* Tests: one integration-test binary per crate
Each crate with several tests/*.rs files keeps them as modules of tests/it/,
so cargo links one test executable per crate instead of one per file. Files
that change process-wide state or time a workload stay binaries of their own.
* Profiles: line tables for workspace crates, no debug info for dependencies
* Windows ARM64: nightly and on release pushes instead of every push to main
* xtask ci: clippy and tests in the dev profile
The release profile's thin LTO and single codegen unit made linking the test
binaries take most of the gate. The codec and audio crates stay at opt-level 3
in the dev profile, so their tests run at release speed.
* xtask ci: nextest, doctests, no examples, one wasm check
Tests run with cargo nextest when it is installed (cargo test otherwise) and
build with --lib --bins --tests, so examples aren't linked; doctests run
separately. The wasm check keeps going past the first failing crate. VP8
fixture generation locks across processes, since nextest runs each test in
its own.
* VP9 synthetic fixtures: write aside and rename
A test in another process could read a fixture ffmpeg was still writing.
* Fuzz tests: a third of each sweep by default, all of it with FILMCRAFT_FUZZ_ITERS
Every seed, mutation kind and thread count still runs. Header bytes each get
one overwrite value and two bit flips in rotation, so every value and bit
index still occurs across the region. FILMCRAFT_FUZZ_ITERS=N runs the whole
sweep plus N random rounds per seed, as before.
* Platform fallback tests: one failure point per hybrid path
The failure-anywhere test injected a failure at four points around every
restart point; the hybrid's paths differ only in the first GOP, around the
second and last restart points, at the end and never. The replay test needs
the first two GOPs.
* Export tests: shorter ranges with the same checks
The bitrate test measures 0.5–1.5 s, where each preset reaches its own
target (on 0–3 s the simple demo left two presets at the same rate). The
every-preset test exports three frames per preset.
* Audio tests: effects in parallel, a table-driven FFT for the Opus quality metric
The extreme-settings test keeps both settings and the 40 s tail for every
effect, one thread each. The Opus quality metric's FFT builds its twiddles
once and allocates nothing per transform (same algorithm and summation order).
* CI: dev-profile gate with nextest
xtask ci no longer builds with --release, so the release-profile overrides
go; nextest is installed for the test step.
* CI: apt packages through the cached install-apt-packages action
* Review fixes: fuzz tests with a panic hook in their own binaries, loud fixture lock
The fuzz tests of av1, h264, hevc, mpeg2v, vp8 and vp9 count panics through
a process-wide hook, so a panic in another test of the same process would
fail them; they are binaries of their own again. The VP8 fixture lock fails
the test instead of skipping it silently, and the full-sweep doc comments say
how many truncation points the full run samples.
* Fuzz tests: rotate the default subset per run (FILMCRAFT_FUZZ_SEED replays one)
* platform tests: decode-switch tests in their own binaries, one NVENC lock
Problem: camera RAW is at 0% (docs/codec-parity.md, #383). CinemaDNG,
the one cinema RAW format with a public specification, could not be
read at all.
Fix: new L0 crate filmcraft-dng, written from the public DNG 1.4-1.7,
TIFF 6.0 / TIFF-EP, ITU-T T.81 and CinemaDNG specifications only (no
other decoder's source read; editions in the crate README):
- TIFF container with SubIFDs, bounded (64 IFDs, 4 levels, cycles end
the walk) and checked offsets; the main raw IFD is the largest
NewSubFileType 0 CFA / LinearRaw image.
- Raw data: 1-16-bit uncompressed (MSB-first packing) and lossless JPEG
(process 14, predictors 1-7, 1-4 components, point transform, line
restarts), strips and tiles, tiles decoded in parallel; allocations
capped by size and by the data present.
- Processing per the DNG model: ActiveArea, LinearizationTable, per-site
black levels with deltas, white levels, as-shot white balance with
neutral highlight clipping, Malvar-He-Cutler Bayer demosaic (bilinear
for other patterns), DefaultCrop, ColorMatrix / CameraCalibration /
ForwardMatrix with illuminant interpolation, Bradford to linear
Rec. 709, BaselineExposure; output scene-linear RGBA floats.
- CinemaDNG FrameRate, TimeCodes and ReelName tags.
- A test-only writer (feature `writer`) synthesises DNGs. Tests: raw
samples bit-exact for every depth / predictor / layout, processing
against hand-computed values, hostile files (every truncation, 12k
mutations, cyclic IFDs, huge sizes, bad offsets), and LibRaw's
unprocessed_raw as an external oracle reading the same samples
(skipped when not installed).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
VP8 WebM showed black video: there was no VP8 decoder. New L0 crate
filmcraft-vp8, implemented from RFC 6386 sections 1-19 (tables extracted
from the RFC text, nothing from the attached reference decoder): every
coding tool, versions 0-3, hidden altref frames, size changes. Bit-exact
with libvpx on 62 libvpx VP8 test vectors, real-world FATE samples and
generated fixtures; never panics on corrupted input.
Wired into codecs for Matroska / WebM (V_VP8, including the alpha layer
of transparent WebM) and MP4 (vp08). VP8 colour is BT.601 (RFC 6386 9.2).
A hidden altref frame is a WebM block of its own with the time of the
frame that follows it: the VP8 time lookup now moves on to that shown
frame instead of repeating the previous one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Keep FLAC beside main's new AAC Audio format (21 formats, both audio-only
presets), use is_mp4_with / is_mp4_video in audio_codec() and the Audio
Format picker so MP4 (incl. AV1) offers AAC or FLAC, and add flacLevel to
main's export.media params.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Problem: `cargo xtask parity` (#781, #782) measured only the menu bar; the
Effects panel's effects and transitions had no measured parity figure
(docs/gaps.md G7).
Fix: `--area effects|transitions` compares the catalogue the engine's
effect definitions list in the Effects panel (new `effect_catalog`
example in filmcraft-project, the panel's own top-folder filter) with a
reference of folder paths or bare names (default the local, never
committed plan/premiere/<area>.json). Same report, caps and `--min` as
menus; placement by folder; Presets / Lumetri Presets and the other
area's items are out of scope, so one panel snapshot serves both areas.
`--area menus` (default) is unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Problem: symphonia rejects every Layer II frame whose bit rate / mode
combination ISO/IEC 11172-3 does not allow (mono above 192 kbit/s, stereo
below 64 kbit/s). ffmpeg writes mono MP2 at 384 kbit/s by default, so such
tracks played as silence in MPEG-TS / PS, Matroska and raw .mp2 files.
Fix: new L0 crate filmcraft-mpa, a clean-room Layer I / II decoder written
from ISO/IEC 11172-3 and 13818-3 (all modes incl. joint stereo, MPEG-1/2/2.5
rates, the disallowed combinations decoded with the allocation table their
bit rate per channel selects). PacketDecoder::mpeg_audio uses it for layers
I and II (packets of several frames too); raw .mp2 files decode with it,
resyncing after damage. Oracle tests vs ffmpeg: 73-84 dB SNR on ffmpeg /
twolame streams and synthetic Layer I / joint stereo frames; fuzzed damaged
frames never panic.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Problem: the menu parity figure in docs/target-app-parity.md came from an
ad hoc label-matching script over string literals; nothing repeatable
measured the real menu bar (docs/gaps.md G7).
Fix: `cargo xtask parity` dumps the shipped menu bar (new
`menus::default_menu_items`, deterministic: English, default prefs,
built-in workspaces; printed by `examples/menu_tree.rs`) and compares it
with a reference list of menu paths, by default the local, never-committed
`plan/premiere/menus.json` (any JSON of paths via --reference; parsing is
depth- and size-capped). Presence and placement are reported separately
(same path, elsewhere, missing); app/Help/recent/account/macOS items are
out of scope. Writes a markdown checklist (target/parity/ or --out) and
--min fails below a presence threshold. Without a reference it prints our
inventory only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Problem: the macOS bundle had no NSMicrophoneUsageDescription and empty
hardened-runtime entitlements, so macOS never asked for microphone access
and voice-over recording got no input.
Fix: declare NSMicrophoneUsageDescription in Info.plist.in and grant
com.apple.security.device.audio-input in entitlements.plist; an xtask test
guards both.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
No conflicts. main's new Chinese catalog requires every tl! string:
added zh-CN for the FLAC labels (format, preset, Audio Format entry,
Compression Level and its hint).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
New L0 crate written from the AVI RIFF file reference and OpenDML 1.02:
avih and per-stream strh / strf (BITMAPINFOHEADER, WAVEFORMATEX(TENSIBLE))
/ strn, and chunk tables from the best index a file has: the OpenDML indx
super index and ix## standard indexes (field indexes too), else idx1
(offsets relative to movi or absolute) plus a scan of RIFF AVIX
extensions it cannot see, else a scan of the movi lists (key frames then
unknown). Frames are read on demand through a ByteSource.
Sizes are clamped to the file, so a cut-off recording opens with what
is there; rec lists nest at most 3 deep and tables are bounded by what
the file could hold. Tests: hand-built files for every index path,
truncation, and every single-byte mutation and truncation (no panic);
ffmpeg-made files against ffprobe; an ignored test on a 1.16 GB OpenDML
file with AVIX extensions.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
New L0 crate filmcraft-flac, a FLAC encoder written from RFC 9639:
constant / verbatim / fixed / LPC subframes (Levinson-Durbin on Tukey-
windowed autocorrelation, up to order 12), partitioned Rice residuals,
left/side, side/right and mid/side stereo, CRC-8/CRC-16, STREAMINFO with
MD5 (RFC 1321, own implementation), levels 0-8, all lossless.
Export:
- Format::Flac, an audio-only .flac (16/24-bit), streamed one second at
a time like WAV/AIFF; STREAMINFO is rewritten at the end with frame
sizes, total samples and MD5. Preset "FLAC 48 kHz 24-bit".
- AudioCodec::Flac in MP4 and QuickTime (fLaC + dfLa), one frame per
sample with the short last frame timed exactly; the dfLa STREAMINFO
is replaced before the moov is written (Mp4Writer::set_sample_entry).
MXF stays PCM.
- AudioSettings::flac_level (default 5); Export panel: FLAC (lossless)
audio format and Compression Level; file.exportMedia `format: "flac"`,
`audioCodec`, `flacLevel`. Spanish and Japanese strings.
ffmpeg oracle: every stream decodes bit-exactly with the STREAMINFO MD5
equal to ffmpeg's hash; .flac and FLAC-in-MP4/MOV exports decode to the
same samples as a WAV export of the range.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- filmcraft-tts (feature kokoro): Kokoro-82M on candle, implemented from
the StyleTTS 2 / iSTFTNet / ALBERT papers, config.json and the
checkpoint's tensor shapes; official .pth read with candle's restricted
pickle reader; voice packs from their zip. Neural Voice: tts-text
phonemize -> Kokoro per chunk -> pauses -> pitch via audio-dsp.
- catalog: 9 US English voices (model card grade C+ or better) + CMUdict,
every file pinned to a revision and SHA-256 (336 MB).
- speech: generic download_files (same verified downloader).
- engine: tts.render (background synthesis into Session::tts_cache),
tts.downloadVoices (background job, cancellable), voices list with
package state; feature neural-voices (default on in the desktop app).
- ui: natural/basic voice groups, consent box (size, sources, licences),
download progress + Cancel, synthesis never blocks the window;
ui.set {tts:{...}} for agents.
Measured: Whisper-base WER on 50 original sentences 2.1-3.6 % for all 9
voices (built-in robotic voice 96.6 %); 2.2-2.3x real time on CPU.
Gates: fmt, clippy, layers, assets, wasm pass; tests pass except
shortcuts_tests::assign_reassigns_conflicts_and_undoes (fails on main)
and the frame pool test (flaky under load, passes alone).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
New L2 crate filmcraft-tts-text: normalize(script, lang) -> Vec<Token> for the
Text to Speech text front end (P2a). English only (EnUs/EnGb), per the amended
brief. Handles cardinals to 999 999 999 999, negatives, decimals, ordinals,
money ($, €, £), percent, years (1100-2099), clock times, numeric and
abbreviated dates, phone-like digit groups, Mr./Mrs./Dr./St./Mt./vs./etc./e.g./
i.e./a.m./p.m., spelled acronyms, &/+@#/, and [pause Ns|ms] markers capped at
10 s. Pure Rust, deps serde + thiserror only, wasm-clean.
Never crashes (AGENTS.md §0): >1 MiB is NormalizeError::TooLong, oversized and
leading-zero numbers are read digit by digit, malformed pause markers become
plain text, all slicing is get()/boundary-safe, and a catch_unwind suite covers
empty/whitespace/oversized/10k-digit/$$$$/mixed-script/emoji/RTL/combining/NUL
inputs plus determinism.
Fix round (Claude review): a '.' after a number now ends the sentence (year
rule and money/decimal follow-through), so "in 1999." reads as a year with a
SentenceEnd and "$1,234,567." gets its final SentenceEnd; new table cases for
sentence-final numbers and the year rule under punctuation (., ,, !, ()); the
remaining direct indexes in non-test code converted to get()/first()/last().
Gates: fmt, clippy -p filmcraft-tts-text --all-targets -D warnings, tests
(25 suites, ~140 table cases), cargo xtask layers, cargo xtask wasm all green.
cargo xtask ci not run this round (Claude runs it when merging); it stops on a
pre-existing filmcraft-engine shortcuts test failure unrelated to this crate.
🤖 Written by Merlin (Hermes Agent, profile merlin)
Write what a narrator says, pick a voice, pitch and pace, and add it to
the timeline. Selecting a narration clip loads it back into the panel;
Save regenerates it in place as one undo step, keeping the clip's start
and length (longer speech kept whole in the file, shorter speech padded
with silence).
- filmcraft-tts (L2): Voice trait, voice catalogue, pause markers
([pause 1s]), and an original formant synthesizer (Basic Female/Male;
robotic placeholder until the neural voices land).
- project: Narration + Project::narrations (schema v13, no-op migration).
- engine: tts.voices/preview/create/edit/inspect; narrations never
overwrite clips (free targeted track, free track, else a new track).
- ui: Text to Speech panel beside Properties, preview playback,
click-to-edit, automation ids.
Gates: fmt, clippy -D warnings, layers, assets and wasm (rustup
toolchain) pass. Tests: all pass except three that fail identically on
main (61238bf) on this machine: shortcuts_tests::assign_reassigns_
conflicts_and_undoes, gpu fx_tests::gpu_effects_match_cpu_exactly, and
the frame pool test, which passes when run alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Desktop/shared:
- Frame workers catch panics per job: a failing decode/render no longer kills the worker
or leaves its frame in flight forever (blank monitors). Counted as perf.stats failedJobs.
- UI pass catches panics and shows a "FilmCraft hit an error" window (Save Project /
Continue) instead of closing the app.
- Panic hook writes <data dir>/Logs/crash-<day>.log with thread, location and backtrace;
a window that can't start shows an error dialog instead of exiting silently.
- GPU compositor only on adapters that render/blend Rgba16Float with >= 4096 textures;
GPU validation errors and oversize frames switch to the CPU compositor instead of
wgpu's default panic.
- Dock layout and Media Browser splitter no longer panic on tiny panels/windows.
Web:
- Export mode / previews / voice-over / log reveal no longer call std::env::temp_dir
(panics on wasm32): new filmcraft_engine::temp_dir(); default export dir /exports.
- Fatal-error overlay with Reload; WebGPU device failure retries on WebGL2 (?webgl);
cache-busted wasm/glue URLs; webcodecs entries[0] guard.
Tests: robustness_ui (worker survives injected panics, UI panic recovery, tiny windows),
crash hook, dock split, GPU plan size, default export dir; web smoke test extended.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- assets/app-icon/: canonical SVG + light SVG, 1024 PNG, macOS runtime PNG, .icns, .ico,
hicolor theme, LICENSE.txt and README (palette, geometry, provenance, regeneration).
- packaging/icons.sh regenerates everything with resvg, iconutil and the new `cargo xtask ico`.
- Native app sets the window icon (with_icon) and the Wayland app id; build.rs embeds the .ico
and VERSIONINFO into filmcraft.exe on Windows (winresource, Windows-host build-dep only).
- packaging/linux/ai.storyteller.filmcraft.desktop with Icon=ai.storyteller.filmcraft.
- Web app: favicon.svg replaced by favicon.png from the same artwork; sidecars and
ATTRIBUTION.md updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Decode fps per codec (H.264, HEVC, VP9, AV1, ProRes at 1080p/2160p),
Program-monitor playback and scrubbing through the real frame scheduler,
timeline UI frame time with a 1000-clip / 20-track sequence under
egui_kittest (+ wgpu render), export throughput (H.264 + AAC, ProRes),
project save/open of 5000 clips, and peak RSS per section (each section
runs in its own process under /usr/bin/time). Writes JSON and Markdown
to target/bench. bench_playback now shares its code (examples/bench/).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- crates/speech (L2): Transcriber trait, FixedTranscriber, Whisper model catalogue (pinned,
SHA-256 verified) with optional downloader (feature `download`: ureq + rustls + RustCrypto),
pure-Rust Whisper inference on candle with word timestamps (feature `whisper`), log-mel, VAD
word tightening, MFCC speaker clustering. Both features off by default and for wasm.
- project: per-media-item transcripts (Project::transcripts); schema v9 with a no-op v8->v9
migration and a v8 fixture.
- edit::transcript: sequence transcript derived from clip transcripts, search, paragraphs,
word ranges, pause/filler detection, ripple delete of many ranges, caption layout.
- engine: transcript.* commands (generate, set, delete, inspect, search, select, extract, lift,
renameSpeaker, removeFillers, removePauses, createCaptions, models, downloadModel);
Session::transcriber for hosts/tests; features `whisper` / `speech-download` (also on the
desktop and CLI apps). Without whisper, transcript.generate fails with a clear error.
- ui: Text panel Transcript tab (Transcribe button, speaker paragraphs of clickable words,
Shift+click selection -> In/Out, extract/lift/remove fillers/create captions, search) with
text.transcript.* automation ids; headless UI test.
- docs/transcripts.md, crates/speech/README.md, architecture and project-files updates.
whisper-tiny measures 12.2% WER over 797 words on local samples.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- eframe web runner on WebGPU (WebGL2 fallback; GPU compositor on WebGPU, CPU on WebGL2).
- fs: virtual file table of File/Blob handles, 1 MiB chunked async range reads with an
LRU cache and read-ahead (pending + retry instead of blocking), downloads for written
files, /opfs/ paths written to OPFS.
- import: File System Access picker / <input type=file>, drag-and-drop anywhere,
container index prefetch (moov wherever it is), retried imports; project open/save.
- recovery: OPFS auto-save snapshot while dirty, media copies, reopened on next visit.
- audio: AudioWorklet output fed ahead from the UI thread; played-frame audio clock.
- webcodecs: H.264/HEVC/VP9/AV1 MP4/MOV decoded by the browser VideoDecoder via a
reader opener (fallback to our decoders on any error), probed with isConfigSupported.
- api: window.filmcraft (execute/request/commands/screenshot/importFiles/importUrl/
openProject/files/readFile/info).
- cargo xtask web [--dev] [--serve PORT] (wasm-bindgen CLI); xtask wasm also checks
filmcraft-ui-egui and filmcraft-web; layer L6 for the web app.
- Original web icon (favicon.svg) with attribution sidecar; docs/web.md; headless-Chrome
smoke test (tests/smoke.mjs).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
New L0 crate filmcraft-dnx, written from SMPTE ST 2019-1:2016 + Amd 1:2023
(tables extracted from the standard's text by tools/extract_tables.py).
Decoder: every compression ID (DNxHD 1235-1260 incl. field-coded and
macroblock-adaptive interlace, thin rasters, RGB 4:4:4; DNxHR 1270-1274
LB/SQ/HQ/HQX/444), 8/10/12-bit, 4:2:0/4:2:2/4:4:4, DCT and lossless alpha,
macroblock rows in parallel. Matches ffmpeg within IDCT precision (max 2 LSB,
<=50 per million beyond 1, 87-99% identical) on 22 ffmpeg fixtures; follows
ffmpeg where it departs from the standard's text (reconstruction offset,
CID 1271 weights, 4:4:4 signalling) - documented in the README.
Encoder: DNxHR LB/SQ/HQ/HQX(10/12-bit)/444, CBR at the eq. 7.1 frame sizes,
uniform-qsf binary search + per-macroblock refinement; ffmpeg decodes every
profile with -xerror (HQ luma PSNR 58.8-66.3 dB).
Wiring: MOV AVdn/AVdh import (codec label from the CID), export
Format::DnxHr (MOV AVdh, dnxProfile lb|sq|hq|hqx) in the export dialog and
file.exportMedia.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- New L1 crate filmcraft-text: fonts (bundled Inter Regular/Medium/SemiBold/Bold/Italic,
JetBrains Mono, Noto Serif; DirectorySource scans system font folders reading only name/OS2
tables; family/style resolution with synthetic bold/italic; per-char fallback), layout
(harfrust kerning/ligatures, unicode-bidi, unicode-linebreak, alignment incl. justify,
leading, tracking, baseline shift, all/small caps, underline, carets/hit-test/selection,
cached), raster (moved from captions; active-edge list, shapes, emboldening), render
(outline + sub-pixel glyph-mask caches; transformed-outline path for rotation/skew),
mask (distance-transform strokes, blur).
- Captions burn-in now shapes and kerns via the engine (font.rs/raster.rs removed).
- Timecode / Clip Name effects draw real text (full clip names, not seven-segment digits);
text_burnin golden re-blessed.
- UI takes its font bytes from the text crate (no duplicate copies in the binary).
- New fonts: Inter Bold, Inter Italic (OFL-1.1, rsms/inter v4.1), Noto Serif Regular
(OFL-1.1, notofonts) with sidecars, ATTRIBUTION.md rows and licence text.
- Perf: 3-line 96 px title at 1080p ~1.6 ms cold / 0.15 ms warm (release).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- crates/ui-egui/examples/bench_playback.rs drives the Program monitor's own FrameServer,
prefetch (now FrameServer::schedule_playback, shared with the monitor) and PlaybackMeter at a
simulated 60 Hz display, optionally compositing on a headless wgpu device. Scenarios:
h264-1080, stack3 (3 × 1080p, scaled/rotated/opacity), h264-2160, demo, after-preview
(live play → Render Effects In to Out → 2 plays), seek-storm. Reports latency percentiles,
on-time ratio, worker/decode wall and thread-CPU time, UI-thread present time, process CPU per
frame, decoder seeks/samples (codecs::gop_stats) and wasted jobs; --json for machine output.
Fixtures are generated with ffmpeg into target/fixtures/playback.
- FrameServer records per-job timings (thread CPU via rustix clocks) while profiling is on.
- Dropped/shown were counted per UI refresh: "shown" counted every refresh that had the frame
(322 "shown" in an 8 s play of 192 frames) and "dropped" counted a frame that arrived a few ms
after its first refresh. PlaybackMeter counts timeline frames: shown when the exact picture was
on screen while due, dropped otherwise, including frames passed over without a refresh.
Baseline (GPU path, 8 s, load avg 104–168 on 14 cores; CPU columns are load-robust):
h264-1080 Full 191/1 shown/dropped, 46 ms CPU/frame; Half 188/4
stack3 Full 189/3, 102 ms CPU/frame; Half 157/35 (3 extra decoder seeks)
h264-2160 Full 8/184, 313 ms CPU/frame, 10 seeks; demo Full 114/78, Half 191/1
after-preview: live 0/192; 1st play after render 124/67, 2nd 119/73 (UI-thread present
p50 21 ms / p95 45 ms); seek-storm: 0 of 40 seeks shown within 150 ms
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
crates/golden (test-only, L3): nine procedural 320x180 scenes (transform
and opacity, blend modes, Gaussian Blur, Lumetri basic, Crop, Cross
Dissolve / Dip to Black / Wipe mid-transition, timecode/clip-name text)
rendered on the CPU and compared with committed goldens (PSNR >= 45 dB,
max <= 12, p99 <= 2); FILMCRAFT_BLESS=1 regenerates them with
attribution sidecars. testkit::golden adds PNG I/O, diff metrics and
failure images. The same scenes are composited on the GPU and compared
with the CPU.
GPU parity exposed a real bug: GpuCompositor cached uploads by
pixel-buffer address without owning the buffer, so a new frame
allocated at a freed frame's address was drawn with a stale texture
(whole wrong frames, PSNR ~12 dB). The cache now keeps the buffer
alive; regression test upload_cache_keeps_buffers_alive.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
