#258 left the Tint row at 🟡 and docs/tint-direction.md as a review note.
The sign now matches the slider's track, the DNG specification and
Lightroom XMP, so the row is back to ✅; the doc says what happens to
saved custom tints (they shift the other way; nothing is rewritten) and
What's new tells users.
Co-Authored-By: Hanmo <letsgo@88.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01912WaK6Rmew3wxZuPyX7E8
A texture's whole-pixel size is slightly off the frame's exact aspect (a
128 px thumbnail by up to ~1 %), so following it always made the loupe
rect jump a little as the thumbnail, preview and full render arrived, and
moved mask and crop overlays with it. Follow it only when it differs by
more than 2 % (an unsupported raw's embedded JPEG with another crop), and
never in the crop tool, whose overlays need the frame's own geometry.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01912WaK6Rmew3wxZuPyX7E8
#287 made app.export refuse unknown params. Stored settings (a user export
preset, prefs.json lastExport) written by another version are not the
caller's typo, and refusing them would break that preset or Export with
Previous for good after an update: their unknown keys (also inside
watermark) are dropped before the strict check. What the caller passes is
still checked strictly.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01912WaK6Rmew3wxZuPyX7E8
Keep embedded and decoded preview proportions consistent in the justified
grid, photo views, alternate detail previews, and culling previews.
Co-Authored-By: GPT-6 Luna <noreply@openai.com>
* Release: FreeBSD x86_64, Flatpak x86_64/aarch64, AppImage zsync updates; README Downloads
- linux: appimagetool -u embeds gh-releases-zsync update information; the .AppImage.zsync is
published beside each AppImage (zsync installed; checked in the job).
- flatpak: new job per arch repackages the Linux tarball into a single-file .flatpak
(packaging/linux/flatpak-bundle.sh + ai.storyteller.lightcraft.bundle.yml); the Flathub
manifest gets the CLI and the same commented finish-args; packaging-lint checks they agree.
- freebsd: new job builds lightcraft-<ver>-freebsd-x86_64.tar.gz in a FreeBSD 14.3 VM with the
freebsd.yml packages (packaging/freebsd/package.sh; layout dry run in packaging-lint).
- version, linux and web no longer use the release environment (they sign nothing), so a
branch dispatch dry-runs every unsigned job; the release job still needs macOS and Windows.
- README: Downloads section listing every build; docs/releasing.md updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Linux packages: describe LightCraft, not PhotoCraft, in the deb/rpm description
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Brandon Thomas <shared2@storyteller.ai>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Bilevel scans, CCITT fax images and 2- or 4-bit gray TIFFs were rejected with
"sample buffer too short". The `tiff` crate returns such samples packed, most
significant bits first, eight, four or two to a byte with each row padded to a
whole byte; the converter expects one byte per sample. They are now spread out
one per byte and scaled to 0–255 (255, 85 and 17 times the sample: exact) before
conversion. 3-, 5-, 6- and 7-bit samples, and planar images with more than one
sample narrower than a byte, are refused with an "unsupported" error.
The new test builds 1-bit (both photometric conventions, a width that leaves row
padding, set padding bits), 2-bit and 4-bit strips in code, plus a truncated
strip that must fail cleanly; it fails without the fix with the old error. A
600×400 bilevel image, uncompressed and CCITT G4, now decodes to the same pixels
PIL reads from it.
The module documentation no longer claims palette support: the `tiff` crate
refuses RGBPalette images on every read path, so the palette branch here has
never been reached. The parity row lists palette TIFF as missing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A WhiteIsZero (PhotometricInterpretation 0) gray TIFF opened as a negative at
every bit depth. The `tiff` crate (0.11) already inverts those samples as it
reads them, for every layout it accepts, and refuses the WhiteIsZero layouts it
can't invert; the decoder here then inverted them a second time. WhiteIsZero is
common in scanner and fax output, and CCITT fax images use it by definition.
The second inversion is removed. The new test builds 8- and 16-bit WhiteIsZero
and BlackIsZero strips in code and fails without the fix (the lowest sample
came out black instead of white).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The filmstrip painted selection chrome only for the active photo, so after
Select All or a ⌘/Shift-click multi-selection the other selected photos
looked unselected, although the selection itself was right.
Paint it as the grid does: every selected photo gets the selected cell
fill; the active one keeps its white frame and the others get the grid's
grey one.
Test: tests_scroll::filmstrip_shows_every_selected_photo samples a cell's
background in the headless render after Select All, after selecting one
photo, and after a toggle (what a ⌘-click runs); before the fix a selected
photo that was not the active one was painted like an unselected one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The detail view's filmstrip shortened a long file name with `&name[..13]`,
a byte slice. For a name with multi-byte characters byte 13 can fall
inside a character (`202407層三限定訂閱圖(4)`: inside '限'), and the
slice panicked the UI thread as soon as that photo scrolled into the
strip, so the app quit.
`film_label` now counts characters: more than 14 are cut to 13 and an
ellipsis, as before for ASCII names.
Tests: a unit test of `film_label` (CJK, exactly 14 characters, ASCII,
empty) and a headless filmstrip test with the name from the issue; both
panicked with the reported message before the fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`--help`, the module doc and the README promise that a `--memory` session
"writes nothing", and the desktop app broke that in two places:
- `PrefsWriter` saved `ui.json` whenever the UI state changed (issue #164),
so a throwaway session's `thumbSize`, window layout or language replaced
the user's saved settings — the README showcase script does exactly that.
- The GPU crash sentinel from #136 armed itself in every session except
`LIGHTCRAFT_NO_PREFS` (issue #169); `with_init_marker` creates the
settings folder to hold `gpu-init.marker`, removes the marker and leaves
the folder, so a `--memory` run on a fresh home left an empty
`Library/Application Support/LightCraft` behind.
Now `PrefsWriter` carries its target path (`None` for `--memory`,
`LIGHTCRAFT_NO_PREFS` or no home folder) and writes nothing without one; it
also writes nothing while the session is the one offered when the library
can't be opened (the question window and Continue Without Saving, which
`docs/control-protocol.md` already describes as writing nothing) and
resumes when `app.openLibrary` ends it. `gpu_crash_check(in_memory)` neither
arms the sentinel nor removes a marker in a `--memory` session; it still
reads one (new `read_init_marker`), so GPU rendering starts off there too
and the next ordinary launch reports and clears it. A `--memory` session
still reads `ui.json`, but a damaged one is reported and left in place
rather than set aside. Normal launches are unchanged.
Tests: `prefs_writer_writes_nothing_in_a_memory_or_temporary_session`,
`memory_session_leaves_the_gpu_marker_and_folder_alone`,
`memory_session_leaves_a_damaged_ui_json_alone` (apps/lightcraft), and
`read_init_marker` coverage in the gpu crate's marker test.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
`ExportOptions::from_json` reads what it recognises and defaults the rest, so
`longEdgee=400` exported at 3000 px, `longEdge=banana` at full size, and a
`watermark` of `5` or `{"text":"x","bogus":1}` passed with `ok:true`, while
`size: 3` was silently clamped to 0.5 of the short edge (#181, #183).
`from_json` stays lenient for stored prefs and dialog prefill. New
`ExportOptions::validate(cmd, params)` / `from_params(params)` check every key
against the option and target lists (unknown ones name the nearest known key),
every value's kind and range, and a `watermark` object's keys, kinds and ranges
(`size` 0.005..0.5 and `inset` 0..0.4 of the short edge, `imageWidth` 0.01..1 of
the width, `opacity` 0..1, `color` [r,g,b]), returning
`invalid parameters for \`app.export\`: …` like the other commands. The three
places that run an export use it — the desktop `app.export`, the headless
backend behind `lightcraft-cli run` and the MCP `export` tool, and
`lightcraft-cli render --opt` — and `export.savePreset` validates a preset's
params when it is saved. Built-in presets, the Export dialog's `to_json`,
Export with Previous and `render --opt` all pass unchanged.
The `app.export` param string (`lightcraft-cli commands`, `engine.commands`),
the MCP `export` tool schema and docs/parity.md LR-EXP-WATERMARK now list the
watermark keys, their units and ranges.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
`tol = 0.04 + 0.16·refine` fully selected anything within 0.06 OkLab of a
sample at the default Refine 50. Through `tonemap_for_select` the report's
teal mid-grey and neutral light grey are only ~0.063 apart (L 0.60 vs 0.72,
a/b ≤ 0.02), so the second patch was 99 % selected and the mask covered the
whole photo; Refine 85 and 100 kept most of it too. Now
`0.015 + 0.06·refine` (full ≤ 0.0225 / gone ≥ 0.045 at Refine 50, gone ≥
0.066 at 85, 0.075 at 100), computed once in
`lightcraft_pipeline::masks::color_range_tolerance` and used by both the CPU
evaluator and the GPU `mask.wgsl` kind-3 parameter so the two keep agreeing.
Test: two flat patches at the report's sRGB values, sampled as
`mask.sampleColor` does; the sampled patch stays fully selected and the other
one is out at Refine 0, 50 and 85 and mostly out at 100.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
`fn snapshot` wrote every reply but never looked at `ok`, so a script with a
bad `ui.set`, an unknown command or a failed `ui.screenshot` exited 0 while
the PNG showed the wrong view or was never written. The run now counts
replies with `"ok": false`, keeps going (every reply still reaches stdout and
the final `-o` screenshot is still written — continue-and-report, like
`run --keep-going`) and returns `N scripted request(s) failed` afterwards,
which `main` turns into exit 1. Usage text and docs/control-protocol.md
"Headless rendering" state the contract.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Split out of #240 without the compressed decoder (crates/raw/src/vendor/orfc.rs
is not part of this change; compressed ORF stays preview-only).
- The layout is read from Exif CFAPattern (0xa302, repeat counts in either
byte order) when it describes a 2x2 Bayer cell. BGGR and GBRG were not
reachable before.
- Without a usable tag the green diagonal is still found from the samples,
now from totals of 32x32 blocks instead of single-pixel differences, so
fine texture no longer outweighs a small red/blue difference. The old
version indexed the samples directly; this one reads them with get().
- probe_info no longer unpacks 12-bit packed ORFs that state their layout
(XZ-2: under 0.1 ms; it ran the same unpack as a decode before, which
takes 11 ms). 16-bit files still read their samples to find the bit depth.
Tests: stated layouts in both byte orders, unusable tags falling back to the
samples, the texture case, hostile areas for the fallback, probe_info equal
to decode; a corpus test comparing the E-1, E-400 and XZ-2 mosaics with
their embedded JPEGs under every layout.
No user-visible change yet for the bodies whose layout the old guess got
wrong (E-M1 and E-620 are BGGR): they write compressed ORF.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`select_photos {"ids": [9999]}` answered `selected: 1` and left the session
with no usable active photo; `develop.get {"id": 9999}` answered too.
`library.select` now checks every id (and `active`) against the catalog
before touching the selection and returns `no such photo N`, so the
previous selection and active photo stay in place; `photo_arg` (develop.get,
photo.inspect, photo.allMetadata, history.list) does the same for an
explicit `id`. Engine and MCP tests cover every select mode.
Fixes#182
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
`moxcms` stamps the ICC header's creation date/time with the wall clock on
every `encode()` (the writer uses `ColorDateTime::now()` and ignores the
profile's `creation_date_time` field), so two exports of one photo with
the same settings differed in the embedded profile and nowhere else — the
pixels matched, the hashes never did. `write_matrix_trc`, which every
embedded profile goes through, now zeros the header's date bytes (24..36)
after encoding, as other deterministic profile writers do. A test asserts
two writes of every named space are identical, the date bytes are zero and
the profile ID is empty.
Fixes#180
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The browser accepted and persisted library.preferences cacheMb, but stored thumbnails stayed over that limit: the worker index pruned with the native default (DISK_CACHE_BYTES), so the setting only reached the desktop cache. The worker group now carries the session cache_bytes(), the app refreshes it every frame, and the next stored thumbnail prunes to it.
On macOS every shortcut that appears in the menu bar was left to the native
menu, and the egui shortcut handler skipped those keys. AppKit only offers a
key press to the menu bar when it carries ⌘ or ⌃, or is a function key; a
plain key (E, C, H, M, I, K, D, 0–9, P, U, Y…) or one with only ⇧ / ⌥ (⇧P,
⇧V, ⌥Y…) goes to the window's winit view instead. So all of those did
nothing on macOS, while keys outside the menus (Space, G, X, ⌫) and ⌘
shortcuts worked.
`publish_shortcuts` now lists only the shortcuts the menu bar really receives
(`menu_delivers`: ⌘, ⌃ or a function key) in `native_shortcuts`. The others
keep their accelerator on the menu item, so the menu still shows the key, and
egui runs them.
Checked in a release build by pressing keys (not via the control channel,
which bypasses the native menu): C opens Crop & Rotate, E Edit, ⇧P Presets,
3 / 0 set and clear the rating; ⌘⇧H toggles the histogram once per press;
typing "c" in the search field types and leaves Crop closed. The same build
without this change ignored C, ⇧P and 3. ⌃ shortcuts (merges) are unchanged
and were not tried.
Regression test: native_menu::tests::single_key_shortcuts_stay_with_egui
(fails with the previous behaviour: "E must be left to egui").
The decoder change alters format-8 renders while the photo's content hash
and settings stay the same, so thumbnails rendered before it (edited
photos keep theirs on disk) would survive. Bump RENDER_CACHE_VERSION.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Custom macOS larger runners in a runner group do not pick up jobs; the
built-in M2 Pro label does. Only release.yml uses it, and the
actions_macos_xl budget caps the monthly spend.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Raw format 8 codes a cell top-left, bottom-left, top-right, bottom-right.
decode_strip placed the values in row order, which put the two green
samples of every RGGB cell on each other's sites: flat areas were right,
but every edge got a one-pixel colour zipper. Measured on a G9 II and the
corpus GH6, S5 II and S9: the Gb plane sat at (+0.5, -0.5) half-res cells
from Gr instead of the Bayer (-0.5, +0.5); with this change all four match.
The test encoder used the same order, so the round trip passed either way;
a new known-answer test hand-codes one cell and checks where it lands.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The standard macOS runners are shared with all CI and queue for an hour or
more. The macos-release runner (GitHub-hosted M2 Pro) is in a runner group
usable only by release.yml on the release branch, with a monthly spend cap
on the actions_macos_xl SKU.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* About: contributors compiled in from contributors.json (Contributors + Models tabs, grab bag + table, sortable, name toggle)
Adds the craftrules standard (standards/contributors.md) to LightCraft: contributors/contributors.json
is generated by craftrules/scripts/contributors.py and baked into the binary by
crates/ui-egui/build.rs (merged next to the tr_format! generation) as static tables. The About
dialog gets tabs About · Contributors · Models. Each contributor line: @username, opt-in
display/real name, PRs, commits, lines +/−/Δ, binary assets +/−, first/last commit. Sortable by
every column; alphabetical ignores case and the leading @. Tabs and the list view are
control-channel widgets (button:aboutTab-*, button:creditsTable / button:creditsGrabBag).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Contributors docs: craftrules is two levels up (../../craftrules)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Contributors: self-submitted names in contributors/people.toml; AGENTS.md credit-yourself section
build.rs reads contributors/people.toml next to the generated contributors.json and shows a name only
for entries the generator verified as written by that person (craftrules standards/contributors.md).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Contributors: regenerate (verified people.toml entries)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Contributors: regenerate after rebase onto main
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Contributors: regenerate (verified people.toml entries)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
