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 <[email protected]>
eframe 0.36's App::on_exit takes the glow context as a second argument.
The macOS-only menu-bar Quit path still called on_exit(self), so the
0.7.0 macOS release job failed with E0061. Pass None; our on_exit ignores
it. macOS isn't built in CI, so only the release build caught it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
libtest has no skipped status and hides the output of passing tests, so a
test that returns early without a GPU, a corpus file, a craft-fonts face or
a model passed silently. `lightcraft-testkit` (the testkit slot the layering
table already reserves, dev-dependency only, std only) adds
`skip!(Need, "reason")`: it prints the note as before and, when
LIGHTCRAFT_SKIP_LOG names a file, appends one JSON line (need, reason, test,
binary, file:line). Plain `cargo test` is unchanged. The 61 existing skip
sites use it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
New crate lightcraft-geocode (L1): parses the GeoNames cities dump and finds the
nearest place within 50 km through a 1-degree cell index; NaN / out-of-range
positions and malformed lines find nothing. Engine commands
metadata.reverseGeocode (Photo menu, Info panel "Fill Location"; fills only
empty fields unless overwrite, one undo step, activity row for batches),
metadata.downloadPlaceNames (background download of the CC-BY 4.0 GeoNames
files into the settings folder, cancellable activity row) and
metadata.placeNamesStatus. No data is bundled; NOTICE records the licence.
Fixes#582.
Sources: GeoNames dump readme (prose); haversine (textbook); PKWARE APPNOTE
(ZIP, prose); cell index own design.
Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
New `inpaint` crate (L1): multi-scale patch-based completion with a PatchMatch
nearest-neighbour field, seeded by a membrane fill. Deterministic, no panics on
hostile sizes, cancellable. Ported from PhotoCraft (MIT OR Apache-2.0).
Remove tool: a fourth spot mode, `fill` ("Content-Aware"), rebuilds the spot
from patches of its surroundings instead of one source; coarse-to-fine, so the
structure is chosen at a scale set by the spot size and preview and export agree
(tested). spot.add leaves its source unset.
Sources: C. Barnes et al., PatchMatch (SIGGRAPH 2009); Y. Wexler, E. Shechtman,
M. Irani, Space-Time Completion of Video (TPAMI 2007); P. Perez et al., Poisson
Image Editing (SIGGRAPH 2003) for the membrane fill; code ported from PhotoCraft
crates/algo/src/inpaint.rs + poisson.rs (a680ed22).
Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
Profiles are interpolated in focal length (vignetting also in aperture and
focus distance), scaled by crop factor, and converted by least squares to the
DNG WarpRectilinear / FixVignetteRadial polynomials the pipeline already
applies. The lensfun database (CC-BY-SA) is only read from a directory the
user supplies; nothing is bundled.
Sources: lensfun's published database format and lens calibration
documentation (no lensfun code read), PanoTools wiki (a, b, c distortion
model), DNG specification 1.6 (WarpRectilinear, FixVignetteRadial);
interpolation and fitting: own design.
Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
ROADMAP.md: take main's restructured version; JPEG XL export is now
recorded in docs/file-format-parity.md and docs/gaps.md (gap 25).
Co-Authored-By: Claude <[email protected]>
- release.yml: new linux-riscv64 job (ported from GridCraft): cross-compiles on
ubuntu-24.04 against ports.ubuntu.com riscv64 libs, packages a tar.gz with
package.sh --formats tar, smoke-tests lightcraft-cli under QEMU. The release
job needs it, so its tarball lands in the draft and SHA256SUMS.
- packaging/linux/package.sh: CROSS_ARCH/CROSS_TARGET/CROSS_COMPILE/EMULATOR
support (target dir, cross strip, emulated --version).
- README Downloads and docs/releasing.md list the riscv64 tarball.
Co-authored-by: Claude Opus 5.5 <[email protected]>
- New crate lightcraft-slideshow (L1, no UI): SlideSettings (margins linked or
independent, guides, aspect preview, zoom to fill, stroke border, cast shadow,
backdrop colour, colour wash, background image; lengths relative to the slide's
shorter side; validated and clamped) and a CPU compose() used by playback, the
Settings preview and slideshow.preview.
- Engine: slideshow.layout (partial merge, saved in prefs.json) and
slideshow.preview {id?, width?, height?, path?} (PNG to path or base64);
RenderJob::with_slide composes on the render worker; the background photo is
rendered once and shared.
- Playback: with a non-default layout each slide is the composed screen-sized
slide, prepared ahead like before; the default layout plays as before.
- Settings > Slideshow: Layout, Options and Backdrop sections and a debounced
live preview of the active photo's slide (with margin guides).
- Tests: composer geometry/pixels/hostile input, command get/set/persist/preview,
headless playback of a composed slide and of the default layout.
Previews looked more saturated than the exported files on Mac screens. The
pixels are the same: preview and export match to within one level on the CPU
and GPU paths, and exports embed a correct sRGB profile. But wgpu leaves the
window's Metal layer without a colour space for sRGB surfaces, and Core
Animation shows such a layer without colour matching ("If nil, no
colormatching occurs", CAMetalLayer.h). On a P3 screen the sRGB previews
were therefore shown too saturated, while Preview and Photos convert the
exports properly.
While no monitor profile is chosen in Settings > Display, the desktop app now
shows previews through the profile of the screen the window is on
(NSWindow.screen.colorSpace.ICCProfileData, safe objc2 calls), through the
existing display-profile path. It is read again about once a second, so
moving the window to another screen is followed. A chosen profile still wins;
"Use Screen Profile" goes back to it. app.displayProfile reports it as
path: null, screen: true. Other platforms and headless sessions are
unchanged.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Export as JPEG XL from the dialog, CLI, MCP and control channel (format jxl): lossy with
quality or lossless, 8/16-bit, effort fast/normal, progressive, EXIF/XMP boxes; HDR edits
as 16-bit Rec. 2020 PQ. Encoder: jixel (pure Rust, BSD-3-Clause OR Apache-2.0, native
only: it spawns its own threads), wrapped in catch_unwind. Lossy files declare an enum
encoding (sRGB / P3 / Rec. 2020; Adobe RGB and ProPhoto written as Rec. 2020); lossless
files embed the ICC profile.
Decoder fix: jxl-oxide desaturated colours outside sRGB when rendering XYB images to
wider primaries; lossy wide-gamut JXL now renders to linear sRGB primaries and is
converted to Rec. 2020 here. jixel's slowest effort is left out: jxl-oxide misdecodes
its (valid) files.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
`encode_avif_pq`: linear Rec. 2020 with SDR white at 1.0 → PQ (SDR white at
203 cd/m², BT.2408), BT.2020 NCL YCbCr 4:4:4, 10-bit, with CICP nclx
(9/16/9, full range) and a clli box. ravif fixes the sRGB curve, so this
drives the rav1e + avif-serialize it already uses (no new crates).
Export: AVIF with `hdr` renders the HDR edit in Rec. 2020 and writes PQ.
Photos without an HDR edit export as before, in every format. CoreImage
decodes the file as HDR (same 3.1 % of the frame above SDR white as the gain
map JPEG of the same edit).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
One config-dir helper (engine::config::config_dir; camera_profiles re-exports it) with both
the face-models and denoise-models defaults; hosts set up both model folders; Settings has
Faces and AI Denoise tabs; both command sets registered; both model dialogs; locale entries
from both; one [features] table per app manifest; Cargo.lock regenerated by cargo; the
DecodedSource cache cost counts both segmentation mattes and the denoised twin.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Recover the unpublished denoise work against current upstream. Preserve later opt-in model setup, shared Rust HTTPS downloads, pending-photo resume and an Amount-independent switch. Keep original RAWs unchanged and share the current color finish between ordinary and denoised sources.
Bound public tiling inputs, refuse corrupt-cache exports, and retain synthetic/real-model coverage. Include two Windows path spelling fixes in existing tests. No model weights, face code, media or Tract runtime are included.
Validation: cargo xtask ci (all seven gates); installed RawNIND CPU/GPU agreement 1.38e-6; native CPU preview/export on a public 24 MP Sony RAW, unchanged original SHA-256.
Co-authored-by: Codex <[email protected]>
Co-authored-by: Claude Sonnet 5.5 <[email protected]>
Generated with Codex.
New L1 crate `lightcraft-faces`, the base of a one-drop model install:
- `ModelManifest` (role, licence with a commercial yes / no / unknown flag and a
plain-language notice, provenance, input preprocessing, output, thresholds)
with strict validation: sizes capped, numbers finite and sane, https-only
addresses, bounded text. A manifest from any source is untrusted.
- `known`: models recognised by SHA-256 (YuNet, SFace, AuraFace v1), with what is
and is not known about their licences and training data.
- `onnx`: a streaming, hostile-input-safe protobuf probe of a model's graph
inputs and outputs. It reads a 248 MB model in about 4 ms and never loads the
weights; truncation at every length and random corruption never panic.
- `suggest`: known hash -> its manifest; an unknown model shaped like a face
embedder -> a draft manifest with the assumptions spelled out; anything else
-> a plain reason it is not supported yet.
- `hash`: streamed SHA-256.
Checked against the real model files with an opt-in test (LC_FACE_MODELS). It
found an old exporter listing every weight as an input, now handled. Running a
model (tract) is not part of this crate yet.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Compression 52546 tiles and strips decode through jxl-oxide (new
default-on `jxl` feature of lightcraft-raw): bare codestreams and
ISO-BMFF containers, 1 or 3 planes, 8-16-bit integers scaled back
exactly from the codestream's own bit depth, 16-bit floats. Edge tiles
may be coded at their nominal or in-image size. Headers are checked
against the IFD in header-only probing too. A single-chunk JPEG XL
preview IFD (RGB or grey) is offered as the embedded preview.
The decoder's memory is bounded by the tile: 16 bytes per sample (three
times that for a lossy grey tile, which the decoder expands to three
channels) plus 64 KiB; jxl-oxide 0.12 was measured at <= 10.3 bytes per
sample for lossless and <= 13.5 per colour sample for lossy tiles. A
codestream needing more (e.g. extra channels) is an error.
Lossless tiles are checked bit-exact on synthetic CFA and LinearRaw
files (zune-jpegxl as a test-only encoder); lossy tiles are unverified
on a real file.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
New crate `lightcraft-heif` (a port of PhotoCraft's photocraft-heif): a thin,
panic-guarded wrapper around heic-rs (pure Rust, pinned =0.1.1) - 8/10-bit,
alpha, grid tiles, container irot/imir/clap, ICC/EXIF/XMP. `lightcraft-codecs`
decodes HEIF behind its non-default `heif` feature (HEVC patents are the
distributor's call, the same policy as PhotoCraft); the apps forward it. HEIC
files stay recognised without the feature: opening one is a clear
"isn't included in this build" error. Fuzz-found heic-rs panics arrive as
regression fixtures (inline bytes, no media committed).
Verified end-to-end on real iPhone HEICs (lightcraft-cli render, rotation and
metadata correct). Note for reviewers: measured here, the engine currently
ignores the decoded image's source-space matrix on export for ANY format (a
true Display-P3 PNG round-trips unchanged); the ICC and space are carried for
all codecs, so this inherits - and does not worsen - existing behaviour.
Co-Authored-By: opencode <[email protected]>
The downloader in lightcraft-segment (mirrors, resume, pinned SHA-256, pure-Rust
HTTPS) was already generic apart from four SAM 3 constants. Move it to
crates/fetch (L0, native only) so other model features can use it without
depending on the SAM 3 inference code. lightcraft-segment keeps SAM3_FILES,
DEFAULT_MIRRORS, MIRRORS_ENV and wraps mirrors(); no behaviour change. The
downloader's 13 tests moved with it.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
AI masks from Meta's SAM 3, run in LightCraft itself: a new
`lightcraft-segment` crate ports the Hugging Face transformers
implementation (Apache-2.0) to candle, on the GPU through Metal on macOS
(CPU elsewhere), with no C/C++ and no `unsafe`.
- Object: click to include, ⌥-click to leave a part out (SAM 3's
interactive prompt encoder + two-way mask decoder).
- Describe: a text prompt selects every instance ("sky", "the red car";
"car, road" merges several) via the CLIP text encoder and DETR detector.
- Both as new masks and as Add / Subtract / Intersect components; + / −
buttons under the selected mask combine selections.
- The segmentation is stored with the mask (288² quantized logits, ~10-30
KB), so renders, exports and the web build never need the model.
- A zoomed-in detail pass (background, ~5 s on an M4 Pro) re-segments the
area around a selection from a sharper source: 5-10x finer edges on
small objects.
- Per-selection Edge slider (hard ↔ soft); hovering a mask in the list
shows it in red (the always-on overlay is now off by default, O toggles).
- The engine feature `sam` (on in the desktop app) gates the model; the
image is encoded once per photo in the background, clicks take ~35 ms.
- Weights are not bundled (SAM License): tools/install-sam3.sh / .ps1
download and verify facebook/sam3; docs/ai-masks.md.
- Clearer errors (missing file, missing model) and a stderr logger in the
app (LIGHTCRAFT_LOG=info).
candle is pinned at 0.9.2: later releases make candle-core depend on
tokenizers with the Oniguruma C library. Every stage is checked against
the Python reference (tests/reference.rs, tools/sam3_reference.py): on
Metal the backbone matches to 7e-5 relative error, masks to ~5e-5.
Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01C8BkS7JDqa1EMcjW6spDGA
Fixes the Windows installer: the Start Menu shortcut and Add/Remove Programs icon were blank, and the shortcut description was wrong.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The macOS malloc_zone_pressure_relief call (returns freed allocator pages
after raw decodes) lived in identical copies inside the desktop app and the
CLI. It now lives in lightcraft-sysmem, an isolated L0 helper crate and the
workspace's only crate allowed unsafe: one #[allow(unsafe_code)] module, a
SAFETY comment, a safe API that returns bytes released (0 off macOS), and
tests (repeat calls, release after large frees, concurrent calls). Both apps'
alloc_release.rs now just install the safe hook, and their crate roots gain
#![forbid(unsafe_code)], so every production crate except sysmem forbids it.
AGENTS.md/CLAUDE.md name the exception, per craftrules standards/never-crash.md.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Brandon Thomas