Adopts ListenBrainz-inspired patterns to speed up imports from ~24h to
under 30 minutes for 49k scrobbles.
Phase 1 - track_lookup cache table:
- New migration (000006) adds persistent entity lookup cache
- Maps normalized (artist, track, album) → (artist_id, album_id, track_id)
- SubmitListen fast path: cache hit skips 18 DB queries → 2 queries
- Cache populated after entity resolution, invalidated on merge/delete
- Benefits both live scrobbles and imports
Phase 2 - SaveListensBatch:
- New batch listen insert using pgx CopyFrom → temp table → INSERT ON CONFLICT
- Thousands of inserts per second vs one-at-a-time
Phase 3 - BulkSubmitter:
- Reusable import accelerator for all importers
- Pre-deduplicates scrobbles by (artist, track, album) in memory
- Worker pool (4 goroutines) for parallel entity creation on cache miss
- Batch listen insertion via SaveListensBatch
Phase 4 - Migrate importers:
- Maloja, Spotify, LastFM, ListenBrainz importers use BulkSubmitter
- Koito importer left as-is (already fast with pre-resolved IDs)
Phase 5 - Skip image lookups during import:
- GetArtistImage/GetAlbumImage calls fully skipped when SkipCacheImage=true
- Background tasks (FetchMissingArtistImages/FetchMissingAlbumImages) backfill
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When a listen arrives with no MBZ IDs and no album title (the common
multi-scrobbler/Last.fm case), search MusicBrainz by artist+track name
to resolve recording, release, and release group IDs. This unlocks
CoverArtArchive album art, proper album association, and duration data.
New file: internal/mbz/search.go
- SearchRecording() method with Lucene query escaping
- Confidence filter: case-insensitive exact match on title + artist credit
- Release selection: prefer Official status, then first available
- Uses existing rate-limited queue (1 req/sec)
Integration in catalog.go:
- Only triggers when RecordingMbzID, ReleaseMbzID, AND ReleaseTitle are
all missing — no impact on scrobbles that already have MBZ data
- Soft failure — search errors don't block the listen
- KOITO_DISABLE_MUSICBRAINZ handled automatically (MbzErrorCaller returns error)
Interface + mocks updated:
- SearchRecording added to MusicBrainzCaller interface
- MbzMockCaller: SearchResults map for test data
- MbzErrorCaller: returns error (existing pattern)
New tests:
- TestSubmitListen_SearchByName — mock search, verify album+duration resolved
- TestSubmitListen_SearchByNameNoMatch — verify graceful fallback
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Root cause of the panic at ~1,693 items: importers created
`&mbz.MusicBrainzClient{}` (nil requestQueue) instead of using the
engine's properly initialized client. When any code path called an MBZ
method, it panicked on the nil channel.
Changes:
- Pass engine's MBZ client to Maloja and Spotify importers
- Change MalojaTrack.Album to pointer type to handle null album JSON
- Continue on error instead of aborting the entire import
- Accept both Maloja export formats ("scrobbles" and "list" keys)
- Extract per-file import into importFile() with its own defer/recover
- Add progress logging every 500 items
Test fixtures:
- maloja_import_null_album_test.json (null album, valid album, empty artists)
- maloja_api_format_test.json (API "list" format)
New tests: TestImportMaloja_NullAlbum, TestImportMaloja_ApiFormat
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* add subtle gradient to home page
* tweak autumn theme primary color
* reduce home page top margin on mobile
* use focus-active instead of focus for outline
* fix gradient on rewind page
* align checkbox on login form
* i forgot what the pseudo class was called
* chore: call relay early to prevent missed relays
* fix: get current time in tz for listen activity (#146)
* fix: get current time in tz for listen activity
* fix: adjust test to prevent timezone errors
