The proxy's largest visible error against the original was the JPEG encoder's
quantised 8x8 DCT DC - each block's own mean, luma and chroma - which shows as
block steps across the frame and a small constant colour shift, and grows with
saturation. Quality 92 -> 97 shrinks those quantisers to about a third (block
-mean dE76 0.138 vs 0.202 on the fidelity test's smooth scene), at about twice
the file size; what remains is the 8-bit sRGB quantisation both encoders share,
which only a lossless proxy format can close.
The header gains an `enc` field (encode version 2) and the folder's look marker
becomes "<look> <enc>": a proxy or marker from before the encode version is
rebuilt/rescanned once, so existing libraries pick the new quality up without
any command. Old files still decode bit-identically (the scale mapping is
untouched, decode ignores `enc`), and a new proxy read by an older build is
left alone by its staleness check.
The fidelity test pins the gain with a frozen copy of the old encoder and a
block-mean dE76 metric (per-pixel dE is dominated by the shared 8-bit floor);
the 0.8 bound is the geometric midpoint between the no-op ratio 1.000 and the
0.632 floor cap the 8-bit quantisation sets on this scene.
Merged HDR DNGs lost the camera's embedded lens correction: load_frame strips the lens opcodes from each
source (frames are merged uncorrected) and write_linear_dng wrote no OpcodeList3, so Optics ▸ Enable Lens
Corrections did nothing on the result (Sony ILCE-7M3/7RM4A, Panasonic/Leica, DNGs with their own opcodes).
Each raw frame now keeps the lens corrections it applies (the DNG 1.6 skip rule already applied),
re-expressed for the default crop develop keeps by the new Opcode::cropped: the centre moves into the crop
and coefficients are rescaled for the crop's farthest corner. HdrResult carries the reference frame's and the
DNG writes them as its OpcodeList3. Frames turned before merging (mixed orientations) and panoramas carry none.
Tests: cropped corrections give the crop of the corrected whole (WarpRectilinear, WarpRectilinear2 with and
without reciprocal, FixVignetteRadial); an HDR merge of cropped, corrected brackets, upright and vertical, gives
the same profile correction as its reference photo (fails without the fix). Real brackets not measured yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015gRtdWF1E9358FfzhzYqCS
- drag_latency (ui-egui example): the whole app driven headless at 60 frames a second; per develop control,
the latency of each drag tick until its draft is the loupe (or, with ZOOM, the zoom window) texture, the
ticks never shown, the UI thread's frame time and the time to the full-quality render after release.
- drag_tick_bench (engine example): the UI-thread work of one drag tick outside the render.
- render_bench: ONLY=window (a 1:1 zoom window cut from the original: first window, pan, slider changes;
GPU_ONLY=1, DUMP=file for pixel A/B), ONLY=load / load1 / jpeg (+ LOADONLY=1): the steps of opening a photo.
The window rows carry no colon in their names, so `cargo xtask bench` records each of them (parser test).
- open_timeline (ui-egui example): opening raws in Detail through the real app logic, headless with the GPU.
Co-Authored-By: Claude <noreply@anthropic.com>
MediaCache::enforce_budget never evicts the original of the photo on screen (set_on_screen), but it still
counted it. A 64 MP original (Sony A7R VI: 9984×6656 linear float, 760 MiB) nearly fills the default cache
share (768 MiB) alone, so the loop evicted everything else instead: the photo's preview and thumbnail sources,
the rendered previews and the prefetched neighbours. Every edit at 1:1 (the whole-frame render and the
thumbnail), every zoom out and every step to the next photo then decoded the raw again.
The pinned original is now left out of what is compared with the budget. The rest still fits the share; the
original sits over it, as the first commit of this PR already allowed.
Co-Authored-By: Claude <noreply@anthropic.com>
Each drag tick changed the edited photo's settings, so the filmstrip asked
for a new thumbnail render every frame. For an unedited raw shown with its
embedded preview, the thumbnail-level source had never been decoded, so each
of those renders decoded the whole raw (a 64 MP ARW: over a second), several
side by side, and the loupe's drafts queued behind them.
- The edited photo's thumbnail is not re-rendered during a slider drag while
it has one; it follows once the drag ends.
- A thumbnail of a photo whose preview-level source is decoded (the one in
the loupe) is cut from that source (SourceRef::Fit) instead of decoding the
file again; not when the photo has a denoised picture.
Co-Authored-By: Claude <noreply@anthropic.com>
Sony ARWs of cameras without measured colour matrices were edited on the
relative scale (6500 K / 0 = as shot). Their SR2SubIFD stores the
multipliers of the camera's presets at 2500, 3200, 4500, 6000 and 8500 K;
lightcraft_raw::kelvin_presets reads the as-shot multipliers on the curve
through them and they take the spectral matrices' place in the #730
design (as-shot Kelvin and tint only; look, Bradford adaptation and the
pixel-identical legacy upgrade unchanged).
Covered: ILCA-99M2, ILCA-77M2, SLT-A33, DSC-RX10, ILCE-3500, ILCE-7S.
Kelvin-dial shots read back 3400 -> 3396, 5200 -> 5193, 4700 -> 4708,
5600 -> 5594 K; Shade 7493 K. Off-curve constants (isotherm tilt 0.74,
tint spacing 1) measured from the 14 Sony bodies with both spectral
matrices and presets.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add correlated dark holdouts that reject misleading training and accept
matching training. Check finite, monotone curves in both cases. Confirm
the regression fails with the error gate removed and passes when restored.
Qualify the improvement claim when no original curve can be fitted, and
explain the source-look process policy and offline smart preview refresh.
Co-Authored-By: Codex <noreply@openai.com>
Use separate luminance pairs for tone fitting so the camera look follows
its JPEG into deep shadows. Keep colour pairs and holdouts unchanged.
Require enough correlated shadow samples and lower held-out error, with
noise and compression floors and the existing median curve fit.
Add synthetic regressions, bump look and render cache versions, and
record the public multi-make and private numeric validation in the docs.
Fixes#1307
Co-Authored-By: Codex <noreply@openai.com>
Resize::apply rounds the size, then the render fit the frame into that box
again: on a frame a hair under 3:2 the refit gave 1333 x 1.4999999 = 1999.49
and lost a pixel of the long edge. Frame::fit now keeps a box that already has
the frame's aspect up to rounding, and output_size rounds the native size to
whole pixels first so float noise can't decide a half-pixel tie.
An uncompressed Leaf Aptus 22 MOS keeps its 16-bit mosaic in IFD0
(compression 1, BlackIsZero, one sample per pixel) with no Make tag and
no CFA photometric, so `probe` returned None and the file imported as a
greyscale image of the mosaic. Leaf's data block, tag 0x8606 (LeafData
in the ExifTool EXIF tag-name documentation), now marks such an IFD0 as
a raw: OtherTiff, preview only with the reason, as the lossless-JPEG MOS
already is. A rendered RGB image that kept the block stays an image.
The Hasselblad Lunar is a Sony NEX-7 that writes an ARW under Make
HASSELBLAD; only SONY files reached the ARW reader, so it opened as its
preview. A HASSELBLAD file with Sony's raw coding (an IFD with
compression 32767 or the tone-curve tag 0x7010, which no 3FR / FFF has)
now goes there and decodes like a NEX-7: black 512, the SR2SubIFD white
balance. The Tag2010 white-balance fallback, the one table arw.rs keys
by model, takes the body from the maker note's SonyModelID (0xb001)
when the file is not Sony-branded.
Docs: the Lunar is verified in docs/raw/evidence.tsv (236 verified),
raw-parity.md, parity-checklist.md, gaps.md and whats-new.md.
RENDER_CACHE_VERSION 61 -> 62: a Lunar's thumbnails were rendered from
its preview.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
For a DNG with its own colour matrix, an identity or absent ProfileToneCurve, no
LookTable and no ProfileGainTableMap, fit only the tone and chroma curves to the
camera JPEG, with the file's colour model untouched (profile tables and embedded
vignetting applied to the sensor proxy as the loader renders it). The fit is used
only when the preview is at least 90% of the raw's long edge, shows the same
picture (structure gate), passes the existing held-out acceptance, and does not
put more of the photo near white than the plain render (highlight gate, at 2x
proxy luminance). Otherwise the render is unchanged. Other formats and
catalogued photos are untouched. Render cache version 62.
library.up / library.down (Up / Down) go to the cell above / below in the
grids (nearest by position in justified rows; the top / bottom row stays
put), and step one photo in the filmstrip.
Sources: own design.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Go to History Step (and new Previous/Next History Step commands) loads the
step's settings and moves a per-photo cursor (`Photo.history_cursor`,
`Op::SetHistoryCursor`, catalog format v8; older catalogs open at the latest
step) instead of appending "History: …". Later steps stay listed, greyed in
the History panel, and can be returned to; a new edit made behind the end
drops them first. Each move is one undo step; undoing an edit restores the
dropped steps and the cursor. `history.list` returns `cursor`.
Also: spots.rs uses as_chunks (clippy chunks_exact_to_as_chunks).
Sources: own design (Lightroom Classic / Photoshop History panel behaviour, observed).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A decoded 64 MP original (Sony A7R VI: 9984×6656 linear float, 797 MB) is bigger than the
default cache share (half of 1.5 GiB). MediaCache::enforce_budget evicted it as soon as the
render handed it over, so every pan of a 1:1 view and every slider move while zoomed decoded
the raw again (and lost the shared GPU copy, keyed by the CPU Arc).
The frontend now tells the cache which photo Detail shows (MediaCache::set_on_screen, set per
frame in panels::compare::prepare); that photo's original is not evicted for the budget and is
let go as soon as another photo or view is shown. Keying it on "no newer preview of another
photo" instead does not work: the neighbour prefetch stores previews after the original.
The decoders keep the alpha plane, but to_working converted only the
colour, so transparent pixels kept the black most encoders store under
them: renders, previews, exports and contact sheets showed it. The
working image is now flattened onto white, in linear light like the
rest of the pipeline.
- Files with no capture date in their metadata or sidecar (e.g. exports saved without
metadata) take the file's modification time, as Lightroom does, instead of all landing
under "Unknown Date" (the loupe overlay showed the import date for them). The import
review and its folder/rename templates use the same date.
- The import review counted files with the same bytes as another file in the source as
"already in the library" (8 on a brand-new, empty library); they are now shown apart as
"identical copies".
- View > Grid Info captions are drawn on Square Grid only; choosing one from Photo Grid now
switches to Square Grid instead of seeming to do nothing.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
print.exportPdf now converts each composed page to the job's colour target and writes it into a
print-ready PDF:
- PrintJob gains `color` (PrintColorSettings): Managed by Printer (sRGB / Adobe RGB, profile
embedded, no output intent) or a printer / paper .icc / .icm converted here and named as the
document's output intent; perceptual or relative colorimetric with black point compensation;
Print Adjustment brightness / contrast; print sharpening Low / Standard / High for Matte or
Glossy; 16 or 8 bits per component. print.set / print.get and print templates carry it;
out-of-range values are refused when set.
- Each photo is sharpened for the paper right after it renders at its print size.
- print/output.rs converts and Flate-compresses one page at a time (memory holds one page's device
samples) into ICC-tagged images through engine::pdf; the PDF is 1.6. The 256 MiB cap counts the
compressed pages.
- A printer profile that can't be used (missing, too large, not a profile) fails the print before
anything renders, naming the file.
- contact_sheet's PDF writer goes back to private: only the contact sheet uses it.
Tests: a printer profile reaches the PDF as ICCBased images plus an output intent, a junk profile
fails naming the file, colour settings validate with the job, sharpening and media round-trip
through print.set / print.get, a missing profile fails before rendering with no activity row;
the existing print and CLI tests follow the PDF 1.6 output. Docs: LRC-PRINT-JOB and gap 15.
engine::print_color turns composed pages into a print-ready PDF:
- Profile: Managed by Printer sends the page in sRGB or Adobe RGB with that profile embedded and
no output intent (the driver converts); a printer / paper .icc / .icm converts the page to the
device's values here and becomes the document's output intent (named by the profile's
description), so the driver must not convert again. Profiles over the size cap, unreadable or
invalid files are errors that name the file.
- Intent (perceptual / relative colorimetric), black point compensation and print adjustment
brightness / contrast, through codecs::print.
- Print sharpening Off / Low / Standard / High for Matte or Glossy paper, per photo at its print
size, with the export sharpener.
- 16 bits per component by default, 8 for smaller files.
PrintColorSettings is serde (camelCase, defaults for missing fields) so print templates can save
it. Page sizes are checked with overflow-safe arithmetic. Not wired to the print layout yet.
engine::pdf writes print-ready documents: pages of any size in points with raster images placed
on them, each image tagged with its ICC profile as an ICCBased colour space (RGB, CMYK or grey),
stored losslessly with Flate at 8 or 16 bits per component (16-bit samples big-endian), or copied
as a JPEG stream where file size matters more. A profile shared by several images is written once.
An optional output intent names the printer / paper profile the file was made for (PDF/X-style
/OutputIntents in the catalog); the Info dictionary carries the title, as UTF-16 when it isn't
plain ASCII.
PDF 1.6 (16-bit images need 1.5), classic cross-reference table, deterministic bytes. Pages, images
per page, page and image sizes are capped; bad sizes, placements, sample counts and component
counts are errors, never panics. Tests read the files back through the xref table: image samples
and profiles inflate to what went in, placements reach the content streams, one profile object is
shared. Not wired to the print job yet.
File → Print… (⌘P) opens the Print module as a dialog: settings on the left,
the page drawn to scale on the right with the photos as they will print (fit
or Zoom to Fill, turned by Rotate to Fit, Photo Border, Inner Stroke, page
background).
Settings column: templates (save, apply, delete), layout style, image
settings, grid / package cell controls (standard and custom cell sizes, Auto
Layout, New Page, Clear Layout), rulers and guides (units in/cm/mm/pt, snap
off/cells/grid, grid size, margins, bleed, page grid, image cells,
dimensions), page setup (paper, orientation, margins), page background and
the print job.
Preview: rulers in the chosen unit, dashed margins and bleed, page grid,
dimension labels, page navigation. In the package styles cells are selected,
dragged (print.cell.move) and resized from a corner handle
(print.cell.resize) with the guides' snapping, arranged from the inspector
row or a context menu, and given a photo (Custom Package). Every change is a
print.* command; the drag is measured from where the pointer went down so
the click-versus-drag slop is not lost.
Print to PDF… picks a file and renders on a background thread with an
activity-stack row, cancellation and original protection, like the contact
sheet, which keeps working unchanged.
Headless tests drive the dialog with real pointer input through the control
channel (print:cell:N, print:resize:N, button:print*): styles and checkboxes,
add / drag / resize / arrange / delete, snapping to a neighbour, Auto Layout
and pages, templates, the PDF, the activity row, and that the dialog and its
buttons fit the window.
The Print module's state lives in the session (Session::print), so the app,
CLI, control channel and MCP share one print: layout style (Single Image,
Contact Sheet, Picture Package, Custom Package), page setup, grid, image
settings, page background, guides and job. 18 commands, none journaled
(nothing here touches the catalog):
print.get / set / reset set merges a partial patch, validates the whole
result and changes nothing on error; a new
paper or margin pulls cells back inside
print.cell.add|move|resize|delete|stack|photo|page
print.layout.autoLayout|newPage|clear
print.exportPdf {path, ids?} composes and writes the PDF with an activity
row, refuses to replace an original
print.template.list|save|apply|delete
Templates keep layout, image settings, page and job (not the photos assigned
to cells, not the guides) and are saved in the library's prefs.json next to
the export presets. They are stored as JSON and checked when applied, so a
template from another version or a hostile one is an error on use and never
makes the preferences file unreadable.
Tests drive every command through the dispatcher (hostile numbers, unknown
keys, caps, persistence across reopening a library) and a CLI end-to-end test
chains set / add cells / Auto Layout / PDF in one run.
print/compose.rs turns a layout and its photos into PageRasters: 16-bit
interleaved RGB encoded in Rec. 2020 at the job's pixel density. Each cell's
photo is rendered with the export pipeline at exactly the size the cell needs,
then fitted (whole photo, centred) or zoomed to fill (centre crop), turned a
quarter clockwise when Rotate to Fit makes it print larger (or crop less),
framed by the Photo Border (package styles only) and finished with the Inner
Stroke drawn inside the photo edge. The page background, border and stroke
colours are given in sRGB and converted to the page encoding.
Photos go to cells by layout style: Single Image and Contact Sheet paginate
the selection, Picture Package repeats the first photo, Custom Package prints
each cell's own photo. The job is prepared on the session and run off it with
progress and cancellation, like the contact sheet; page and job pixel counts
and page counts are capped so a hostile resolution or paper cannot exhaust
memory.
print/output.rs holds the interim PDF writer (Rec. 2020 to 8-bit sRGB JPEG
pages sized to the paper in points) through the contact sheet's Pdf, now
pub(crate). PrintJob carries the print resolution (72 to 720 ppi, 240 default).
Tests draw on synthetic images for exact pixels (borders, strokes, crop,
rotation direction, resampling) and on the demo library for whole pages.
crates/engine/src/print/layout.rs is the pure page model the Print module
builds on: ten named papers plus custom, orientation, margins and the
printable area; Single Image, Contact Sheet (grid with spacing and square
cells), Picture Package and Custom Package styles; package cells with a
stacking order (send to back / backward / forward / bring to front); a
deterministic MaxRects packer behind Auto Layout (largest first, earlier
pages filled before a new one opens, a cell turned a quarter only when it
fits nowhere as given); snap-to-cells and snap-to-grid helpers; ruler units
and guide options.
All lengths are inches in page space, after orientation. Every number from
outside is checked (NaN, infinity, negative, oversized, cell and page caps)
and Layout::validate guards saved layouts.
No command or UI reaches it yet, so the parity rows keep their status.
With XMP auto-write on, photo.readMetadataFromFile committed what it read
and the command handler then wrote the sidecar back, adding lc:settings to
a sidecar that had only another app's crs: settings. Later reads prefer
lc:settings, so LRTimelapse's next crs: edits were ignored. The read now
sets the existing one-command skip_auto_write switch (as photo.removeRegion
does); edits made afterwards are still auto-written.
Tests: a read leaves the sidecar byte for byte and a later develop edit is
still written; the issue's crs:Exposure2012 0.5 -> 1.2 round trip is read.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A manual album shown in its own order sorted with position() per photo
(O(n^2): 2.0 s for 20k photos, now 18 ms). album.addPhotos, removePhotos
and toggleTarget test membership with sets. The 100k scale test measures
all three.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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 <noreply@anthropic.com>
EngineError gains Cancelled and ModelMissing { model: AiModel } and a
kind() -> ErrorKind, so frontends decide on what an error is rather than
on its words. The face detector, SAM 3 model checks and the Lightroom
job produce them at the source; the error text is unchanged, so control
channel and MCP clients see the same `error` strings.
Background results that cross the wire as JSON get an additive flag
instead: SAM 3 download status `cancelled`, Apple Photos import outcome
`cancelled: true` next to `skipped`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A panic inside RenderJob::run or QuickJob::run (a raw decoder or pipeline bug) unwound the
job pool's worker without sending a result. The slot stayed pending and its key was never
asked for again, so the loupe or the thumbnail rendered forever; and the pool never
replaces a dead worker, so after as many such panics as it has workers (2–6) no photo
rendered until LightCraft was restarted.
Both now catch the panic (guard::catch) and return it as the job's failure, answering the
request it came from, so the renderer records it like any failed render: no retry until
the catalog changes, and the workers live on. run() is the one place every render goes
through (the pool, export, Build Previews, contact sheets, face and AI-mask workers);
callers that had their own guard keep working, and the watermark preview's, now
redundant, is gone. The web build aborts on panic, so nothing changes there.
Co-Authored-By: Claude <noreply@anthropic.com>
A lensfun profile is scaled by the crop factor of the recorded frame. `lookup` took the body's from the
database whenever the camera was in it, so a full-frame body shooting a crop of its sensor (APS-C /
Super 35 mode, or switched there by an APS-C lens) had the profile applied as if the crop were the full
frame: the crop's corners got the full frame's corner distortion and vignetting. On a Sony ILCE-1M2 in
APS-C mode with the FE 100-400mm F4.5-5.6 GM OSS at 400 mm, the warp's r³ coefficient was the full
frame's 0.0215 instead of 0.0096.
The 35 mm equivalent focal length shows the crop: when it implies a crop factor more than 10 % above the
body's, that wins (`frame_crop_factor`); within 10 % the difference is the tag's whole-millimetre
rounding and the body's value stands. Without the body in the database the 35 mm equivalent is used, as
before.
Tests: `in_camera_crop_mode_uses_the_frames_crop_factor` (fails on main: in its profile the APS-C frame's
vignetting gain is 0.40 instead of 0.18) covers APS-C mode, full frame, a 35 mm equivalent rounded by a
millimetre, no body in the database and a hostile focal length. Checked on 19 private ILCE-1M2 ARWs
(FE 100-400mm GM, Sigma 24-70mm F2.8 DG DN | Art, full frame and APS-C, firmware 2.01-4.00): every
APS-C coefficient is the full-frame one scaled by 1/1.5².
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Photo Merge decoded each raw itself and wrote the reference frame's own colour
tags into the merged DNG. A CR2 (and NEF, ARW and the other raws whose colour is
fitted from the camera JPEG) has none, so an HDR merge of CR2s rendered with no
camera colour: green and about 16 L* dark.
`lightcraft_merge::load_frame_with` takes the raw's colour tags from the caller,
and the engine passes `starting_colour_tags`: the colour the raw opens with
(`starting_colour`) as DNG tags. That is one forward matrix with a colour matrix
and AsShotNeutral that agree with it at the as-shot white, the look's hue/sat
table as ProfileHueSatMap and its tone curve as ProfileToneCurve. Raws with
their own matrices (DNG) keep their tags. The look's chroma curve has no DNG
tag and is left out.
On two CC0 CR2s (EOS 550D, 40D) an HDR merge of three copies against the photo
alone: mean CIE76 dE 26.15 -> 1.07 and 22.89 -> 0.47.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conflicts in crates/ui-egui/src/lightroom_import.rs: main moved the
Lightroom import's progress from its own window to the shared activity
stack, and this branch had added a second bar to that window.
Resolved by keeping main's activity row and feeding it this branch's
step-by-step progress (LightroomProgress): the row counts the step under
way (the catalog, reading the photos, adding them) and names it, so it
moves while the photos are read. The progress window stays removed; the
whole-import fraction is still in the task's status.
The engine half of this branch merged without conflict and is what main
still lacks: reading the photos is counted, and Cancel stops it.
main's two activity tests now drive the task through LightroomProgress.
"Reading photos…" and "Adding photos…" are added to the language
catalogs, cut from the existing "… N of M" translations.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sources: TIFF 6.0 and own black-box analysis of the three CC0 samples;
white-balance estimate from the embedded preview: own design.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Nikon bodies from before ~2016 record no CropArea, so the decoder kept
their whole sensor data, less trailing dark columns. Where that data
holds masked lines above black or optical black, they were developed as
a coloured or dark bar on that edge: the D7000 and D5100's 36 columns at
~600 (#976), the D800's 14 (#914, the top in portrait), and others.
MEASURED_BODIES (crates/raw/src/vendor/nef.rs) now gives 36 model /
data size / depth groups the frame of their camera JPEG inside the data:
D7000, D5100, D800, D800E, Df, the D4's 12-bit mode, D3, D3S, D700,
D300, D300S, D600, D610, D810, D4S, D5600, Coolpix A, D3000, D40X, D60,
D80 and D2Xs. An entry matches model, data size and depth exactly; the
measured white level becomes optional (only the D4's 14-bit mode has
one). Bodies whose data holds only picture beyond the camera frame get
no entry and render as before.
`cargo xtask raw-frame FILE...` (crates/engine/examples/raw_frame.rs)
measures a body: it aligns each raw's embedded camera JPEG with the
sensor data (half-size luminance, gradient correlation, sub-pixel peak;
weak, ambiguous or between-pixel peaks are flagged), counts the edge
lines that hold no picture and reports what the decoder's crop keeps
beyond the camera frame. It prints entries only with evidence: at least
3 clean files agreeing exactly, of the model (12- and 14-bit pooled) or,
with at least one of its own, with sibling models of the same readout
(data and JPEG size) that agree on the same frame. On the D7500, which
records its CropArea, it finds that area exactly; it declines the Z f,
whose JPEG is distortion-corrected.
It works for any make whose raw embeds a full-size camera JPEG: on
Sony, Canon CR2 / CR3, Pentax and Samsung files it finds the decoder's
own crop (relative to the active area).
docs/nikon-bodies.md surveys every Nikon raw on raw.pixls.us (98
models, 291 files; 138 CC0 samples measured): 22 models record
CropArea; of the 88 groups without it, 36 have entries, 13 are fine, 25
could not be aligned and 7 do not decode. The D3200, D3100, D200, D50,
D750 and Nikon 1 V1 / V2 are pending, each listed with the files it
needs; the doc also describes how to add a body. raw-parity.md no
longer claims every NEF coding but HE decodes.
measured_bodies_are_consistent checks every entry lies inside its data
and is listed once.
The evidence is committed in docs/nikon-bodies/: headers.json (every
file's raw size, depth and maker-note crop tags), samples.json (the 138
measured CC0 files with URL and sha256) and frames.jsonl (the tool's
per-file results). `cargo xtask corpus --download nikon` fetches the
samples (sha256 checked) so `cargo xtask raw-frame` reproduces every
count.
Fixes#976Fixes#914
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
New export format `rendereddng` (export dialog: Format DNG, "DNG of: Rendered
photo"; `app.export`; MCP export tool): the rendered photo, with the edit,
crop, size, output sharpening and watermark baked in, written by the new
`lightcraft_raw::write_rendered_dng` as a LinearRaw DNG of 16-bit float Deflate
tiles in the export colour space. ColorMatrix1 (XYZ -> space), ForwardMatrix1
(space -> XYZ D50), a neutral AsShotNeutral and an identity ProfileToneCurve
make raw converters show it as rendered. HDR edits keep highlights above white.
Bad input (wrong sample count, zero or huge sizes, NaN/inf, out of half-float
range) gives an error or is clamped, never a panic.
Also fixes the Rust 1.99 clippy lint chunks_exact_to_as_chunks in
pipeline/spots.rs that fails `cargo xtask ci` on main.
Sources: Adobe DNG Specification 1.4+ (LinearRaw, ColorMatrix/ForwardMatrix,
ProfileToneCurve); own design.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Brush strokes record a pen pressure (0..1) per point from egui touch
force while Brush > Pen Pressure is on (default). Each dab's radius goes
from a tenth to all of the brush, and its flow is proportional to the
pressure, in the committed stroke and the live preview alike. A mouse
stroke stores no pressure, so documents and renders are unchanged. The
GPU mask kernel leaves pressure strokes to the CPU. mask.brushStroke takes
`pressure`; bad lists are refused. Also satisfies clippy 1.99's
chunks_exact_to_as_chunks in spots.rs.
Sources: own design.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The header-only path added in #755 answers for a Fuji RAF, whose preview pointer sits in the first
bytes. A Canon CR3 keeps its preview in a JPEG track, and a TIFF raw (CR2, ARW, DNG, RW2, 3FR …) can
keep its preview tag in an IFD the prefix cuts short: both fell back to `std::fs::read` of the whole
file — tens of megabytes for a JPEG of a few, on the format of the largest bucket in gaps.md (Canon,
33 % of photographers).
`preview_window` now asks `lightcraft_raw::sparse::preview_plan` — the module #961 added for the
import path, which had no production caller — which ranges `embedded_preview` reads, and reads
exactly those: `planned_preview` fills a buffer of the file's length, reads the ranges the plan names
and asks again until it says `Ready`. The buffer is the file's length, which the memory gate counted
already, and its untouched pages stay zero, so what this path costs is the bytes read — on macOS a
freed buffer of that size may sit in the allocator's cache rather than go back to the OS (measured on
a 38k-photo import, #832). The loop gives up past 60 % of the file, the bound step 2's partial read
uses, and a plan that does not fit, a file past 1 GiB or a window that does not decode all end in the
whole-file fallback.
The quick path hands the JPEG to the decoder alone, so two container facts the whole-file path applies
have to be visible in the bytes it is handed, or the file declines and the whole file keeps the
preview:
- a preview colour space of Adobe RGB (`decode_raw_preview` applies it; a JPEG-only decode has the
JPEG's own space, else sRGB). A container that says sRGB does not decline — that is what the decode
assumes anyway, and declining it took the fast path away from a real Sony ARW;
- the file's own EXIF orientation, when the preview JPEG carries none of its own: the plan's header is
a bounded 64 KiB (the header route grows its prefix to the preview's declared extent instead), and
an orientation tag it cannot see would show the preview unrotated.
Measured over the whole CC0 corpus (111 raws): 52 files take this route, reading a median 7.8 % of
their bytes (0.9-27.7 %) — RW2 3.3 %, Leica RWL 4.1 %, ARW 7.5 %, CR3 8.4 %, DNG 9.7 %, CR2 14.3 % —
while the 21 RAFs stay on the header route (median 4.5 %). The other 38 read whole: NEF, NRW, ORF, PEF
and SRW because the planner does not describe them (an SRW plans to the whole file, which the 60 %
budget then declines), plus the ARWs, CR2s, CR3s and DNGs that declare Adobe RGB or whose plan wants
more than that budget — the bytes they read are the ones the plan would have read anyway. Across all
111 files the plan handed on the same preview as the whole file, with the skipped bytes zeroed and
again 0xFF. `LIGHTCRAFT_PROFILE=1` prints
[profile] preview …/arw-sony-a7rm4a-compressed.arw: read 688128 of 61832192 bytes (plan, window 483441)
Tests: a synthetic TIFF/DNG whose preview only its tags locate (identical pixels to the whole-file
path, the CFA strip unread), the same structure without the tag that makes it a raw, a file whose
orientation only the whole file shows, and a file that states Adobe RGB against one that states sRGB;
the opt-in real-file test reports which path answered.
Co-Authored-By: DeepSeek V3.2 <noreply@deepseek.com>
bflatastic