Extraction always used Matroska (.mka) because it takes essentially any
audio codec. That is right for most codecs but wrong for AAC: MP4 records
the AAC encoder priming (the ~1000 warm-up samples every AAC encoder emits)
in an edit list, and the edit list does not survive into Matroska. The
remuxed track then decodes ~23 ms longer than the source, with the priming
samples played at the head instead of discarded.
Measured on a 5s 640x480 fixture: source audio decodes to 962980 bytes of
PCM, the .mka to 967440 — 4460 bytes / ~23 ms extra, peaking at -66 dBFS.
Inaudible in practice, but it also puts the track fractionally out of step
with its own reported duration, for no reason.
Pick the container by codec instead: aac -> .m4a (keeps the edit list),
everything else -> .mka as before. If the preferred container refuses the
codec, retry into .mka before falling back to keeping the whole video. AAC
is worth the special case because mp4 / mov / m4v — what people actually
upload — almost always carry it.
Adds the strongest available test of the "lossless" claim: decode the audio
straight out of the source mp4, decode the stored extract, assert the PCM is
byte-for-byte equal. Forcing .mka fails it with exactly the 4460-byte delta.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
After extracting the audio track, uploadAudio assigned the record's `size`
from the new .mka BEFORE deleting the source video:
size = statSync(extracted).size;
rmSync(filePath, { force: true }); // can throw EBUSY/EPERM on Windows
filePath = extracted;
rmSync with force:true only swallows ENOENT — a briefly locked file (exactly
what the existing scheduleRetry machinery in this file exists to handle)
throws. The catch then discards the extract and keeps playing the original
container, which is correct, but `size` had already been overwritten with the
much smaller extracted size while the whole video stayed on disk. That makes
totalBytes() under-count and lets the upload directory grow past its quota.
Commit filePath and size together, only once the source is actually gone.
Adds a regression test that partially mocks node:fs to make rmSync throw for
the source .mp4 and asserts the persisted record (index.json — `size` is not
exposed through search()/toSong) still describes the retained file. With the
old ordering it records 27894 bytes for a 104544-byte file.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The embedded QQ Music API sidecar could bind a different port than the one
the client base URL (getQQMusicBaseUrl) targets. The upstream
@sansenjian/qq-music-api package derives its default port from
process.env.PORT (falling back to 3200) and, in some historical versions,
auto-started that server as an import side effect. When an old build listened
on 3300 while the client requested 3200 (issue #122), fetching the QQ login QR
failed with ECONNREFUSED on 127.0.0.1:3200, so the QR never showed and login /
cookie persistence silently broke.
Align process.env.PORT with the configured qqMusicApiPort for the duration of
the import (restoring the previous value afterwards so nothing else in the
process is affected), reuse an already-listening instance instead of racing a
second listen, and log the port actually bound (read from the socket) so any
mismatch is visible in the logs.
- src/music/api-server.ts: PORT alignment + reuse-on-auto-start + bound-port log
- src/music/api-server.test.ts: regression coverage that the sidecar follows
qqMusicPort (not an injected PORT) and restores PORT afterwards
- README.md: QQ login FAQ clarifies the sidecar and client share qqMusicApiPort
and points stale-latest-image users (who saw 3300) at re-pulling the image
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
config.spotify is a single process-wide object shared by every BotInstance, so
config.spotify.deviceName was identical for all bots. On the Rust (librespot)
backend each bot spawned `librespot --name <deviceName>` with no per-bot
uniqueness, registering two Connect devices with the same name under the one
shared account. findDeviceByName()/waitForDevice() match by name, so bot A's
transfer()+play() could drive bot B's librespot (misroute) and a bot could
report ready on seeing the OTHER bot's same-named device (false readiness).
Fix: derive a per-bot-unique Connect identity from the shared base name.
- controller.ts: new exported pure helper perBotDeviceName(base, instanceId?)
(`${base}-${instanceId}` when an id is given, else base). Add optional
instanceId to SpotifyControllerOptions; buildBackend() computes the effective
name once and passes the SAME value to both the Rust and go backends so
--name, findDeviceByName, and waitForDevice all key on one identity.
- instance.ts: pass instanceId: this.id into buildController(); add instanceId
to the spotifyControllerFactory param type (test seam).
The user-configured config.spotify.deviceName base is left untouched; the suffix
applies only to the backend/Connect identity. Behavior-preserving for callers
that pass no instanceId (base name used unchanged).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
rustPresent/goPresent and getBackendInfo used existsSync(findX()), which for a
bare PATH command name resolves against cwd, not $PATH — so a scoop/choco/cargo/
apt install was invisible and Spotify was gated off. Add a sync PATH-aware
resolveExecutable() + isLibrespotPresent()/isGoLibrespotPresent() in binary.ts
and route controller.ts and web/server.ts through them.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Implements the Stage-2 SpotifyAudioBackend over Rust librespot: spawns
librespot with --backend pipe (no --device => s16le/44100/2 on stdout, no
--passthrough), pipes stdout -> ffmpeg (44100->48000 s16le), waits for the
Connect device to register before emitting "ready", and controls playback
(transfer/play/pause/resume/seek) plus track-end/position/metadata via a
polled SpotifyConnectApi. child_process/connect/oauth/ffmpeg injected for
fully mocked, no-network unit tests (Windows-targeted; not e2e without Premium).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Wraps an injected axios instance with a live user Bearer token from
getToken(): getDevices/findDeviceByName, transfer/play/pause/resume/seek,
and getPlaybackState (null on 204). Read-only calls degrade gracefully;
mutating calls no-op when unauthorized. Fully unit-tested with a mocked
AxiosInstance (no network) — Windows-targeted, not e2e-testable (no Premium).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Stage 3 Task 2. PKCE (S256) authorize URL, code exchange, and refresh with
rotated-refresh-token persistence + invalid_grant store-clear. axios/http and
token store injected for fully mocked, network-free unit tests.
Corrections C3.2 (require the operator's own client_id; no librespot public
client / :5588 default; buildAuthorizeUrl throws + isAuthorized false without
it) and C3.7 (delete the state->verifier map entry in a finally on every
terminal path) applied.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Append isRustLibrespotSupported/pickLibrespotPath/findLibrespot/
checkLibrespotAvailable/resetLibrespotBinaryCache to binary.ts, mirroring
the go-librespot resolver. Supported on all platforms (pipe->stdout),
resolves librespot.exe on win32, caches only positive --version probes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Per-bot orchestrator: isAvailable() gates on config.enabled + platform +
binary presence; ensureStarted() starts the backend once (idempotent, retries
on failure); playTrack/pause/resume/seek/stop delegate; getPcmStream() proxies
the backend PCM; re-emits backend trackEnded/metadata. backendFactory injected
for tests (fake backend, no real binary/network).
Correction C3: the controller does not re-emit a raw "error" event (Node's
EventEmitter throws on an unhandled "error"); it logs the backend error and
marks itself not-ready so the next ensureStarted() relaunches. getPcmStream()
returns the backend's single persistent stream (no per-attach PassThrough).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Implements SpotifyAudioBackend over a go-librespot sidecar: start() mkfifos
the pipe, spawns the FIFO->48k s16le ffmpeg reader BEFORE go-librespot, writes
config.yml, polls the REST / until ready, then connects the WS event stream.
Maps not_playing/stopped -> trackEnded and metadata -> SpotifyNowPlaying;
play/pause/resume/seek delegate to the REST client. All child_process/fs/REST/WS
seams are injectable so the lifecycle is fully unit-tested without a real binary.
Correction C1: ffmpeg is resolved via getFfmpegCommand() (now exported from
src/audio/player.ts) so the bundled ffmpeg-static fallback is honored in Docker;
the ffmpeg command is overridable via deps.ffmpegCommand for tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
GoLibrespotRestClient wraps axios (injectable via deps.http) for
/player/play|pause|resume|stop|seek, GET /status, GET / ping; ping/getStatus
swallow errors to false/null, mutating ops reject. GoLibrespotEventClient
(EventEmitter) parses {type,data} /events frames and re-emits type with data,
reconnects on close with capped backoff, stop() tears down. TDD with a mock
AxiosInstance and a fake WebSocket (no real binary/network).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Mirror youtube.ts findYtDlp/checkYtDlpAvailable (bin/ then PATH,
cache-positive-only availability, reset test hook) and add
isGoLibrespotSupported() Linux gate for the Stage 2 audio backend.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds src/music/spotify/webapi.ts (client-credentials token, catalog
mappers, 429 retry) + tests, and threads the new "spotify" platform id
through the type unions in provider.ts and database.ts. Also widens the
downstream QueuedSong.platform union (audio/queue.ts) and the
getProviderFor parameter (bot/instance.ts) so tsc --noEmit stays clean;
these two are the necessary call-site fixes for the new union member.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
A loose `^2.2.10` range let npm pull a newer build of the QQ Music API. The
library went ESM in 2.3.x: an ESM-only 2.3.0/2.3.1 throws ERR_REQUIRE_ESM on
load, so the embedded server never binds port 3200 and every QQ request
(including /getQQLoginQr) fails downstream with ECONNREFUSED — the QR code never
appears and cookies look broken.
- Pin to `~2.4.0` (verified: loads via the bot's native ESM import, exposes the
Koa app, and every endpoint qq.ts calls returns the exact shapes it parses —
QR, recommend, lyric, play, playlist detail). Blocks the broken 2.3.0/2.3.1
and any future 2.5 migration. NOTE: 2.4.x requires Node >=20.17 (or >=22.9).
- Add describeQqApiStartupError(): on startup failure, log an actionable error
(ERR_REQUIRE_ESM → version-pin hint; engine mismatch → Node-upgrade hint)
instead of a generic warning, so this is obvious from the logs next time.
- README: troubleshooting entry for "QQ 二维码不弹 / 登录失败 / cookie 无法使用"
and the version/Node note in the dependency table.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Implements the NetEase-parity login features for the Kugou provider now that
QR login works:
- getDailyRecommendSongs (每日推荐), getUserPlaylists (我的歌单), and a real
getRecommendPlaylists (推荐歌单, was a stub) ported from the reference API.
- mapKugouSong now extracts cover art per endpoint (sizable_cover / cover /
trans_param.union_cover, resolving the {size} template) — Kugou songs had no
artwork before.
- Fix the playlist-song shape (combined "歌手 - 歌名" in `name`, mixsongid as the
audio id) so opened playlists show real titles instead of 未知歌曲.
- New defensive playlist mappers keyed on global_collection_id (the only id
getPlaylistSongs can open); dedup user playlists in case the list endpoint
ignores pagination; firstStr() so an empty-string field can't mask a real one.
Frontend wires Kugou as a third home-discovery source (Source type, store
caches/auth, availableSources, fetchHomeData, Home FM card + source tabs,
SourceTabs label, persisted-tab whitelist). SourceTabs now highlights the
fallback-corrected source so the active tab shows when a logged-out source was
persisted (newly possible with 3 sources).
Adds kugou.test.ts coverage for the new mappers, the cover/empty-string and
playlist-shape handling, and id openability.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds a self-contained Kugou provider (bilibili-style: direct API calls, no
embedded API server, no new npm dependency) plus full backend + WebUI wiring.
Provider (src/music/kugou.ts): search, song-url (with device registration),
lyrics (KRC decode), song detail, playlist, album, personal FM, QR login +
cookie persistence, and quality. Request signing / crypto / KRC decoding are
ported from the MIT-licensed MakcRe/KuGouMusicApi using Node's built-in
crypto and zlib (no third-party crypto packages).
Wiring: the "kugou" platform is threaded through the provider contract, queue,
play-history, bot instance/manager dispatch (getProviderFor + the -k command
flag), index/server composition, the music/player/auth routers (unified
/search/all, /quality, the platform coercions, QR login), the cookie store,
and the WebUI (search source tab + badge, SongCard badge, brand token, and a
Kugou QR/cookie login card in Settings).
Verified live during development: search, lyrics, and album playback resolve
correctly. NOT verifiable in CI (Kugou anti-bot blocks the build host's IP):
play-URL resolution, QR login, and VIP audio — these are built faithfully to
the reference and need end-to-end testing on a non-flagged IP / a Kugou
account. See the header comment in kugou.ts.
Includes src/music/kugou.test.ts (mappers + KRC→LRC). An adversarial review
pass fixed: pagination truncating on filtered counts, an ms/seconds duration
heuristic, dfid soft-fail caching, the /v5/url random-dfid fallback, the FM
body identity, and an unguarded nickname decode.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Uploaded local files were deleted whenever a track left the current slot,
with no check on whether the file was still needed — causing data loss in
several flows. Replace with reference-aware cleanup: a file is deleted only
once it has been played AND is no longer referenced by ANY bot's queue
(BotManager.getReferencedLocalSongIds wired into the provider via
setInUseResolver), with the sweep run AFTER each queue mutation.
Fixes:
- play-song replay no longer deletes the file it is about to play
- loop / repeat-all / prev no longer destroy uploads mid-cycle
- a shared upload queued on multiple bots is not deleted while still in use
- !play / play-playlist / play-album clean the whole replaced queue, and an
empty/failed playlist/album load keeps the previous queue + files intact
- bound disk use with an upload quota (evict oldest UNREFERENCED files)
- validate uploads by extension against the audio whitelist (never trust the
client Content-Type); the stored extension is always a known audio type
Deletion now unlinks the file FIRST and drops the record only on success,
with a bounded non-blocking retry for briefly-locked files (Windows/ffmpeg),
so a failed unlink never orphans a file or diverges index.json. The quota
never evicts the just-uploaded file, and long filenames keep their extension.
Adds src/music/local.test.ts covering the cleanup lifecycle, quota eviction,
and upload validation.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
xxmod