci profile kills a test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conflict: crates/engine/src/atn_cmds.rs was split into cmd/atn_cmds/ on main.
Kept main's tree there: the PR's set_to guard for a `layers` list is dead on
main (the MAPPED path already refuses it), and its new test in
layer_props_tests.rs repeats main's own refusal test, so both stay out. The
stricter corpus round-trip in tests/it/corpus_atn.rs comes in as written.
Conflicts with main's module moves resolved: driver_printable takes main's
spooler-held settings_for_layout (which already wraps the page query in
with_spooler) and drops the now unused settings/pg lines; take_turned_page and
its test move to dialogs/file_ui/print.rs and dialogs/file_ui/tests.rs;
crates/engine/src/cmd/print_cmds/tests.rs keeps main's version (the PR did
not change it in its own commit).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A double-click on a filter row of a smart object that isn't the active
layer queued `layer.select` for the Layers panel to run after the rows
were drawn, but opened the filter's editor in the same pass. The editor
captures the document revision as its target, and the later select
raised it, so OK on the reopened Liquify filter (and on the generic
parameter form) was refused with "The document or Smart Filter changed"
and the Liquify strokes were lost. This happens when both clicks reach
one frame (a real mouse on a slow frame).
The double-click now selects the layer first, as the Layers panel runs
the select (committing an open type session), and drops the queued one.
Two Layers panel tests deliver both clicks in one pass; both failed
without this.
Co-authored-by: .CooCoo <1206598+dotCooCoo@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Audited all 22 catalogs (83886 entries) for duplicates, malformed rows, placeholders, plural form counts and ellipses. Keep every test and failure exit status; disable fail-fast so a single run exposes remaining regressions.
The typed zoom glided through a leftover log distance in egui temp data that only Fit/Fill and a
scrub cleared. Any other zoom right after Enter was multiplied by what was left: type 1600 at
100 %, Enter, then View > Actual Pixels ended at 87 %, not 100 %. The glide also never landed
exactly (1600 % ended at 16.000002, 6400 % past pixel_aligned's tolerance) and changed the field's
value by a few ulps each frame, which wiped text typed during the glide.
- set_from_field sets the zoom to the target and calls View::request_zoom_animation with
General > Animated Zoom, as Zoom In/Out and Actual Pixels do: it glides with the preference on,
jumps with it off, lands exactly, and any later zoom replaces it. The private glide, its DocId
bookkeeping and field_percent go.
- A zoom field keeps the keyboard through a click on a menu title, so the menu's arrow and letter
keys were dead (menu_nav/menu_keys treated it as typing in the menu). They work now, except in
Help, whose Search field keeps its letters (#5996). A letter or Space the menu takes no longer
also reaches the zoom field as text (an alert, or a digit typed into the zoom).
- Enter in the field within two frames of a click on the pasteboard, title bar or a menu was
vetoed and the typed zoom lost; egui draws only those frames with the mouse still.
- The separator check's copy of the text restarts when the value changes under the focused field
(ArrowUp, another zoom), so `1.5`, ArrowUp, End, `.` no longer alerts.
- Doc comments: the Esc/focus path drops only the text copy; no platform sets Services::alert.
Co-authored-by: sn-o-w <2406819+sn-o-w@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Its translations are part of the combined catalog; this merge records the
branch so the PR closes as merged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Its translations are part of the combined catalog; this merge records the
branch so the PR closes as merged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Its translations are part of the combined catalog; this merge records the
branch so the PR closes as merged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Its translations are part of the combined catalog; this merge records the
branch so the PR closes as merged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Its translations are part of the combined catalog; this merge records the
branch so the PR closes as merged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The estimate charged every tile a solid layer can reach at full size, but a solid fill stores
one tile shared by every covered position and Surface::translated shares it again at any
offset, so only the edge is written. A filled 12000 x 12000 16-bit layer was estimated at
about 1.1 GiB and lost the live drag it has on main. The estimate now mirrors translated:
one target per 2 x 2 block of source tiles, and a block inside the content with four stored
tiles is charged once per distinct set of tiles, capped by the tiles the content can reach.
It also counts a smart object's filter mask, which shift_smart shifts too.
New tests: a filled 12000 x 12000 16-bit layer stays live through the real drag path (fails
at the PR head); the filter mask is counted; content off the tile grid is capped by its
reach. A 36 MP 8-bit and a 24 MP 16-bit photo layer stay live through the drag path (pins
the common case; fails with the earlier x4 estimate).
Co-authored-by: Max Witt <131544883+SiFuMax09@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fifteen rows for the Select and Mask workspace (view and refine modes,
Color Aware / Object Aware, Edge Detection, Global Refinements,
Decontaminate Colors, the Refine Edge Brush Tool and the two messages
shown when the workspace closes), so `cargo xtask i18n-coverage` reports
Swedish complete again (2,202 / 2,202).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Main's noop_step now takes the step's params (an Arrange that cannot move the layer is a no-op too); it is public for the CLI's run and batch, which pass them.
- Smart-object contents (Replace Contents, Convert to Layers, Stack Mode,
internal Edit Contents) refused any SVG whose canvas area times its
node count passed 24 MP, e.g. a 1920x1080 layout with 12 shapes.
decode_source now uses the bounded image parse with File > Open's
layered import (import_svg_bounded_images). Convert to Layers' shape
cache cap is 1 GiB in the document's format, the import's own budget,
not 24 MP (four full-canvas shapes in a 4K document failed).
- The bounded resolver failed the whole drawing on an embedded image it
can't read (an unknown MIME type, a corrupt PNG); it now leaves it out
as usvg does, so such SVGs place and render again. Byte and decode
overruns stay errors.
- resvg gives each embedded SVG image a scratch pixmap the size of its
target and frees it after, so the scratch budget counts the deepest
chain of nested SVG images, not every sibling: 12 SVG icons on a 2 MP
canvas, or one at a 12 MP render, no longer fail.
- Main's shared nesting budget (#9397) now covers the bounded path too:
511 + 511 levels across a drawing and its embedded SVG overflowed a
1 MB stack.
- An AI working copy is regenerated through VectorCraft only when
VectorCraft is the editor; Illustrator or another program gets the
original bytes. VectorCraft is recognised by bundle id, a known
install, or its name (vectorcraft[.exe], vectorcraft-<version> AppImage,
ai.storyteller.vectorcraft).
- updateSvg refuses an AI/PDF/EPS source (updateVectorSource saves it
back in its own format); the doc no longer calls updateSvg "existing".
- Edit Contents on contents already open in the internal editor switches
to that document, as main does, instead of an error.
- Stack Mode on an SVG refuses up front when one float frame per layer
would pass 1 GiB (the layered import no longer refuses for it).
- A stored display preview still fails on an embedded image it can't
read (strict budget, its own import_svg_preview_with_work; external-
editor saves and conversions leave the image out as placement does); skipped embedded SVGs still use up the XML byte
budget; an image under an unknown MIME type or with no pixels is left
out uncharged.
- Merge: the PSD import reads a preview's native source from the
document's global blocks; main's owned open moves them out of the
file, so every stored preview was dropped.
- Main's strict Clippy rules on the PR's app code (explicit import, a
Unix-only test helper, one long opt-in test).
Co-authored-by: Mocky Digital <282319807+mockydevs@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The bounded parser for embedded SVG checked each nested SVG image
against MAX_XML_DEPTH on its own. Nested images are parsed on top of
their parent's recursion, so 32 levels of images could reach the XML
parser about 16,000 elements deep, past what its stack holds. Pass the
depth the enclosing drawings leave down to each nested image, as main
now does for the ordinary parser.
Resolve conflicts with main at c645d5fc:
- svg.rs: keep the bounded rasteriser on main's new render() helper,
keep main's rasterize_rgba8, and inflate with bufread::MultiGzDecoder
under the caller's limit.
- psd_import.rs: move the vector preview into main's smart_content().
Main now moves global block data out of the file before layers are
read, so look the native source up in the importer's own copy.
- smart-object rendering, shortcuts, menu dispatch, strokes: take main's
structure and keep the editor intercept and modal checks.
- Parity docs and Cargo.lock: keep both sides.
Adapt to main: explicit imports for the new wildcard_imports lint, a
smaller real-editor acceptance test for cognitive_complexity, the new
Group::frame_stroke field and options()' depth argument.
Adds a Finnish (Suomi, `fi`) catalog with 4,169 entries: every key of
the other bundled catalogs, written in sentence case from the English
labels with ordinary Finnish image-editing vocabulary. It is registered
with complete_menus, and xtask's source-key coverage check includes it.
Plurals are one/other with the partitive singular after any number but
1 (1 taso, 2 tasoa).
Any fi-* locale selects Finnish; sv-FI does not. Tests cover locale
tags, system-language resolution, id and context overrides, and plurals.
Clean-room, AI-assisted (Claude Code) translation; a native speaker
checked it in the running app, but it has not been proofread line by
line.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#8896 added 17 strings to every catalogue it knew; ar.tsv landed just
before it and got none, so main failed
the_reason_a_stale_workspace_closed_is_translated. Adds the 16 missing
rows (Remember Settings was there), wording as the catalogue already has
it (Select and Mask, Refine Edge, Content-Aware).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* perf(algo): row-parallel local_sharpen with fast-path protect_detail (#5426)
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.
* bench(engine): add bench_sharpen_tool example
Conflicts: svg.rs (main's bufread inflate and shared render helper, kept
with the PR's bounded raster path), psd_import.rs (the PR's vector
preview moved into main's smart_content), menus/invoke.rs (the SVG
intercept moved into main's split invoke_unguarded), rendering.rs and
the ui-egui import lists (main's explicit imports), Cargo.lock, docs
status lines (unions).