The thumbnail cache wrote .jpg files into the library's thumbs folder, which is usually inside
Pictures: photo galleries listed every thumbnail as a photo (issues #610, #1066), and every read
brought the file's date forward, so each launch made them all look new.
- Cache files are named <32 hex>.lcthumb (JPEG data as before). A .jpg of before is taken over
under the new name when it is asked for; the rest when the cache is tidied, which the engine does
once per cache folder, on a thread of its own with a row in the activity stack that shows its
progress and can stop it. Nothing is rendered again.
- Reading writes nothing: what was read is remembered for the session (a full cache lets it go
last), and only a file that went unused for a week has its date brought forward.
- A folder that holds nothing but the cache gets a CACHEDIR.TAG, when the cache makes it and when
it is tidied; one that also holds somebody's files is not tagged (issue #98).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The left panel shows a count beside every album. Each catalog revision had
every album counted again, and a smart album's count asks every photo of
the library whether its rules hold: one pass over the library for each smart
album after each keyword, rating or slider tick (about 25 ms an album in a
library of 67,000 photos).
`AlbumTally` (`crates/catalog/src/tally.rs`) keeps the counts and, for each
smart album, which photos it holds. After a small edit each smart album asks
only the photos the edit touched (`Catalog::touched_since`); an album of
chosen photos is counted again only when a photo came, went or was deleted.
Counted from every photo as before: new rules, a change to an album, a stack
or the keyword list, an edit to more than 256 photos, another library, and
each new minute while an album asks about "the last N days".
What an album remembers is in proportion to the photos it holds: a bit a
photo id while ids are close together, a set when they are far apart.
The frame after a keyword in that library, both side panels open: about
54 ms -> 27 ms (156 ms before #1660).
Tests: 900 random edits against eight smart albums (keywords, flags,
ratings, labels, sliders, capture dates, file names, an album of chosen
photos) and an ordinary one, the followed counts compared with
`album_count` after each; a delete followed, three hundred photos counted
whole; the members' bits at the edges and for ids far apart.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Everything derived from the catalog was keyed on its revision, which every
op raises. So each edit (a keyword, a rating, every tick of a slider) had
the grid's list queried and sorted again from every photo, the source
counted by a second full query, and the side panels tally the whole library
again. In a library of 67,000 photos the frame after one keyword took about
156 ms with both side panels open (an idle frame: 2 ms).
The catalog now remembers which photos its latest ops touched and what of
them they changed (`Catalog::touched_since`, `aspects_since`, `Aspects`). It
answers "not known" for an op that changes more than photos of its own (an
album, a stack, the keyword list), for changes too long ago, and for a
revision it did not make itself (a library just loaded).
- The view (`Session::visible`): when the same view is asked again after a
small edit, the photos the edit touched are taken out of the list it had,
asked about again, and put back at their place in the order
(`Catalog::follow`). 37 ms -> 0.6 ms. Worked out whole as before: more
than 256 photos touched, a change that is not photos' own, an album in its
own order, Recently Added, Missing Photos, and a view that asks about "the
last N days" (it changes with the clock too).
- The source's count: taken from the list when no filter is set, else
counted without sorting (`Catalog::count`). 37 ms -> 0.
- The side panels (`Caches`): the folder tree, keyword tree, date groups,
people's names and library counts are kept through an edit that changed
nothing they read. A rating or a slider changes none; a keyword leaves the
folders, dates and counts.
The frame after a keyword: about 156 ms -> 54 ms with both side panels
open, 71 ms -> 2 ms with the grid alone (`tests_view_cost.rs`, a synthetic
library, run with --ignored --nocapture).
Still paid on each edit: the count beside every album and smart album in the
left panel (about 25 ms for a smart album there), and the filter bar's lists.
Tests: 1,440 random edits of every photo-local kind across sources, sorts,
filters and the hold, each compared with the list worked out whole; 600
random edits comparing every kept tally with a fresh one; the clock; the
record itself, with its failure paths. `ui.inspect` -> `perf` reports
`viewFollowed` and `viewWorkedOut`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The scan asked the disk about every library photo in the folder: is its
file there, and has it a sidecar under any of four names. On a network
share each question is a round trip, and the scan had just listed the
folder. Measured on a hard disk over SMB, for unchanged folders of 1,600
to 2,400 photos: 4.6 to 7.1 s of questions after a listing of 4 to 5 s.
- The walk hands each folder's listing over (`walk::files_in_noting`:
every entry's name and kind, hidden ones too, and whether that is all of
them). The scan walks the folder once, then checks the photos beside
the probing of new files.
- `sync::Listing` answers "is it there" / "is it a file" where that holds
on any file system: a name the folder lists is there; a plain ASCII name
is not there when the folder was listed to its end, all its names are
ASCII and none is that name in another case. Everything else is asked
of the disk as before (a link, a name Windows reads another way, a
folder that was not walked, and every name not listed in a folder that
holds a name outside ASCII).
Now under 0.05 s for those folders. Two things differ on purpose: a file
the folder lists is there though it can't be opened by its path (a folder
that may be listed but not searched, an over-long path), where such
photos were listed as missing; and the answers are the folder as the
walk saw it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A view is asked again after every edit, so a photo that an edit makes stop
matching it leaves the grid at once and every photo after it moves up: a
keyword given in a smart album of photos without keywords takes the photo
away from under the cursor, and in a random order, with no landmarks, the
shift reads as a new shuffle (the seed does not change).
View > Hold View Until Refreshed (`library.holdView`, off by default, kept
with the library's preferences): a photo the view shows stays in it when an
edit makes it stop matching a smart album's rules, Picks or the filter bar.
View > Refresh View (`library.refresh`, also in the sort menu while the hold
is on), Reshuffle, another source, filter or sort, or new rules for the
smart album shown, shows what matches now.
- Photos that start matching still appear at once.
- A photo that is deleted, or taken out of the album or folder shown, still
leaves at once: a held photo has to be where the view looks
(`Catalog::query_keeping`, `within`).
- A source with no filter over it holds nothing and is queried as before.
- A refresh keeps the seed.
- The grid's count no longer reads "3 of 2" when a held view shows more than
now match its source.
Each edit still asks the whole library again; that cost is not addressed
here.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Photo ▸ Photo Merge ▸ HDR for Each Bracket Set (`merge.hdrBatch`) merges many brackets in one go with the last HDR
options, as Lightroom merges a selection of stacks: one HDR per stack among the photos, then per bracket among the
unstacked ones (`Catalog::bracket_sets`: photos taken less than about a second after the previous one's shutter
closed, at exposures at least ½ EV apart; bursts at one exposure and earlier merge results are left out). The app runs
the merges one after the other in the activity stack (cancelling one stops the rest; a failed set doesn't); the
command runs them inline for the CLI and MCP (`lightcraft-cli merge hdr-batch FILES…`) and reports merged and failed
sets.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A 7-frame bracket of 67 MP raws ran out of memory: the merge decoded the frames four at a time, and the AHD demosaic
collected every tile before copying them into a second full-size picture. Full-size frames now decode one after the
other (the demosaic is parallel inside; previews stay parallel), and AHD fills the picture by rows of tiles. Peak memory
of 67 MP HDR merges: 3 frames 4.8 → 3.8 GB, 4 frames 6.3 → 4.7 GB, about the same time; AHD output is pixel-identical.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`ExportOptions::validate` accepted `metadata`, `sharpen`, `sharpenAmount` and
`conflict` case-insensitively, but `from_json` read them with serde's
exact-case names and silently fell back to the default when that failed.
`{"format": "jpeg", "metadata": "NONE"}` therefore passed validation and
exported with `MetadataPolicy::All`: GPS, location and camera data went out
in a file the caller asked to carry none. MCP, the control channel and the
CLI all go through `from_params`, so all were affected.
Each of the four enums gets a case-insensitive `parse` (the pattern
`HdrCurve::parse` and `JxlEffort::parse` already use), and both `validate`
and `from_json` call it, so validation and conversion can no longer
disagree. Accepted inputs, error messages, defaults and `to_json` output are
unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Activations stay on the device between layers (net/program.rs lowers the graph; every
operator is a WGSL kernel, one submission and read-back per run); faces of a photo go
in batches of up to 16; Settings > Faces shows each model's device and a CPU-only
switch (faces.gpu); the scan's pace limits GPU use; the GPU check is an activity row.
YuNet 58 -> 21 ms, SFace 40 -> 9.6 ms, AuraFace 295 -> 71 ms (heavily loaded machine).
Sources: own design.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
On a case-sensitive file system, `DSC_1.nef.xmp` or `DSC_1.Xmp` beside `DSC_1.NEF` was never
found, so the photo came in with only the file's own metadata (its IPTC and keywords) and lost
the sidecar's rating and crop. The lookup now tries the photo extension in lower and upper case
and `.xmp`/`.XMP`/`.Xmp`; Move takes such a sidecar along.
Sources: own design.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A raw imported while its format couldn't be decoded kept preview_only until a
manual Reload, so new decoders in a release never reached existing photos.
When a library opens in another app version, its preview-only photos whose
stamp differs are probed again in the background and get exactly what Reload
from Disk gives them (shared reload_ops): a file that decodes now loses the
flag, one that still doesn't gets the current reason and a fresh stamp. The
result is one journaled system batch, not an undo step; develop settings and
history are kept. Unreadable files are left for next time, photos changed
while probing are skipped, and a result for a library closed meanwhile is
dropped. A toast says how many raws now develop from their raw data.
- engine: cmd/recheck.rs (stale / probe / apply) and
library.recheckPreviewOnly {all?} -> {checked, updated, cleared}; refused
without file access, where every probe would clear every flag
- desktop: recheck::tick once per library generation, through tasks::spawn
("Checking raw files" in the activity stack)
- the preview-only notice (Edit, Info) gains a Reload from Disk button
- strings translated in all nine catalogs; docs/mcp.md and the parity row
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Catalog format v9 adds Photo.preview_checked and Op::SetContent.preview_checked:
the app version (CARGO_PKG_VERSION) whose decoder set preview_only, empty when
it isn't set. Import and content_op (Reload, Relink) stamp it; undo swaps it
back. Catalogs from before v9 read an empty stamp, so their preview-only photos
count as stale (Photo::preview_only_stale) and will be rechecked once.
The version guard test now also accepts a format version that only adds fields.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sony ARWs record shading, lateral chromatic aberration and distortion as a camera setting plus a radial
table in the raw image IFD (0x7031-0x7037), with copies in the encrypted SR2SubIFD (0x797c-0x7982).
They are now decoded for every body that records them and carried through RawInfo and RawImage as
`lens_metadata.sony` (types in crates/raw/src/lens_metadata.rs, decoder in
crates/raw/src/vendor/arw_lens.rs):
- Raw-IFD fields are the ones the TIFF reader parsed, which is what the image itself uses. A scan of
the directory adds diagnostics only: repeated tags, and why an entry the reader skipped was dropped
(values outside the file, an unknown type, past the reader's limits). SR2SubIFD fields are read
from their first entry. Only the exact layouts are accepted: SHORT[1] or non-negative SSHORT[1]
settings, SSHORT[17] headed by 16 for shading and distortion, SSHORT[33] headed by 32 for CA.
Nothing is allocated from a count in the file.
- The raw IFD's copy wins. The SR2SubIFD copy fills in only for a tag the raw IFD lacks, never for a
malformed one, and the two copies' agreement is recorded.
- Settings keep their number and their documented meaning, Unknown(value) otherwise (never Off).
- Framing evidence (stored frame, Sony and default crops, reference rectangle, coding, 3:2) and a
per-component status. Only Applied components become opcodes.
Applied corrections are unchanged: the ILCE-7RM4A distortion conversion receives exactly the table it
received before, and no other body or component gains a correction. The ILCE-7CR distortion misses its
camera JPEG at 24 mm, and shading and CA semantics are not validated on any body; see
docs/sony-lens-corrections.md.
Locating the SR2SubIFD, now done for every ARW that has one, selects the SR2Private offset, length and
key with the TIFF reader's own rules, which lightcraft-tiff now exposes and builds its reader on
(directory_entries, entry_span, ParseOptions::budget_for, decode_value). Only each field's first value
is decoded, so the directory's other entries cost no memory sized by the file: a 16 MiB file whose
SR2Private holds eight fields aliasing it peaks at 37 kB. DirectoryEntries keeps its layout private,
and its positions() iterator uses checked arithmetic.
Tests: synthetic decoder tests; a differential test of the SR2Private selection against the TIFF
reader's in both byte orders (0 to 8,193 filler entries, unreadable and unusual first occurrences of
each field, a used-up value budget); a dhat peak-heap test; public-corpus checks against the values
ExifTool prints, with pinned and anchored SHA-256s (LIGHTCRAFT_CORPUS_REQUIRED=1 makes absent files
fail); and an opt-in local inventory check (crates/raw/tests/sony_lens_inventory.rs).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An accelerator hook in the face interpreter takes Conv nodes; lightcraft-gpu implements it with a WGSL
implicit-GEMM convolution (any stride, padding, grouping; split-K for small maps; a direct kernel for depthwise
layers), each model's weights resident on a device of its own. After loading, detection and recognition check
the GPU in the background (agreement within 1e-3 and faster than the CPU) and otherwise stay on the CPU, which
remains the reference; the first GPU error drops it. Sources: own design.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Docs conflicts resolved on upstream's text; MRW edits reapplied with
coverage re-measured by docs/raw/coverage.py (243 verified, 946
unverified, 221 preview). Drops stale duplicate rows upstream's docs
carried in raw-parity.md and gaps.md.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Docs recounted with docs/raw/coverage.py: 241 verified, 35 unsupported. The MCP
headless extension list now comes from engine::import::EXTENSIONS upstream, which
already carries mef.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Keep main's spots.rs implementation and preserve the catalog reader and
updated capability documentation. Readers are byte-identical to the PR.
Validation: 48 catalog and 9 spot tests pass on candidate and main.
Formatting, Clippy, parity, layers, assets and WASM checks pass.
Full CI remains blocked by the same mask-overlay test failure reproduced
on main; no assertions or tests were changed to hide it.
Refs #798.
