How NOVA is tested

Prerequisites (Windows)

Rust (MSVC toolchain), Node 20+, pnpm, Visual Studio Build Tools, WebView2
FFmpeg with libx264, libx265, libvpx-vp9, libopus on PATH (fixture generation only)

bash scripts/fetch-libmpv.sh        # pinned libmpv → vendor/libmpv (hash-checked)
bash scripts/generate-fixtures.sh   # synthetic media → tests/fixtures/media
pnpm install

Suites

Command	What it proves
`cargo test -p nova-playback-api`	State reducer rules, stale-session filtering, input validation, JSON contract shape
`cargo test -p nova-playback-mpv`	Real libmpv (headless: `vo=null`, `ao=null`) against the fixture corpus: load, duration, tracks, pause/seek/resume, rapid seeks, EOF and replay, invalid/truncated media, missing files, rapid file switching, track selection, volume/mute/speed confirmation, stop, and engine create/destroy leak check
`cargo test -p nova-domain`	Queue rules: ordering, cursor after remove/move, ends
`cargo test -p nova-application`	Queue service against a fake backend: auto-advance once per own session, previous-restarts, validation
`cargo test -p nova-storage`	SQLite migrations, newer-schema refusal, corrupt file, identity (moved/copied/replaced), constraints; `resume_restart`: real libmpv + services + SQLite file across two simulated restarts
`cargo test -p nova-library`	Scanner on real folders + SQLite: rescan no-op, rename/move, deleted folder + recovery, unplugged root, cancel, interrupted scan, junction loop, ScanManager. Opt-in 20k benchmark: `cargo test -p nova-library --release -- --ignored --nocapture`
`pnpm --filter @nova/desktop test`	Frontend: snapshot ordering, formatting, shortcuts, virtual list, HomeView/LibraryView component tests (jsdom; IPC mocked at the client boundary)
`pnpm --filter @nova/desktop check`	TypeScript strict type check

Override the engine with NOVA_LIBMPV_PATH=....

Media compatibility harness

See docs/compat/README.md. The harness is opt-in (--ignored) because the corpus is about 650 MB and a run takes about 75 s.

Manual / visual verification (rendering gate)

Headless tests cannot prove pixels reach the screen or A/V sync. For that:

pnpm --filter @nova/desktop dev     # in one terminal (Vite)
NOVA_STATS=1 target/debug/nova-desktop.exe tests/fixtures/media/h264_aac_1080p30_30s.mp4

NOVA_STATS=1 logs engine-measured avsync, dropped frames, video output, GPU context and active hardware decoder every 2 s. NOVA_HWDEC=no forces software decoding.

The fixtures carry a sync pattern: a white bar flashes for the first 100 ms of every second while the audio beeps. Watch/listen for alignment.

Caution for automated GUI checks

Synthetic keyboard input goes to whatever window is in the foreground. Any GUI driver must verify NOVA is the foreground window before sending input or capturing the screen, and should not run while someone is using the machine.

Corpus

All media is synthesized by scripts/generate-fixtures.sh from FFmpeg lavfi sources (no third-party content), so it is freely distributable and reproducible. Generated files are git-ignored.

File	Purpose
`h264_aac_1080p30_30s.mp4`	Baseline H.264/AAC, timecode + sync pattern
`hevc_aac_1080p24_10s.mkv`	HEVC/Matroska
`vp9_opus_720p30_10s.webm`	VP9/Opus/WebM
`h264_aac_360p30_180s.mp4`	3-minute file for resume rules (≥ 60 s)
`multitrack_h264_2a_2s_10s.mkv`	2 audio (eng/fra, titled) + 2 subtitle tracks
`truncated_h264.mp4`	First 200 KB of the baseline (damaged file)
`not_media.mp4`	Text file with a media extension