The gpu group (at most 4 at a time) was chosen by name: every ui-egui
integration test and every test whose path contained canvas, gpu, mip or
type_tool. On main's gate that put about 680 ui-egui tests that never
open a device on four slots.
The group is now photocraft-gpu plus every test in a module named
`gpu_tests` or `<name>_gpu_tests`:
- ui-egui integration tests that draw on a device move to
tests/it/gpu_tests/; mixed files and unit tests keep their GPU tests in
a nested `gpu_tests` module or a `*_gpu_tests` file.
- photocraft-automation's GPU backend tests are `gpu_backend::gpu_tests`.
- batch_export_options checks menus and exported bytes, so it runs on the
CPU canvas instead of a GPU harness.
No test assertion changes. The Windows priority list follows the moved
test names.
The gate's archive of ~98 test binaries would be 10-18 GB compressed, so
every shard restores the dependency cache, builds the tests and runs its
partition (scripts/ci-test-shard.sh). `cargo xtask ci tests` runs the test
steps on $NEXTEST_PARTITION, `ci build` builds exactly what a shard builds
(main's cache warm-up), and `ci doctest` runs in one shard. The `ci archive`
part, the archive's xtask binary and the extraction are gone.
The Linux GPU shard keeps the gpu group alone on lavapipe; shards 2-4 split
the rest. Windows uses six shards: ~55 min of tests on one runner, and its
slowest test alone takes ~9.6 min, so a seventh would not shorten the
slowest shard.
- xtask starts every cargo without the variables `cargo run` sets for xtask
(CARGO_MANIFEST_DIR, CARGO_PKG_*, ...), so a build script that declares
`rerun-if-env-changed` for them (ring) no longer reruns, with the whole
craft facade above it, between builds inside and outside xtask.
- `cargo xtask ci` splits into the gate's parts without a second step list:
`lint` (fmt, egui-fork, clippy, layers, panel-commands, scorecard, wasm,
i18n report), `archive` (test build into the nextest archive, doctests),
`tests --from <dir>` (test and panic_hunt from the extracted archive, on
$NEXTEST_PARTITION; --gpu / --no-gpu select by the gpu test group) and
`build` (cache warm-up). Plain `cargo xtask ci` still runs every step.
PHOTOCRAFT_CI_EXCLUDE leaves packages out of the test build and doctests.
- The `ci` nextest profile's archive carries the xtask binary, so a shard
builds nothing (scripts/ci-test-shard.sh).
- Linux gate: lint leg, one build, four shards. Shard 1 runs the gpu group
alone on lavapipe and checks its output (scripts/ci-gpu-lavapipe.py now
wraps that run); the second GPU run and its rebuild are gone.
- Windows unit, integration, clippy and warm legs move onto the shared gate
(Windows runner, lint leg for Clippy, four shards), keeping the PR
selection from the platforms job, rust-lld, /OPT:NOICF, /DEBUG:NONE and
the D:\t target dir.
- The slowest-tests-first priorities apply on Linux too.
- Swatinem/rust-cache v2.9.3 by commit for this repository's own cache
steps, with fresh keys; cache-version v2 for the gate and docs entries.
`test (without heif)` restores the gate's entry through the shared action.
Dispatch-only jobs no longer save caches no pull request reads; the quick
perf suite restores perf-nightly's entry.
- windows_delay_loads reads the app from NEXTEST_BIN_EXE_photocraft, as the
other app tests do, since archives relocate binaries.
PhotoCraft's UI moves onto the shared craft UI framework in storytold/craft-libs (next branch): shell, dock and panels, hosted dialogs, keyboard, focus, IME and pointer routing, the canvas's view commands, themes and tokens. The engine, document model, file formats, commands and automation stay. PhotoCraft pins storytold/craft-libs next.
- Without the AI-MCP flag plain `photocraft-cli mcp` is headless as before: a
Control Port saved before the agents switch no longer makes it look for the
app, start it or apply When PhotoCraft Is Closed. Test in
mcp_without_consent_works_headless (fails without the gate).
- `prefs::read_integrations_section` (migrate + lenient read) is shared by the
app's launch read and `photocraft-cli mcp`: one bad value resets only itself,
so a saved "error" closed-app choice no longer falls back to Launch in the CLI.
Replaces the app's hand-written per-field reader. Test in
integrations_come_from_the_saved_preferences_or_default (fails without it).
- docs/development.md: the headless client examples pass --headless.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Plain `photocraft-cli mcp` now works in the open window while the ArtCraft
Launcher's agents switch is on, and may start PhotoCraft, also on a
developer's machine. AGENTS.md says scripts, tests and agent sessions pass
`--headless` (and tests give the CLI their own settings, launcher and flags
folders); the headless examples in docs/development.md, docs/mcp.md,
docs/control-protocol.md, the book and the README use it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
As agreed in craft-ai-integrations#195: one switch for every Craft app, in
the launcher; only the launcher writes MCP client configurations; plain
`<app>-cli mcp` works in the open app.
- Consent: with the AI-MCP flag, the launcher's switch alone decides,
read through `craft-ai-agents` (craft-ai-integrations#197), the crate the
launcher writes the app list with and the gateway reads it with. The
"Let AI agents connect" box is gone; without the flag everything is as on
main (a saved Control Port keeps its connection).
- `photocraft-cli mcp`: works in the open window while agents may use it
(what `--live` did), and headless otherwise, exactly as before (only the
roots given to it; the app is never started). `--headless` forces that;
`--bridge` / PHOTOCRAFT_CONTROL_PORT as before. `--live` and `mcp install`
are gone. Several read roots work again, invalid roots stop the start, and
`--gpu` applies to the headless session.
- Client setup is the launcher's: the client code (Claude Desktop, Claude
Code, Cursor, JSON), Copy Configuration, its service and its docs are
removed (about 740 lines net).
- CLI tests run with their own settings, launcher and flags folders: plain
`mcp` would otherwise find (or start) the developer's own PhotoCraft.
- Docs, book, README, scorecard row, packaging comments follow.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Resolve public occurrence identity and main playback budgets, nested Batch context and Pen history together. Migrate the deleted local parser to main shared API while preserving public legacy-suffix refusal. No unpublished behavior fixes or local gate runs.
The Integrations page holds more than the agent connection: open PRs add the
AI services for Generative Fill and other AI features to it (#10903, #12391,
#7007). The menu item stays "AI Integrations…" and the page "Integrations", as
on main; messages and docs say Edit › Preferences › AI Integrations.
* Use craft-psd from craft-libs; remove crates/psd
* Regenerate the scorecard (28 crate roots)
* Take craft-psd at craft-libs 55d120f2
* Regenerate lockfile for current shared format crates
* Pin shared format crates to current craft-libs main
* Match brush import cap to shared decoder budget
* xtask: update layer fixtures after PSD extraction
The page only holds the MCP connection, so Edit › Preferences › AI Integrations… is now
Edit › Preferences › MCP… (command id and the `integrations` preference keys unchanged).
Copy Prompt for an Agent puts a text on the clipboard that an agent with a terminal
(Claude Code, Codex, Cursor's agent) pastes to register `photocraft-cli mcp --live` itself
and check the connection; Claude Desktop has no terminal and keeps Add to Claude Desktop.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Ported onto current main and #11363:
- The dialog moved into prefs/prefs_ui/agents.rs (prefs_ui.rs was split on main); the port
field is labelled for the accessibility check.
- One agent connection: Preferences start and stop it live (agent_link service); only
--control opens a port at launch. #11363's token reuse, folder rules (absolute paths,
warnings) and readRoot/writeRoot names are kept; a saved controlPort from before loads
with connectAgents on.
- agent_link.rs: #11363's connection file code plus #5525's client install code.
- CLI: mcp --live, --when-closed and mcp install --client on main's clap parser.
- control_server: listen()/Listening on top of main's challenge authentication.
- app_dirs: main's resolver moved to photocraft-automation so the CLI finds the same folder.
- Interface catalogs moved to craft-i18n on main; the PR's rows are not carried over.
Co-Authored-By: Furoxfufu <Furoxfufu@users.noreply.github.com>
Co-Authored-By: ingling <ingling@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
craft-cms was extracted from crates/cms (storytold/craft-libs#779), and both
copies changed afterwards. storytold/craft-libs#PRNUM ports the 41 commits made
here since the extraction, so the shared crate now has everything this copy
had, and PhotoCraft depends on it.
craft-cms returns errors where the local crate panicked on caller input:
transforms, tables and LUT constructors return Result. Call sites propagate
the error where the function already returns one, and fall back where the
arguments are fixed by construction (each with a comment saying why).
Lut3d's fields are private there; the proof and HDR display table is built
with Lut3d::from_fn.
The pin is the head of the craft-libs branch and moves to main when that
pull request merges.
Keep all 34 catalog deletions; sync crop options and fill rasterization text via craft-i18n 4a642c06dbcb8172f307f827c1167cd5dd8ac341. Adapt the fill prompt catalog test to embedded catalogs and preserve both parity-document updates.
The i18n module becomes a thin Lang wrapper over craft_i18n::I18n and the
catalogs embedded from craft-i18n. Plurals use trn_plain and PhotoCraft
keeps its own fmt; @id rows come only from PhotoCraft's own rows that keep
the label's shape. Detection, PHOTOCRAFT_LOCALE, tag bounds and the 35
offered languages are unchanged.
Serbian: sr, sr-RS, sr-Cyrl-RS and sr-ME reach sr-latn, the only Serbian
catalog, and a saved interface.language "sr" migrates to "sr-latn".
The menu completeness test becomes a report that never fails, and
`cargo xtask i18n-coverage` is replaced by `cargo xtask i18n-report`, which
reads CRAFT_I18N_DIR through the crate. Docs describe the new system.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The System Title Bar preference hands the frame to winit. On Wayland
compositors without server-side decorations (GNOME) winit draws the
decorations itself: the font-free Adwaita controls when its
wayland-csd-adwaita-notitle feature is on, sctk's plain fallback frame
otherwise. The winit dependency that turns the feature on sits in the
Linux-only block of the app manifest, so the FreeBSD build gets winit
through eframe alone and ends up with the fallback frame. The default
integrated title bar is not affected.
Add the same winit line for FreeBSD with just the decoration feature (the
x11 one serves the Xwayland pen path, which is Linux-only). sctk-adwaita and
its dependencies are already in Cargo.lock through winit, so the lockfile
does not change and sctk-adwaita is the only crate that joins the FreeBSD
dependency graph; the Linux, Windows and macOS graphs are unchanged. Name
FreeBSD in the NOTICE attribution and the development notes.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Keeps main's Fade preview path in filter_preview.rs and passes the PR's
tool colours to the non-Fade preview; catalogs keep both sides' rows.
Co-Authored-By: Firstbober <firstbober@tutanota.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Linux runner has no GPU, so the GPU tests ran on Mesa llvmpipe over
GL, which can't render Rgba32Float: 57 GPU checks (50 GPU parity, 3
canvas, 2 adjustment preview parity, 2 effect-map recovery) returned
early and passed without comparing anything. A new step after Test
installs mesa-vulkan-drivers and runs the nextest `gpu` group on
lavapipe (Vulkan), reusing Test's build, and fails unless the parity
tests report the Vulkan backend and no test printed a skip.
No code change: photocraft's tests take whatever adapter wgpu returns,
software adapters included. Follows the CI half of lightcraft#1148.
Co-authored-by: FredApps <FredApps@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* PSD write recipes: keep every exported candidate and pin the base file
A case that failed after export dropped out of manifest.json, losing the
file a reference-editor check most needs. The CLI also reopened the corpus
path after the hash was taken.
- Keep the candidate and its hash once export succeeds.
- Run the CLI on a copy of the hashed bytes, so it never opens the corpus.
- Compare numbers by value, so 17 equals 17.0.
- tracking-split also checks runs 0 and 2 and all three run lengths.
Part of #6130.
* PSD write recipes: compare large integers exactly, reject case-only id clashes
Converting to f64 made 9007199254740993 equal 9007199254740992. Ids that
differ only by letter case name one output directory on a case-insensitive
filesystem, so they are now rejected as duplicates.
* PSD write recipes: test that the CLI receives the hashed bytes
* PSD write recipes: reject the contour library commands
Their path param can read or write any file, including the pinned base and
the corpus, and they edit no document.
With 1 GiB or less of free RAM, or 512 MiB or less of commit headroom, the
hard limit was exactly 0, so even a small file was refused with a message that
didn't say memory was low.
The reserve now shrinks with the measurement: up to a quarter of free RAM opens
automatically, the rest goes to approval, and an approved import always leaves
at least 256 MiB free. Limits are unchanged for automatic opening above 4/3 GiB
free and for approval from 4 GiB free. A refusal set by measured memory now
gives the free amount and says to close other apps and try again, and the
approval prompt warns when little memory would be left.
An unreadable cgroup still measures Some(0) and stays refused.
The 32-bit message ended mid-phrase. It now says the cap applies or less if
less memory is free, and separates the cause from the next step. Windows
scratch tests check the returned numbers belong to the temp directory's volume.
PngImport decoded the whole image and then copied it into a second, tile surface (about
+400 MB for a 100 MP RGBA8 file, enough to kill a browser tab). A PNG whose rows come in
final order now streams: PngDecodeRows::step_stream hands each group of rows on, they
become tiles at once (the PngStream the one-shot streamed import also uses) and are
dropped, then the prune scan runs in batches. decode_rows is the same step_stream driven
to the end, so there is one path. Interlaced, animated and early-eXIf-oriented files are
decoded whole as before; a late eXIf sends a streamed import back to a whole decode.
The turn upright of an EXIF orientation 2-8 was one whole-buffer pass in the poll of the
last rows; OrientSteps (photocraft-codecs) does it in bands of 64 rows through the same
band writer as Image::oriented, and PngImport runs it as a stage of its own.
Memory-budget errors keep their type and text: the stream is checked as the one-shot
streamed import is, and, like the whole-image path, a layer under 32 MiB is not.
Docs for the stepped place updated.
The remaining commits of the stack, applied on top of the squash-merged #10147:
- PNG decodes in resumable steps of rows (PngDecodeRows, shared with decode_rows)
- PNG imports as a document in resumable steps (io PngImport, text metadata and
document build in bounded steps)
- Jobs can run a slice at a time where there are no threads (StepWork,
cooperative sessions, Session::poll_jobs)
- Place decodes, flattens and resizes in slices (PlaceSteps), reading a path
and keeping the frame, memory gate and typed memory-budget errors main has
- The web build places dropped and picked files a slice per frame, with a
bounded place inbox
- A measurement of the stepped place; docs for jobs without threads
* Raster: ask the allocator whether a big operation fits; PSD save reserves its whole output
New photocraft_raster::headroom: ensure(need) tries to reserve the need plus a margin and frees it
at once (web only; native always passes), below(bytes) tells whether less than that can be
allocated, and simulate_available fakes the answer for tests and diagnostics.
PsdFile::to_bytes reserved only the pixel data, so embedded smart-object data and other tagged
blocks grew the output by doubling, which on the web is one unrecoverable huge realloc. It now
reserves the exact encoded length.
* Big operations that won't fit in memory fail with an error instead of trapping
Place, Image Size, a smart-object re-render and Save now ask the allocator whether their estimated
need can be allocated before they start (photocraft_engine::memory_gate, estimates per operation,
photocraft_io::estimate_export_bytes for saves). A refusal changes nothing and says what the
operation needed and what to do (save, Edit > Purge > All, close documents).
A standing notice warns when less than 512 MiB can be allocated, with hysteresis, so the user can
save before anything traps. Notices can now be sticky, which keeps them past the auto-hide delay.
* Web build: a trap says so instead of freezing the page, and ?free-mb simulates low memory
index.html holds a hidden, static 'PhotoCraft stopped' message with a Reload link. web.rs reveals
it when the module traps (the window error event, or an unhandled rejection from a spawned
future), tells running out of memory from an internal error by the panic message a chained panic
hook kept, and skips the Leave site prompt once stopped. ?free-mb=N makes the memory gates and the
low-memory warning see only N MiB, for checking them in a browser.
* Docs: web build memory handling in the gaps list and the parity assessment
* Engine: use is_multiple_of in the megabyte formatter
* Cargo.lock: the web shell depends on photocraft-raster
* Web boot page: carry the stopped message too, as the test that keeps the two pages equal requires
* Save check counts one copy of the document, and runs once per save
The encoded file's buffer is reserved fallibly, so a save it doesn't fit
in fails with an error instead of trapping; counting it as well refused
saves that would have fitted. On the background-save path the check ran
in write_document and again in start_save, which other callers reach
directly; it now runs only in start_save there.
* Review fixes: warning check allocates nothing, PSD merged image counted, automation saves checked
The low-memory warning's periodic check reads the size of linear memory
instead of reserving a block: wasm memory never shrinks, so a trial every
few seconds raised the high-water mark by up to 768 MiB for nothing. The
operation checks keep their trial reservation.
PSD and PSB saves always build a canvas-sized merged image (a placeholder
without maximum compatibility) and hold it raw and encoded; the save
estimate now counts it. The synchronous automation save checks memory
before encoding, as File > Save does, and its test asserts the memory
error.
The trap listeners go in before the app is built, so a trap while
building it shows the stopped message too, and a memory access out of
bounds reads as unexpected rather than as running out of memory.
* Cover layered PSD preflight limitation
* Fix layered PSD test iteration
* Docs: layered PSD preflight limitation in the gaps list
* Web OOM gate: gate direct exports, truthful advice for unbounded PSD sources (#10024)
* Web OOM gate: estimate layered PSDs per layer instead of refusing every real PSD
The PSD check admitted a file only when its color-mode data, image
resources and layer section were all empty. Every Photoshop file has
image resources (at least the resolution), so on the web nearly every
real PSD was refused for Place, PSD smart objects could not render,
and Image Size was refused for documents holding one, whatever the
free memory (u64::MAX never fits).
The color-mode data and image resources hold no pixels, so they are
now only skipped. A layered file reads its layer count from the layer
info (signed, absolute value) and adds a canvas-sized surface and mask
per layer to the flat estimate. Files that are not 8-bit RGB or
grayscale, keep their layers in a Layr tagged block, or whose sections
can't be read stay unbounded and refused, as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Web stopped page: match trap messages ignoring case
Safari reports the allocation-failure abort as "Unreachable code
should not be executed" and Firefox says "out of memory", which the
case-sensitive matching missed: the trap was not recognised, or was
shown as an unexpected stop instead of running out of memory.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Export estimate: count the merged image OpenRaster writes
The ORA writer keeps the layers but also flattens the document for
mergedimage.png, a 16 B/px composite of the canvas, which the
estimate left out.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: FredApps <153173854+FredApps@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
On a 32-bit build, free RAM or commit headroom can set the limit instead of the
cap, and then 'install the 64-bit build' was wrong. The message now mentions the
cap and says the 64-bit build may open the file.
The verbatim prefix is now removed through path components, so a scratch path with
a non-UTF-8 name still finds its disk. The UNC rewrite is gone: sysinfo lists only
fixed and removable volumes, so a network path can never match. That limitation is
now documented.