4203 Commits
Author SHA1 Message Date
Dylan M. Taylor d6cc257f55 chore: bump craft-media-source to 6a061a30 (#5158) 2026-11-21 15:10:17 -05:00
Terra854 f77815001f deps: align the shared media cancellation context (#5149) 2026-11-22 02:45:15 +07:00
Terra854andDylan M. Taylor 3d7465142f media: use shared OpenEXR scanline inflation (#5132)
* media: share OpenEXR scanline inflation

* chore: repin craft-exr to storytold/craft-libs#2853

---------

Co-authored-by: Dylan M. Taylor <[email protected]>
2026-11-21 13:28:24 -05:00
Terra854andDylan M. Taylor 176f63972b Delegate FilmCraft media cancellation to the shared request context (#5134)
* Delegate media cancellation to the shared request context

* chore: repin craft-media-source to storytold/craft-libs main (ba25ccd2)

---------

Co-authored-by: Dylan M. Taylor <[email protected]>
2026-11-21 13:22:46 -05:00
FredApps e11edd27ab Merge pull request #3214 from zsoli69/pr/vaapi-derive-image
platform: read VA-API pictures through vaDeriveImage
2026-11-21 17:33:05 +07:00
FredApps f4578faacb #3214: preserve both hardware notes when merging main 2026-11-21 11:08:33 +01:00
584bac5113 engine: migrate legacy still sizes and Motion anchors on open (#5005)
* engine: migrate legacy still sizes and Motion anchors on open

Projects saved before stills applied EXIF Orientation store a portrait
still at the size the file was stored at, unturned, and a clip placed
from it baked its Motion anchor from that size. The picture decodes
upright today, so the project reopens with the still scaled up and
off-centre. On install, a still whose stored size is the transpose of
the size its file decodes to now is recognised as legacy: its stored
size is swapped to the oriented one and the anchor points of clips
placed from it are transposed to match, keyframes included.

* engine: turn legacy still anchors by the file's EXIF Orientation

The migration's plain axis swap (x, y) -> (y, x) is the right mapping
only for Orientation 5. For Orientations 6, 7 and 8 the stored pixel
position reaches the upright picture through a rotation or flip, which
subtracts from a dimension: a stored (4, 6) on a 40x20 still lands at
(14, 4) for Orientation 6, not (6, 4). Centre anchors are invariant
under all four mappings, so the original tests could not see it.

The migration now reads the orientation the decoder applied from the
decoded source itself (MediaSource::still_orientation, reported by
StillSource from the raster it decoded), and maps each anchor, static
and keyframed, by that orientation. The pool's slate wrappers delegate
it, so a pooled source still reports what its inner source decoded.

Two coupled fixes in the same area. A clip placed from a subclip of
the still keeps the subclip's id, so the migration skipped it; it now
collects every item that resolves to the still through resolve_media.
And placing from a subclip baked the anchor from the sequence frame
instead of the parent media's size, so without that fix the transpose
would have moved a sequence-centre anchor: place_item now resolves a
subclip's size from its parent.

Tests: a noncentre anchor with a noncentre keyframe for Orientations
5-8 (fails without the fix at Orientation 6, (6, 4) instead of
(14, 4)); a clip placed from a subclip of the still (fails without
either half of the fix); a DNG carrying an Orientation tag, which the
migration leaves alone because DNG ignores EXIF Orientation. The media
orientation suite now also asserts the orientation each decode
reports.

`image` moves from the engine's dev-dependencies to its dependencies:
the migration and the pool wrappers name `image::metadata::Orientation`
in production code, and a lib build without dev-dependencies did not
compile before the move.

* engine: probe still geometry for the legacy migration and recentre the saved subclip anchor

Two points from the re-review of the #4613 migration.

The migration decided legacy state by resolving a full source for
every nonsquare still in the project, which decoded and retained a
frame for stills that turned out to be current. It now asks
MediaPool::still_geometry, which probes the file's headers through
the new filmcraft_media::still::probe_geometry: each format answers
from the same header its decoder reads, under the same pixel
ceiling, and reports the size and orientation the decode would
apply. No picture is decoded and the pool retains no source.

A subclip placed before the subclip source-size fix saved its anchor
as the centre of a 1920x1080 fallback frame, (960, 540), which is
not a point in the still's coordinates. On a clip placed from a
subclip, an anchor still holding exactly that value with no
keyframes now migrates to the upright centre; edited and keyframed
anchors turn as before.

The media suite proves the probe agrees with the decode for every
orientation across JPEG, PNG, WebP, RGB TIFF and YCbCr TIFF. The
migration suite gains the saved fallback fixture, a second
save/reopen proving idempotence, and a check that opening a current
or legacy project caches no source for its stills.

* engine, media: use bounded header-only reads, require EXIF 5..=8, and expand migration tests

---------

Co-authored-by: fanyzp <[email protected]>
Co-authored-by: Dylan M. Taylor <[email protected]>
2026-11-21 07:09:30 +01:00
FredApps 960a661001 Merge pull request #2351 from inspiretelapps/feat/slip-monitor
Slip Monitor: a live four-up in the Program monitor while slipping
2026-11-20 21:05:40 +07:00
FredApps 15b9187d6f Merge pull request #2366 from inspiretelapps/perf/frame-request-views
perf(ui): Trim Monitor pictures at a cut inside one take stop cancelling each other
2026-11-20 21:04:01 +07:00
0afb2cf22f Adopt the shared AGENTS.md (RFC-0072) (#5090)
* Adopt the shared AGENTS.md (RFC-0072) and test rules (RFC-0071)

* Point comments and doc headers at the shared standards

* AGENTS.md: correct the repository facts

There is no cargo xtask corpus here: say corpora go in photocraft-corpus and that cargo xtask fixtures generates target/fixtures/.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

---------

Co-authored-by: FredApps <[email protected]>
Co-authored-by: Claude Opus 5.5 <[email protected]>
2026-11-20 18:09:44 +07:00
FredApps ae0a71d303 Fix PSD merged row decoding and memory use (#778) (#5074) 2026-11-20 05:45:27 +07:00
.CooCoo 2c2570f41d Pin craft-libs crates to craft-libs main dc6adc3f (#4917)
* Pin every craft-libs crate to craft-libs main 3af2280e

The workspace pinned craft-libs at thirteen different commits, so
Cargo.lock carried two or three copies of craft-audio, craft-media,
craft-cfb, craft-time, craft-timeline and craft-xml. Every craft-libs
dependency (the workspace list, filmcraft-codecs' craft-jpeg and the web
build's craft-web-api and craft-web-host) now uses 3af2280e, and each
crate is in the lock once.

From the crates FilmCraft uses, this brings in craft-libs #733: the web
build's window.filmcraft refuses a call with a busy error while 64 calls
already wait for the app. craft-libs #1503 makes the updater treat an
extracted nextest test binary as a development build. The other changes
are in code FilmCraft does not call (the EXR deep and display readers,
the craft-video Motion JPEG codec, which is not enabled here) or add new
API. craft-font-catalog now depends on harfrust 0.13.3.

* Pin craft-libs crates to craft-libs main 453b0da7

Every craft-libs git dependency moves from 3af2280e to 453b0da7, except
craft-fcproj-format. At 453b0da7 that crate writes project schema 18,
and released builds do not open schema 18 projects. Schema 18 goes in
with #1078, which adds the camera raw Source Settings that schema 18
stores. craft-tiff and the YCbCr alias now resolve to one craft-tiff
package.

- The HEVC picture fixture in crates/codecs sets the new
  bit_depth_chroma field.
- The update prompt tests offer a release and tick the updater until
  its prompt opens, since craft-update now inspects an offered release
  on a thread of its own (storytold/craft-libs#1149).

* Pin craft-psd to craft-libs main 453b0da7

The merge of main brought craft-psd in at 06496f5b. It moves to
453b0da7 with the other craft-libs crates. craft-psd has no changes
between the two revisions.

* Pin craft-libs crates to craft-libs main 08da5e14

Every craft-libs dependency except craft-fcproj-format, including the craft-jpeg pin in crates/export, moves to 08da5e14. Cargo.lock holds each craft-libs crate once.

* Pin craft-libs crates to craft-libs main dc6adc3f
2026-11-20 05:11:01 +07:00
Dylan M. Taylor a8ed9e8516 still: decode PSD composite via craft-psd Merged (#5069)
* still: decode PSD composite via craft-psd Merged

* still: cargo fmt

* still: validate PSD header before Merged::open
2026-11-19 13:51:30 -05:00
Dylan M. Taylor f67d38b35c chore(interchange): bump craft-aaf and craft-timeline to craft-libs main (#5054)
* chore(interchange): bump craft-aaf and craft-timeline to craft-libs main

* fix(captions): convert between filmcraft_time::Tick and craft_captions::Tick

* fix(captions): use filmcraft_captions::Tick in ttml_language test

* fix(engine): adapt caption Tick/FrameRate to craft_captions types

* style: rustfmt
2026-11-18 21:59:34 -05:00
Dylan M. Taylor 3d37c0f1f1 chore(dng): bump craft-raw and craft-tiff to craft-libs main (#5051) 2026-11-18 19:04:22 -05:00
inspiretelappsandClaude Opus 5.5 00fa496944 Slip Monitor: a live four-up in the Program monitor while slipping
With the Slip tool (Y), dragging a clip showed nothing until release. Now the Program monitor
shows the previous clip's Out, the new In and Out (large, side by side, with their source
timecodes) and the next clip's In while the drag runs, plus the offset as a signed timecode and in
frames ("Slip: +25 frames", red "at the media limit" when the media's end holds it back); the
Timeline shows the offset at the pointer. Release commits `timeline.slip`, Escape drops it.

- `timeline.slipPreview` (`engine::slip::preview`) runs `slip_linked`, the function
  `timeline.slip` now calls too, on a copy of the sequence: the preview is the commit at every
  delta, linked picture and sound clamped together (#529, #594), with no project change or undo
  step.
- `panels::slip_monitor` draws with the Trim Monitor's `picture()`, each picture in its own
  frame-server view (`FrameServer::request_in_view`, #2366), so a fast drag keeps about one decode
  per picture waiting and the newest frame wins.
- The slipped sound's waveform slides on the Timeline while dragging.
- New strings in every full-interface catalog (de, es, fr, id, it, ja, ko, pt-br, ru, zh-cn); uk
  is still a small catalog and shows them in English.

Builds on #2366 (frame-server views); #2365 (the `slip_together` i64::MIN fix) is on main.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-11-18 16:43:15 +02:00
inspiretelappsandClaude Opus 5.5 99b5e12d0b test(ui): the Trim Monitor baseline waits for the setup's frame jobs to finish
CodeRabbit on #2366: a cancelled job is only counted when its worker finishes it, so a setup
job cancelled before the baseline could be counted after it. FrameServer::pending_jobs (queued
plus taken by a worker and not yet recorded, counted under the queue lock) gives the test a
barrier: it steps frames until no job is pending, then takes the baseline.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-11-18 16:43:14 +02:00
inspiretelappsandClaude Opus 5.5 b355ec1ec5 perf(ui): a running prefetch a picture asks for survives stopping playback; Trim Monitor test baseline
CodeRabbit on #2366:
- FrameServer::request_job: a request that finds its frame already running clears the job's
  prefetch flag as the queued path does (j.prefetch &= prefetch), so stop_prefetch no longer
  cancels a frame a Trim Monitor picture waits for. Regression test
  a_running_prefetch_a_picture_asks_for_survives_stopping_playback (failed before).
- trim_monitor_frames_ui: the "nothing given up" check counts from just before the edit point
  is selected, so the Program monitor's setup frames can't fail it.
- The 4K benchmark's doc says its job count is jobs finished, not started.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-11-18 16:21:05 +02:00
Dylan M. Taylor 6aa6768494 Merge pull request #4924 from storytold/dot/shared-hdr-20261117
Stills: replace local Radiance HDR decoder with craft-hdr
2026-11-18 09:13:51 -05:00
inspiretelappsandClaude Opus 5.5 ff01b256f8 Merge main into perf/frame-request-views
Conflicts:
- crates/ui-egui/src/frames.rs: main moved request_job onto a FrameRequest struct and
  PoisonError::into_inner, and added the server-drop cancel in run_job; the frame-server view
  (request_in_view, SHARED_VIEW, same_view, the `replaced` count) is carried onto that as a
  `view` field.
- docs/testing.md: main's tests/it paths kept, the Trim Monitor row added with its new path.

Also moves crates/ui-egui/tests/trim_monitor_frames_ui.rs into tests/it/ (one test binary per
crate, docs/testing.md).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
2026-11-18 15:05:10 +02:00
FredApps 688444e75d Merge pull request #5029 from storytold/fix/fcpxml-unusable-lane-warning
fix(interchange): report unusable FCPXML lanes on the live import path
2026-11-18 19:52:04 +07:00
FredApps 6df14ec66f fix(interchange): apply FCPXML adjust-conform on the live import path
#2300's tests call import()/export(), which go through craft-timeline
(pin 57288eda). That pin still ignores adjust-conform, so a missing
element left scale-to-frame off, fill was silent, and export wrote no
type="none". The legacy #[cfg(test)] parser already had the intended
rules.

After a successful live import, the wrapper now applies those rules to
the clips craft-timeline built, and after export it writes
adjust-conform type="none" for clips that keep their native pixels.
roxmltree moves to a lib dependency so the document's DOCTYPE can be
parsed.

Tests: fcpxml_conform::* and the existing interchange fcpxml suite.
Signed-off-by: FredApps <[email protected]>
2026-11-18 18:30:20 +07:00
FredApps 81c351f826 fix(interchange): report unusable FCPXML lanes on the live import path
#2255 added a warning to the legacy FCPXML parser when a connected clip's
lane is not an i32, plus a test that calls import(). Import already goes
through craft-timeline, which still skips those attributes silently, so
the new test failed on main as soon as the PR merged.

Warn from the craft-timeline wrapper instead of bumping the craft-libs
pin (main is hundreds of commits ahead of the pin).

Signed-off-by: FredApps <[email protected]>
2026-11-18 17:04:20 +07:00
NetterSilver 176d072983 Read a hyphen as minus only at the start of a number
A hyphen before digits is read as "minus" only when it starts the text
or follows a non-alphanumeric character. Names and model numbers such as
COVID-19 and F-16 keep a plain hyphen. Negative numbers after whitespace,
brackets, or other punctuation still read as minus.
2026-11-18 16:49:31 +07:00
Will Bunker 9ad3e41397 fix(fcpxml): import and export adjust-conform (fit by default) (#2218)
A clip with no adjust-conform means "fit" in FCPXML: the picture is
scaled into the sequence frame before its adjust-transform. The importer
ignored the element, so a 4K clip in a 1080p sequence landed at scale
100 and was cropped.

Import now sets the clip's scale-to-frame-size flag for fit (and a
missing element) and clears it for none. fill has no equivalent in the
model: it is reported and imported as fit. Export writes
adjust-conform type="none" for a clip without the flag, so FilmCraft
clips keep their native size on a round trip.

An adjust-conform on the <video> inside a <clip> is read first; the
clip's own element overrides it. An inner fill is reported when the
clip does not override it.

Fixes #2218.
2026-11-18 16:37:49 +07:00
FredApps 427992f178 Merge pull request #4741 from ivanao92/export-reuse-renderers
export: parts of a job reuse the frame renderers of the previous part
2026-11-18 16:30:26 +07:00
FredApps f937482d8f Merge pull request #3834 from Zaxaerith/fix/3820-active-dock-tab-highlight
fix(ui): make active panel tabs visually distinct (#3820)
2026-11-18 16:28:48 +07:00
FredApps 1c3e01b8ac Merge pull request #2954 from earlvanze/codex/security-project-media-dimensions
fix(format): reject oversized persisted media dimensions
2026-11-18 16:27:14 +07:00
FredApps 796eec1133 Merge pull request #2708 from kk376/fix/2474-animated-cache
fix(media): bound decoded animated GIF and APNG frames in shared cache (#2474)
2026-11-18 16:25:49 +07:00
FredApps 92551059da Merge pull request #2425 from 00xJS/fix/2421
#2421: the layout in use when the app closes comes back at the next launch, saved workspace or not
2026-11-18 16:24:45 +07:00
FredApps 18c73aed84 Merge pull request #2255 from Wbunker/fix/fcpxml-unbounded-lanes
fix(interchange): FCPXML connected clip with an unusable lane is reported, not silently dropped
2026-11-18 16:23:41 +07:00
FredApps 5491c734ed Merge pull request #1854 from 00xJS/fix/1540-control-no-dialogs
#1540: engine.execute over the control channel never opens a parameter dialog
2026-11-18 16:21:49 +07:00
0b3f30edaa Add a hex field to colour parameters in Effect Controls (#2359)
* Add a hex field to colour parameters in Effect Controls (#1561)

Ultra Key and Color Key could only be given a colour by swatch, by
eyedropper, or by the picker's R/G/B fields in 8-bit or float. None of
those takes `#00ff00`, which is how colours are actually exchanged: a
green screen's published value, a brand colour, one read off a style
guide.

A `#rrggbb` field now sits beside every colour parameter, next to the
swatch and the eyedropper. The typing behaviour matches the hex field in
Settings - Appearance (#722): while the field has focus it shows what is
typed, each valid colour is applied as it completes, and leaving shows
the kept one again, so a half-typed value never sticks. Six digits say
nothing about alpha, and the swatch is Alpha::Opaque, so the parameter
keeps the alpha it had.

Parsing and formatting reuse filmcraft_color::{parse_hex, to_hex}.

Co-Authored-By: Claude Opus 5 <[email protected]>

* Read colour parameters as sRGB, not linear

Review on #2359 pointed out that the swatch used egui's linear `Rgba`
while the new hex field reads the parameter directly. Checking the
renderer settled which is right: `ultra_key` runs each pixel through
`effects::enc` (linear to sRGB) before comparing it with the key colour,
which it takes exactly as stored, so the stored components are sRGB.

The swatch was therefore showing the wrong colour. Ultra Key's default
`[0.0, 0.8, 0.2]` is `#00cc33`, but the picker opened on `#00e77c`.
Reading and writing it as `Color32` through `color_edit_button_srgba`
puts the swatch, the picker and the hex field on the same value, and on
the one the keyer actually uses. Alpha is untouched, as before.

Adds a test pinning the convention, so a later change back to linear
fails instead of silently shifting every colour parameter.

Co-Authored-By: Claude Opus 5 <[email protected]>

* Cover the picker itself, and fix Lumetri's tints too

Review asked for a test that goes through the picker: the previous one
read the stored parameter, so a return to the linear widget would have
left it passing. The swatch now carries an automation id, and the new
test opens the picker and reads what it shows. Turning the widget back
to `color_edit_button_rgba` makes it fail with
`R 0 | G 231 | B 124`, which is the linear misreading.

Looking for other callers turned up the same pattern in the Lumetri
Color panel, where Shadow Tint and Highlight Tint went through
`Rgba` as well. Same parameters, same encoding, so they get the same
treatment; their alpha handling is unchanged.

Co-Authored-By: Claude Opus 5 <[email protected]>

* #1561: the eyedropper and the transition colour picker use sRGB like the colour field

Co-Authored-By: Claude Opus 5.5 <[email protected]>

---------

Co-authored-by: Claude Opus 5 <[email protected]>
Co-authored-by: skylar taylor-barrick <[email protected]>
2026-11-17 22:20:17 -06:00
MarekSpodBiedryandskylar taylor-barrick b6a37d19db Fix clippy expect_used under --all-features and the no-threads build (#2813) (#2838)
* Fix clippy expect_used under --all-features and the no-threads build (#2813)

* xtask: root() falls back to the manifest dir instead of expect, so the whole-workspace expect_used gate passes (#2813)

---------

Co-authored-by: skylar taylor-barrick <[email protected]>
2026-11-17 22:11:29 -06:00
Dylan M. Taylor 1b0f9f8e7b Merge pull request #5006 from storytold/dot/mjpeg-export-shared-20261117
Export: software MJPEG through shared JPEG encoder
2026-11-17 19:40:28 -05:00
Dylan M. Taylor 894a773207 Use merged tokenless JPEG encoder for bounded MJPEG scratch 2026-11-17 17:45:44 -05:00
Dylan M. Taylor 9e15e2a8e2 M2: encode software MJPEG through shared craft-jpeg 2026-11-17 15:46:35 -05:00
Dylan M. Taylor 73ae4dfac0 Merge remote-tracking branch 'origin/main' into HEAD
# Conflicts:
#	crates/media/Cargo.toml
2026-11-17 08:23:01 -05:00
Dylan M. Taylor 93a88a839a Merge pull request #4926 from storytold/dot/film-shared-psd-packbits-20261117
Stills: share strict PSD PackBits rows without allocation overhead
2026-11-17 00:05:26 -05:00
Dylan M. Taylor f5cff8c85e media: pin PackBits decoder to merged shared prerequisite 2026-11-16 23:11:10 -05:00
Dylan M. Taylor a04c021988 Merge pull request #4927 from storytold/fix/ycbcr-metadata-budget
Stills: bound YCbCr TIFF metadata before typed array allocation
2026-11-16 23:09:17 -05:00
Dylan M. Taylor 17beb2d574 Merge remote-tracking branch 'origin/main' into HEAD
# Conflicts:
#	crates/media/Cargo.toml
2026-11-16 23:05:26 -05:00
Dylan M. Taylor 4da80fa02d Merge remote-tracking branch 'origin/main' into HEAD
# Conflicts:
#	crates/media/Cargo.toml
2026-11-16 23:03:59 -05:00
Dylan M. Taylor 39c075a96a media: bound decoded YCbCr metadata before allocation 2026-11-16 22:35:43 -05:00
Dylan M. Taylor 737060c0b4 media: pin shared HDR decoder to merged craft-libs main 2026-11-16 21:59:50 -05:00
Dylan M. Taylor d896af61ce Stills: integrate shared YCbCr TIFF decoder and contributor orientation work (#4925) 2026-11-16 21:58:27 -05:00
00xJ0 04b2b6540c #1540: engine.execute of voiceover.settingsDialog answers the settings, not the window
From the menu, Voice-Over Record Settings… still opens its dialog. Over
engine.execute it now answers the voice-over settings (as
audio.voiceover.settings does) instead of opening a window nobody can see.
The every-dialog-route test checks the window's automation elements, since
the dialog's draft lives in egui memory outside the inspected UI state.
2026-11-16 19:19:58 -07:00
Dylan M. Taylor 1a4e060cdb media: reuse shared PSD PackBits decoder without row allocations 2026-11-16 21:05:11 -05:00
Dylan M. Taylor 2aae01f68c Merge pull request #4083 from vangel76/fix/platform-nvenc-tests-serial
platform tests: take NVENC_SERIAL in every NVENC test, fixes hangs under plain cargo test (#4075)
2026-11-16 20:31:35 -05:00
Dylan M. Taylor 544723c75b Stills: use shared bounded Radiance HDR decoder 2026-11-16 19:17:12 -05:00