fix: FLAC, Ogg and Opus tracks carry their MusicBrainz recording id from the tag: one reader in library gives the scanner, the backfills and the importer the same ids, and the next scan re-reads the library (#5637)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

internal/library/mbids.go

@@ -7,54 +7,57 @@ import (
 	"github.com/dhowden/tag/mbz"
 )
 
-// extractMBIDs reads MusicBrainz album and artist IDs from a tag.Metadata.
+// TagMBIDs is what a file's MusicBrainz Picard tags name. Empty means the
+// tag is absent.
+type TagMBIDs struct {
+	// Release is musicbrainz_albumid: one edition, what albums.mbid stores.
+	Release string
+	Artist  string
+	// ReleaseGroup is musicbrainz_releasegroupid. Wants name albums by
+	// release group, so re-acquisition and request completion key on it
+	// (#5241).
+	ReleaseGroup string
+	// Recording is what tracks.mbid stores and the ListenBrainz Labs
+	// similar-recordings API keys on.
+	Recording string
+}
+
+// ReadTagMBIDs reads a file's MusicBrainz IDs, for the scanner, the MBID
+// backfills and the importer alike.
 //
 // Delegates to dhowden/tag/mbz, which knows the per-format Raw() conventions:
-//   - Vorbis (FLAC/OGG) keys are lowercased by the parser.
+//   - Vorbis (FLAC/OGG/Opus) keys are lowercased by the parser.
 //   - ID3v2 (MP3) TXXX frames are stored under "TXXX"/"TXXX_N" with
 //     a *tag.Comm value whose Description holds the Picard tag name.
 //   - MP4 (M4A) freeform iTunes atoms are stored under their bare sub-name.
 //
-// Then sanitizes: dhowden's readTextWithDescrFrame (used for ID3v2 TXXX)
-// does NOT strip the trailing/embedded NUL bytes that ID3v2.4 uses as
-// frame terminator and multi-value separator — so the Text often comes
-// out as "uuid\x00" or "uuid1\x00uuid2". Postgres rejects NULs in text
-// columns (SQLSTATE 22021). We split on NUL and take the first non-empty
-// segment — Minstrel uses the primary release artist for MBCAA lookup,
-// so first-wins matches the original semantics.
-func extractMBIDs(m tag.Metadata) (albumMBID, artistMBID string) {
+// mbz maps keys by name alone, and Picard names the recording differently
+// in Vorbis: MUSICBRAINZ_TRACKID there is the recording (the release track
+// is MUSICBRAINZ_RELEASETRACKID), so mbz files it under mbz.Track, its name
+// for the release track. Read mbz.Recording alone and every FLAC has no
+// recording id (#5637).
+func ReadTagMBIDs(m tag.Metadata) TagMBIDs {
 	info := mbz.Extract(m)
-	return cleanMBID(info.Get(mbz.Album)), cleanMBID(info.Get(mbz.Artist))
-}
-
-// extractRecordingMBID reads the MusicBrainz *recording* ID — Picard's
-// musicbrainz_recordingid, surfaced by dhowden/tag as mbz.Recording
-// (tag display name "MusicBrainz Track Id"). This is the id the
-// ListenBrainz Labs similar-recordings API keys on and what tracks.mbid
-// stores.
-//
-// Deliberately NOT mbz.Track ("MusicBrainz Release Track Id"): that is
-// per-release-track, whereas similarity is per-recording. Kept separate
-// from extractMBIDs so its existing (album, artist) signature and unit
-// tests stay untouched.
-func extractRecordingMBID(m tag.Metadata) string {
-	return cleanMBID(mbz.Extract(m).Get(mbz.Recording))
-}
-
-// extractReleaseGroupMBID reads the MusicBrainz *release-group* ID —
-// Picard's musicbrainz_releasegroupid, surfaced by dhowden/tag as
-// mbz.ReleaseGroup. mbz.Album (what albums.mbid stores) is the release: one
-// edition. Wants name albums by release group, so re-acquisition and
-// request completion key on this one (#5241). Separate from extractMBIDs for
-// the same reason extractRecordingMBID is.
-func extractReleaseGroupMBID(m tag.Metadata) string {
-	return cleanMBID(mbz.Extract(m).Get(mbz.ReleaseGroup))
-}
-
-func cleanMBID(s string) string {
-	if s == "" {
-		return ""
+	ids := TagMBIDs{
+		Release:      cleanMBID(info.Get(mbz.Album)),
+		Artist:       cleanMBID(info.Get(mbz.Artist)),
+		ReleaseGroup: cleanMBID(info.Get(mbz.ReleaseGroup)),
+		Recording:    cleanMBID(info.Get(mbz.Recording)),
 	}
+	if ids.Recording == "" && m.Format() == tag.VORBIS {
+		ids.Recording = cleanMBID(info.Get(mbz.Track))
+	}
+	return ids
+}
+
+// cleanMBID takes the first non-empty NUL-separated value. dhowden's
+// readTextWithDescrFrame (used for ID3v2 TXXX) does NOT strip the
+// trailing/embedded NUL bytes that ID3v2.4 uses as frame terminator and
+// multi-value separator — so the Text often comes out as "uuid\x00" or
+// "uuid1\x00uuid2". Postgres rejects NULs in text columns (SQLSTATE 22021).
+// Minstrel uses the primary release artist for MBCAA lookup, so first-wins
+// matches the original semantics.
+func cleanMBID(s string) string {
 	for _, part := range strings.Split(s, "\x00") {
 		if t := strings.TrimSpace(part); t != "" {
 			return t