Root cause of all Android player bugs: two competing queue managers.
ExoPlayer held all tracks and auto-advanced with its own index while JS
tracked S.qIdx separately. seekToIndex() is async via runOnMain, so
skips would interleave with ExoPlayer's own advances — causing desync,
wrong-song display, and skips that silently did nothing.
Fix: ExoPlayer always holds exactly ONE MediaItem. JS is the single
source of truth for queue order, shuffle, and repeat.
JS changes (player.js):
- playQueue/playCurrent/nextSong/prevSong/toggleShuffle: replace
setQueue(allTracks, idx) + seekToIndex with setQueue([single], 0)
- _pushQueueIfBridge: send only current track
- nativeInsertAt: no-op (radio songs land in S.queue; JS picks them up)
- fillRadioQueue: remove addToQueue native call
- onTrackChanged: simplified to just schedule scrobble + radio top-up
- onPlaybackStateChanged: remove index-reconciliation block
- onPositionUpdate: remove reconciliation tick (no queue drift possible)
- onQueueEnded: call onEnded() — JS drives what plays next
- toggleRepeat: setRepeatMode('one') only for repeat=one, else 'off'
- Add onNotificationNext / onNotificationPrev to NSPlayer
- Remove _skipping/_skipTimer/_reconTick/_setSkipping (all dead code)
Android changes:
- MusicService: add onSkipToNext/onSkipToPrevious to StateListener;
add MediaSession.Callback to intercept notification skip buttons and
delegate them to JS instead of letting ExoPlayer handle them
- MainActivity: implement new StateListener methods → callJs()
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Reads ffprobe metadata from every audio file and places it into:
{dest}/{AlbumArtist}/{Album}/{NN - Artist - Title.ext}
Exactly mirrors the sanitize(), folder and filename rules in
backend/lib/ytdlp.js. Dry-run by default; --move to execute, --clean
to remove empty source dirs afterwards.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- edit-metadata: use albumArtistSort (when provided) for artist folder naming
instead of auto-converted romajiAlbumArtist — ensures manual overrides sent
by the download pipeline also place files in the correct folder
- Add POST /api/fix-artist-sort endpoint: searches Navidrome for all songs
where artist or albumArtist matches the kanji, re-writes ARTISTSORT +
ALBUMARTISTSORT via ffmpeg (-map_metadata 0 preserves all other existing
tags), moves each file to the correct artist folder, cleans up empty old
dirs, triggers Navidrome rescan — returns a job ID for progress tracking
- addRomajiOverrideUI: after saving a new override, calls fix-artist-sort and
shows a job card with live progress
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Fix missing closing brace for `if (ni >= 0...)` in nextSong — the removed
`else { _nsBridge('next') }` left the inner if-block unclosed, causing
vite/rollup to reject player.js as invalid JS
- Apply romaji override map to iTunes (executeDownload) and Spotify
(executeSpotifyDownload) download payloads — sends artistSort/albumArtistSort/
titleSort/albumSort fields so the backend writes correct sort tags to the
downloaded file and can organize into the right artist folder
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds a maintained list in Settings where you can map kanji artist/album/title
strings to their correct romaji reading (e.g. 照井順政 → Terui Yoshimasa).
Overrides are applied last (highest priority) when the edit-metadata modal
opens, so saving the song writes the correct sort tags permanently to the file.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When reaching the last shuffled song, the UI changed but audio kept playing
the same song. Root cause: ExoPlayer still had its own shuffle mode on (set
in onReady), so its auto-advance and _nsBridge('next') picked random indices
that conflicted with the pre-shuffled JS array.
- setShuffleMode always false: we own the playback order via pre-shuffled
JS array, ExoPlayer must advance sequentially through it
- Remove _nsBridge('next') fallback: never let ExoPlayer decide the next
song autonomously. End of queue with no repeat = let ExoPlayer stop.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The pseudo-element added for the flicker fix had position:absolute;inset:0
with no pointer-events:none, so it sat over the entire overlay and consumed
every click — queue items, lyrics lines, buttons. Add pointer-events:none.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
getPlaylists returns stale/wrong songCount from Navidrome's index.
After rendering the grid, fire one getPlaylist request per playlist
in parallel and update each card's subtitle as responses arrive.
Counts are visually corrected within a second of the page loading.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
NP overlay flicker (stronger fix):
- Remove contain:layout style paint (causes layer issues on some Android WebView)
- Add ::before pseudo-element with solid background so no frame can be transparent
- Add -webkit-transform prefixes for both open and closed states
Shuffle fully random:
- playQueue no longer pins startIdx when shuffle is on
- All songs shuffled into a completely random order, S.qIdx=0 always
- toggleShuffle mid-queue still pins current song first (Spotify toggle behavior)
Playlist count wrong in overview:
- Navidrome's getPlaylists returns stale/wrong songCount in its index
- Add data-pl-id to each playlist card in loadPlaylists
- After loadPlaylistDetail gets the accurate entry count, backpatch the
matching overview card's subtitle so the grid shows the real song count
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace will-change:transform with translate3d(0,100%,0) / translate3d(0,0,0).
will-change triggers a layer-creation/destruction cycle on WebView — when the
player bar DOM updates during a skip, the compositor can briefly expose the
layer's clear back buffer, flashing the content below through the overlay.
translate3d pins the element to a stable GPU layer from the start.
Also add contain:layout style paint to isolate the overlay's paint context
from external repaints, and -webkit-backface-visibility for older WebView.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- playQueue: unified shuffle for both web and Android — pre-shuffles the
songs array with the tapped song pinned first, saves _origQueue for restore.
Passes S.qIdx (0 when shuffled) to NSBridge.setQueue instead of startIdx.
- _computeNextIdx: removed the random-pick shuffle branch entirely.
Since the queue is already shuffled, sequential traversal IS the shuffle order.
- toggleShuffle: removed !HAS_BRIDGE guard so queue reorder runs on Android too.
Re-sends reordered queue to ExoPlayer so native order stays in sync with JS.
Removed setShuffleMode call (we manage order ourselves via pre-shuffled array).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Remove await on fetchRomajiIndex() at startup so navigation is not blocked
by the romaji index network call (was the main cause of 30s Android startup)
- Add window._imgRetry() with 3 retries (1.5s/3s/4.5s backoff) for all card/
song/cover images that fail to load; falls back to emoji only after exhausting retries
- playQueue with HAS_BRIDGE + shuffle: pick a random startIdx so album/playlist
playback starts at a random track, not always index 0
- Reduce mobile genre grid limit from 12 to 6
- Fix "Global Suche" badge text: was "iTunes Store + Local", now "iTunes Store"
- deleteSong: always refresh loadPlaylists() unconditionally (not just when
page is active) with a 3s delay so song counts are current when navigating back
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Guard both snapshot-reconciliation paths (_skipping window can return the
old index before ExoPlayer commits the new one):
- onPlaybackStateChanged: skip the safety-net snapshot read while _skipping
- onPositionUpdate: reset _reconTick (defer recon) while _skipping
- _setSkipping: always reset _reconTick on every new skip/seek
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Five separate root causes fixed:
1. buildTitleKeywords (ytdlp.js): For Japanese titles \w strips all
kana/kanji leaving no keywords. Now uses the first whitespace-
delimited segment as a literal regex keyword so --match-filter
title ~= requires the YTM video to contain the Japanese text.
2. Tokenizer wait (ytdlp.js): If kuromoji hasn't finished loading when
a Japanese download starts, romajiForStorage returns the original
Japanese text unchanged — producing Japanese folder/filenames and
skipping sort tags. Now waits up to 30 s for the tokenizer.
3. titleOnlyMatch guard (metadata.js): iTunes titleOnlyMatch (no artist
check) was used as a fallback for all searches. For Japanese
titles/artists this matched wrong US songs with identical titles
(common in anime OSTs). Now titleOnlyMatch is skipped when the
search title or artist contains Japanese characters.
4. squid.js rawArtist: rawArtist was not in squid's signature and
never forwarded to embedMetadata, so Qobuz-downloaded J-Pop tracks
stored the iTunes-resolved English name as artist instead of the
Japanese original. Fixed by accepting and forwarding rawArtist.
5. squid.js romaji paths: Folder and filename used raw artist/title
with no romaji conversion, creating Japanese-named paths
inconsistent with the yt-dlp path. Now uses romajiForStorage same
as ytdlp.js.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When Navidrome has a Japanese path in its DB but the file was saved with
a romaji filename by the download pipeline, the existing fallback chain
compared the Japanese title stem against romaji filenames and always
failed. Added _findFileByStemOrRomaji which converts the stem via
romajiForStorage before the final recursive search, covering all three
endpoints: edit-metadata, delete-song, song-sorttags.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Call reconcileNativeState() on visibilitychange→visible so the player UI
updates immediately when the user opens the app after a background auto-advance,
before deferred evaluateJavascript callbacks have had a chance to fire
- After deleteSong() succeeds, refresh the active playlist view (grid or detail)
with a 2s delay to let Navidrome's background scan complete
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- /api/delete-song endpoint: resolves path via same fallback chain as
edit-metadata, validates file is inside music root + is an audio file,
deletes it, then prunes the album folder if empty and the artist folder
if that too becomes empty — no recursive rm, depth strictly limited to
two levels above the file so no accidental mass-deletes
- Safety: path.resolve() guard ensures no path-traversal outside music root;
rmdirSync (non-recursive) fails if dir is non-empty as extra safety net
- Frontend: delete button (red outline, trash icon) at bottom of edit mode
with confirmation dialog showing title + folder-cleanup warning
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Three improvements:
- openNowPlaying(): reads NSBridge.getState() (always-fresh @Volatile snapshot)
via ID search to reconcile S.qIdx before render; fixes NP showing old song
when deferred evaluateJavascript callbacks haven't fired yet after backgrounding
- onPositionUpdate(): periodic check (~every 2s) detects track mismatch via
snapshot ID; catches missed/delayed onTrackChanged when NP is open
- updatePlayerUI(): pre-warms next song's 1024px cover so auto-advance NP
swap is instant from browser cache instead of a 1-second load delay
- styles: add backface-visibility:hidden + isolation:isolate to .np-overlay
to reduce GPU composite flicker on Android WebView on manual skip
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Added a play history stack (_playHist) that records S.qIdx before each
forward advance (manual next, auto-advance via onTrackChanged). prevSong
now pops from this stack instead of blindly doing S.qIdx--, so pressing
back during shuffle always returns to the song that was actually playing
before, regardless of where shuffle jumped. History clears on new queue.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two fixes:
1. Remove snapshot read from openNowPlaying() — the MediaController snapshot
can show the old song at the old index (both still valid entries in the
queue), so the ID check does not detect staleness; after the onTrackChanged
fix S.qIdx is always authoritative so there is no need to read the snapshot
2. Call syncNowPlaying() explicitly in onTrackChanged so the NP overlay is
refreshed immediately when the song changes while it is already open
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
openNowPlaying() was blindly trusting the MediaController snapshot index,
which can lag one track behind ExoPlayer on auto-advance. This overwrote
the already-correct S.qIdx that onTrackChanged had just set, causing the
NP overlay to render the previous song while the mini-player showed the
right one. Now only accept the snapshot index when the song ID at that
position matches — stale snapshots are silently ignored.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace saturated blur filter with actual dominant-color extraction:
fetch cover as blob (avoids canvas CORS taint), draw on 24x24 canvas,
find most vivid pixel, darken to ~28% for a moody solid background.
Blurred bg layers hidden when mode is active. Transitions smoothly
between songs via CSS transition on background.
Renamed setting label to "Einfarbiger Player-Hintergrund".
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Now Playing: add "Dynamische Hintergrundfarbe" toggle in Theme settings;
when on, np-bg uses higher saturation/brightness so the cover color really
shows through the blurred background
- Settings modal: extend to bottom nav on mobile (was leaving ~74px dead space
for the player bar which the overlay already covers at z-index 200)
- Android: fix onTrackChanged always staying on the first song — the snapshot
from MediaController can lag behind ExoPlayer on auto-advance, causing the
stale currentIndex to overwrite the correct p.index; now only advance to
a higher live index (for rapid-skip catch-up) and use p.trackId directly
instead of the potentially-stale snapshot id for the queue-mismatch guard
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Handles "File not found" for songs with non-standard folder/artist names
(e.g. artist separator • or year-only album folders) by doing a deep
recursive title-stem search as a last resort after rebase and dir scan.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
/api/itunes/lookup was returning 400 for any request without artist+title,
including id-based lookups (?id=...&entity=album) used by the artist page.
Use next() to fall through to the /:endpoint proxy for those requests.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When iTunes enrichment returns a jingle/preview as the first result (e.g.
trackTimeMillis=9000), the yt-dlp duration filter becomes impossible
(duration > 30 & duration < 24), all filtered steps fail, and the final
unfiltered fallback downloads whatever YouTube returns first — often a
short clip.
Fix 1: spotify.js — only copy iTunes trackTimeMillis to durationMs when
it is > 30 seconds (genuine song, not a jingle or preview).
Fix 2: ytdlp.js — only enter the duration-filtered search branch when
durationMs > 30 000ms, so a corrupted short value silently falls back to
the unfiltered search path which still finds the correct full song.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When a file is renamed or moved to a new artist/album folder during
metadata edit, Navidrome assigns a new song ID, breaking all playlist
entries that referenced the old ID.
Before the edit: collect which playlists contain the song.
After rename + rescan: poll Navidrome until the new song appears,
then update each affected playlist (add new ID, remove old index).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When Navidrome has a stale index (e.g. file was renamed from
'01 - Idea 10.mp3' to '01 - Gibran Alcocer - Idea 10.mp3'),
_findFileInDir now falls back to the original file's directory
and matches by track-number prefix ('01 - ') so the renamed file
is found even without a Navidrome rescan.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Spotify enrichment only copies album/trackNum/duration from iTunes.
Now also fetch totalTracks, discNum, discTotal from iTunes for:
- alreadyEnriched playlist tracks
- single track downloads where album came from Spotify
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
HoneyandClaude Sonnet 4.6