chore(release): v0.4.0

Reviewed-on: #9

.gitea/workflows/build-and-publish.yaml

@@ -45,7 +45,8 @@ jobs:
 
       # The suite runs inside the image, against the interpreter that ships,
       # rather than against whatever the runner happens to provide. A failing
-      # test fails the build. Layers are shared with the push build below.
+      # test fails the build. The runtime stage below is built from the same
+      # daemon afterwards, so its layers are already in cache.
       - name: Run the test suite inside the image
         run: docker build --target test -t music-curator:test .
 
@@ -124,9 +125,6 @@ jobs:
           echo "release=${release}" >> "$GITHUB_OUTPUT"
           echo "Computed bump=${bump}, release=${release}, base=${base}"
 
-      - name: Set up Buildx
-        uses: docker/setup-buildx-action@d7f5e7f509e45cec5c76c4d5afdd7de93d0b3df5 # v4
-
       - name: Log in to the Gitea container registry
         if: github.event_name != 'pull_request'
         uses: docker/login-action@650006c6eb7dba73a995cc03b0b2d7f5ca915bee # v4
@@ -135,21 +133,33 @@ jobs:
           username: ${{ github.repository_owner }}
           password: ${{ secrets.PACKAGES_TOKEN }}
 
-      - name: Build and push
+      # Plain `docker build` rather than buildx. buildx boots its own buildkit
-        uses: docker/build-push-action@f9f3042f7e2789586610d6e8b85c8f03e5195baf # v7
+      # in a container with a cache of its own, so it shared nothing with the
-        with:
+      # test build above and rebuilt the image from the base image up -- two
-          context: .
+      # full builds per run. It earns that cost when building for several
-          # Without this the last stage in the Dockerfile -- the test stage --
+      # platforms; this only ever targets the amd64 NAS, so it does not.
-          # would be what gets published.
+      #
-          target: runtime
+      # `--target runtime` is a strict prefix of the test stage, so every layer
-          # The NAS is the only host this runs on. Building arm64 as well would
+      # is already in the daemon's cache and this resolves in seconds.
-          # mean emulating it under QEMU for no consumer.
+      - name: Build the runtime image
-          platforms: linux/amd64
+        run: |
-          push: ${{ github.event_name != 'pull_request' }}
+          set -euo pipefail
-          tags: ${{ steps.version.outputs.tags }}
+          tags=()
-          labels: |
+          while IFS= read -r tag; do
-            org.opencontainers.image.source=${{ github.server_url }}/${{ github.repository }}
+            [ -n "$tag" ] && tags+=(-t "$tag")
-            org.opencontainers.image.revision=${{ github.sha }}
+          done <<< "${{ steps.version.outputs.tags }}"
+          docker build --target runtime \
+            --label "org.opencontainers.image.source=${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}" \
+            --label "org.opencontainers.image.revision=${GITHUB_SHA}" \
+            "${tags[@]}" .
+
+      - name: Push
+        if: github.event_name != 'pull_request'
+        run: |
+          set -euo pipefail
+          while IFS= read -r tag; do
+            [ -n "$tag" ] && docker push "$tag"
+          done <<< "${{ steps.version.outputs.tags }}"
 
       # Record the release: write the computed version into pyproject.toml, then
       # commit and tag it, so the packaging metadata always matches the release

README.md

@@ -7,9 +7,10 @@ which keeps an MP3 copy of a lossless library for an iPod. This one answers the
 question that mirror cannot: which of it is worth carrying, and which of it has
 not been played in years.
 
-**This is stage two.** It ingests the scrobble history, indexes the library from
+**This is stage three.** It ingests the scrobble history, indexes the library
-Lidarr, and matches one to the other. There are no playlists yet, and it writes
+from Lidarr, matches one to the other, and writes playlists into the mirror.
-nothing back — every Lidarr call is a `GET`. See "Where this is going" below.
+Nothing is written back to Lidarr — every call there is a `GET`. See "Where this
+is going" below.
 
 ## What it does today
 
@@ -18,6 +19,7 @@ nothing back — every Lidarr call is a `GET`. See "Where this is going" below.
 - Indexes every artist, album and track Lidarr knows about, with file paths and
   the date each file landed.
 - Ties the two together and reports how well it managed.
+- Writes M3U playlists into the mirror, from the listening history.
 
 ## Matching
 
@@ -29,7 +31,14 @@ Two tiers, and no third.
 | `name` | Normalised artist and title | Everything the first tier could not carry |
 | `none` | — | Recorded as a miss, never guessed at |
 
-The normalisation is the load-bearing part, because the two sides disagree in
+The name tier does almost all of the work. A recording MBID is exact when it
+lands, but MusicBrainz holds a separate recording per release, and Last.fm and
+Lidarr rarely pick the same one: on a real library, two thirds of scrobbles
+carry a recording id and barely a twentieth of them join on it. The report
+counts how many carried an id and matched on name anyway, which is the measure
+of that disagreement.
+
+So the normalisation is the load-bearing part, because the two sides disagree in
 predictable ways. It folds case and accents, drops guest credits (`Yellowcard
 feat. Tay Jardine` against a tag of `Yellowcard`), strips a trailing
 version suffix (`(Remastered 2011)`, `- Live`), expands `&`, and removes a
@@ -102,10 +111,66 @@ matcher. The line to watch is:
 unmatched by an artist the library holds: N pairs, M plays
 ```
 
-That is a track that was played, sitting beside a file it should have matched.
+That is a track that was played by an artist the library holds. It is then
-Those are the matcher's real misses, and every one is a candidate for being
+split three ways, because owning an artist is a weak proxy for owning a track
-wrongly called cold in stage four. The report lists the worst fifteen by play
+and a shared title is a weak proxy for a shared song:
-count so they can be eyeballed.
+
+- **the library's own title credits the scrobbled artist** — `Voodoo People
+  (Pendulum Remix)` against a play credited to Pendulum. Same song, filed
+  under the original artist. These are the genuine misses.
+- **the same title under an unrelated artist** — a collision, not a miss.
+  Across fifty thousand tracks these are constant: `Everyday` is Rusko and
+  also Def Leppard, `Kaleidoscope` is Delta Heavy and also Chappell Roan.
+  Matching on title alone would be far worse than missing them, which is why
+  there is no such tier.
+- **the title is nowhere in the library** — never bought.
+
+Only the first is worth chasing. Counting all three as matcher failures
+overstates the problem and would over-block the cull.
+
+## Playlists
+
+Written into `<mirror>/_playlists/` as extended M3U, rebuilt every pass. Six
+rules, capped at `--playlist-limit` tracks each:
+
+| Playlist | Rule |
+| -------------------- | --------------------------------------------------------- |
+| `heavy-rotation` | Most played over the last twelve months |
+| `all-time` | Most played ever |
+| `neglected` | Played heavily once, silent for twelve months |
+| `deep-cuts` | Never played, from albums whose other tracks you play constantly |
+| `unheard-favourites` | Never played, by the artists you play most |
+| `unheard` | Never played, anywhere in the library |
+
+Ninety days was the obvious window for "recent" and is the wrong one: on a real
+history it holds a few hundred plays spread thinly across a twenty-thousand
+track rotation, so nothing ranks meaningfully. Twelve months does.
+
+The two `unheard` playlists rotate **weekly**, not per pass. A pass runs every
+few hours, and a playlist that reorders itself each time is one that has to be
+re-imported each time — the Music app imports a snapshot of a file, it does not
+track it.
+
+### Paths
+
+Lidarr knows where the lossless source is; the playlists have to point at the
+MP3s music-mirror made from it. The mapping strips a library root from Lidarr's
+track paths and re-roots them under the mirror, with the suffix changed.
+
+`--library-root` is derived from the common parent of the indexed artist folders
+when unset, so it agrees with Lidarr by construction rather than by being kept
+in step by hand. Override it if that guess is wrong.
+
+Entries are written **relative to the playlist file**, so one playlist works
+from the NAS, from a Mac over SMB, and from Linux, without rewriting.
+
+A track is only listed once its mirror file has been confirmed to exist. Lidarr
+holding the FLAC says nothing about whether the MP3 has been encoded yet. If a
+large number are missing, the run says so — that is what a wrong `--library-root`
+or `--mirror` looks like, since the paths then map to nothing at all.
+
+`_playlists/` survives music-mirror's prune: it only deletes `*.mp3`, and its
+empty-directory sweep skips a directory holding M3Us.
 
 ## How the ingest works
 
@@ -161,6 +226,9 @@ music-curator --report-only   # report on the store, fetch nothing
 | `--backfill-limit` | `MUSIC_CURATOR_BACKFILL_LIMIT` | `0` | Cap backfill requests per pass; 0 for no cap |
 | `--lidarr-url` | `MUSIC_CURATOR_LIDARR_URL` | unset | Lidarr base URL, e.g. `http://lidarr:8686` |
 | `--lidarr-api-key` | `MUSIC_CURATOR_LIDARR_API_KEY` | unset | Lidarr API key |
+| `--mirror` | `MUSIC_CURATOR_MIRROR` | unset | Root of the MP3 mirror; playlists go here |
+| `--library-root` | `MUSIC_CURATOR_LIBRARY_ROOT` | derived | Prefix to strip from Lidarr's paths |
+| `--playlist-limit` | `MUSIC_CURATOR_PLAYLIST_LIMIT` | `100` | Most tracks in any one playlist |
 | `--skip-index` | — | off | Match against the index already held |
 | `--report-only` | — | off | Report without fetching |
 
@@ -209,7 +277,8 @@ nix shell nixpkgs#python3Packages.pytest -c pytest
 | ------------------------------------------------ | ------------ |
 | Last.fm ingest and store | done |
 | Lidarr index and the scrobble-to-track matcher | done |
-| M3U playlists written into the mirror | next |
+| M3U playlists from the listening history | done |
+| Genre and mood playlists from Last.fm tags | next |
 | Cold-music report, unmonitoring what is not played | last |
 
 The cull will unmonitor cold albums in Lidarr and tag their artists. It will

compose.yaml

@@ -30,5 +30,11 @@ services:
       # Cap the backfill at this many requests per pass. Unlimited by default,
       # which finishes a long history in one go.
       # MUSIC_CURATOR_BACKFILL_LIMIT: "0"
+      # The MP3 mirror music-mirror maintains. Playlists are written into
+      # _playlists/ inside it; leave unset to skip them.
+      MUSIC_CURATOR_MIRROR: /mirror
+      # Most tracks in any one playlist.
+      # MUSIC_CURATOR_PLAYLIST_LIMIT: "100"
     volumes:
       - /mnt/tank/apps/music-curator:/data
+      - /mnt/tank/media/music-mp3:/mirror

music_curator.py

@@ -26,6 +26,7 @@ import re
 import signal
 import sqlite3
 import sys
+import tempfile
 import time
 import unicodedata
 import urllib.error
@@ -62,6 +63,16 @@ REQUEST_DELAY_SECONDS = 0.25
 BACKOFF_SECONDS = 2.0
 BACKOFF_CEILING_SECONDS = 60.0
 
+# What music-mirror names its output, and where playlists are written inside the
+# mirror. music-mirror's prune only deletes `*.mp3` and only removes directories
+# it finds empty, so a directory of M3Us survives it untouched.
+MIRROR_SUFFIX = ".mp3"
+PLAYLIST_DIRECTORY = "_playlists"
+
+# Group-readable, for the same reason music-mirror sets it: the mirror is read
+# back by whatever serves it, and a playlist nobody can read is not a playlist.
+GROUP_READ = 0o040
+
 SCHEMA_VERSION = "2"
 
 # Versions this build upgrades in place. Everything added since version 1 is a
@@ -141,6 +152,11 @@ CREATE TABLE IF NOT EXISTS lidarr_track (
 );
 CREATE INDEX IF NOT EXISTS lidarr_track_recording ON lidarr_track (recording_mbid);
 CREATE INDEX IF NOT EXISTS lidarr_track_norm ON lidarr_track (norm_artist, norm_title);
+-- Separate from the composite above, which a lookup by title alone cannot use:
+-- its leading column is the artist. The report searches by title on its own, and
+-- without this it scans every track for every unmatched key -- forty-seven
+-- seconds on a library of eighty-four thousand.
+CREATE INDEX IF NOT EXISTS lidarr_track_title ON lidarr_track (norm_title);
 CREATE INDEX IF NOT EXISTS lidarr_track_album ON lidarr_track (album_id);
 
 -- One row per distinct thing listened to, with the verdict on whether it could
@@ -187,18 +203,28 @@ VERSION_WORDS = (
     "anniversary",
     "reissue",
     "instrumental",
+    # Drum and bass and its neighbours mark versions their own way.
+    "vip",
+    "bootleg",
+    "rework",
+    "extended",
 )
 _VERSIONS = "|".join(VERSION_WORDS)
 BRACKETED_VERSION = re.compile(
     rf"\s*[\(\[][^\)\]]*\b(?:{_VERSIONS})\b[^\)\]]*[\)\]]\s*$", re.IGNORECASE
 )
-TRAILING_VERSION = re.compile(rf"\s+-\s+[^-]*\b(?:{_VERSIONS})\b.*$", re.IGNORECASE)
+# The suffix is matched lazily rather than as a run of non-hyphens, because the
+# thing being stripped frequently contains hyphens of its own -- "Gold Dust -
+# Shy FX Re-Edit", "Back To Your Roots - Friction & K-Tee Remix".
+TRAILING_VERSION = re.compile(rf"\s+-\s+.*?\b(?:{_VERSIONS})\b.*$", re.IGNORECASE)
 
-# Last.fm routinely carries the guest credit in the artist field where the file
+# Last.fm routinely carries the guest credit where the file tag holds only the
-# tag holds only the primary artist -- "Yellowcard feat. Tay Jardine" against a
+# primary artist -- "Yellowcard feat. Tay Jardine" against a tag of
-# tag of "Yellowcard". `with` is deliberately absent: it appears in far too many
+# "Yellowcard", or "Self vs Self (feat. In Flames)" against "Self vs Self". The
-# real titles to cut on sight.
+# opening bracket has to be allowed for: requiring whitespace immediately before
-GUEST_CREDIT = re.compile(r"\s+(?:feat|ft|featuring)\b.*$", re.IGNORECASE)
+# the word misses every bracketed credit, which is most of them. `with` is
+# deliberately absent -- it appears in far too many real titles to cut on sight.
+GUEST_CREDIT = re.compile(r"[\s(\[]+(?:feat|ft|featuring)\b.*$", re.IGNORECASE)
 
 LEADING_ARTICLE = re.compile(r"^the\s+")
 # Deleted rather than spaced, so "Don't" and "Dont" agree. Every other mark
@@ -1031,6 +1057,225 @@ def format_time(uts):
     return datetime.fromtimestamp(uts, tz=timezone.utc).strftime("%Y-%m-%d %H:%M")
 
 
+# Playlists written into the mirror. Each is a rule over the listening history
+# and the library index; none of them look at the audio.
+#
+# The rotating ones are shuffled by week rather than by pass. A pass runs every
+# few hours, and a playlist that reorders itself every time is one that has to
+# be re-imported every time -- the Music app does not track a file, it imports a
+# snapshot of one.
+ROTATION_PERIOD_SECONDS = 7 * 86400
+SHUFFLE_MULTIPLIER = 2654435761
+SHUFFLE_MODULUS = 104729
+
+# Common table expressions the rules share. `played` is every library track that
+# has ever been matched to a scrobble, with its count and the last time it was
+# heard; `recent` is the same restricted to a window.
+PLAYED_CTE = """
+WITH played AS (
+    SELECT sk.track_id AS track_id, COUNT(*) AS plays, MAX(s.uts) AS last_uts
+      FROM scrobble s
+      JOIN scrobble_key sk ON sk.artist = s.artist AND sk.track = s.track
+     WHERE sk.track_id IS NOT NULL
+     GROUP BY sk.track_id
+)
+"""
+RECENT_CTE = """
+WITH recent AS (
+    SELECT sk.track_id AS track_id, COUNT(*) AS plays
+      FROM scrobble s
+      JOIN scrobble_key sk ON sk.artist = s.artist AND sk.track = s.track
+     WHERE sk.track_id IS NOT NULL AND s.uts >= :year_ago
+     GROUP BY sk.track_id
+)
+"""
+SELECT_TRACK = """
+SELECT t.id AS id, a.name AS artist, t.title AS title,
+       t.duration AS duration, t.path AS path
+  FROM lidarr_track t
+  JOIN lidarr_artist a ON a.id = t.artist_id
+"""
+PLAYABLE = " t.has_file = 1 AND t.path IS NOT NULL AND t.path <> '' "
+SHUFFLE = f" ((t.id * {SHUFFLE_MULTIPLIER}) + :week) % {SHUFFLE_MODULUS} "
+
+PLAYLISTS = (
+    (
+        "heavy-rotation",
+        "most played over the last twelve months",
+        RECENT_CTE + SELECT_TRACK + f"""
+          JOIN recent r ON r.track_id = t.id
+         WHERE {PLAYABLE}
+         ORDER BY r.plays DESC, a.name, t.title
+         LIMIT :limit
+        """,
+    ),
+    (
+        "all-time",
+        "most played ever",
+        PLAYED_CTE + SELECT_TRACK + f"""
+          JOIN played p ON p.track_id = t.id
+         WHERE {PLAYABLE}
+         ORDER BY p.plays DESC, a.name, t.title
+         LIMIT :limit
+        """,
+    ),
+    (
+        "neglected",
+        "played heavily once, silent for twelve months",
+        PLAYED_CTE + SELECT_TRACK + f"""
+          JOIN played p ON p.track_id = t.id
+         WHERE {PLAYABLE} AND p.last_uts < :year_ago
+         ORDER BY p.plays DESC, a.name, t.title
+         LIMIT :limit
+        """,
+    ),
+    (
+        "deep-cuts",
+        "never played, from albums whose other tracks you play constantly",
+        PLAYED_CTE + """,
+        album_plays AS (
+            SELECT t.album_id AS album_id, SUM(p.plays) AS plays
+              FROM played p JOIN lidarr_track t ON t.id = p.track_id
+             GROUP BY t.album_id
+        )
+        """ + SELECT_TRACK + f"""
+          JOIN album_plays ap ON ap.album_id = t.album_id
+          LEFT JOIN played p ON p.track_id = t.id
+         WHERE {PLAYABLE} AND p.track_id IS NULL
+         ORDER BY ap.plays DESC, t.id
+         LIMIT :limit
+        """,
+    ),
+    (
+        "unheard-favourites",
+        "never played, by the artists you play most; rotates weekly",
+        PLAYED_CTE + """,
+        artist_plays AS (
+            SELECT t.artist_id AS artist_id, SUM(p.plays) AS plays
+              FROM played p JOIN lidarr_track t ON t.id = p.track_id
+             GROUP BY t.artist_id
+        )
+        """ + SELECT_TRACK + f"""
+          LEFT JOIN played p ON p.track_id = t.id
+         WHERE {PLAYABLE} AND p.track_id IS NULL
+           AND t.artist_id IN (SELECT artist_id FROM artist_plays
+                                ORDER BY plays DESC LIMIT 50)
+         ORDER BY {SHUFFLE}
+         LIMIT :limit
+        """,
+    ),
+    (
+        "unheard",
+        "never played, anywhere in the library; rotates weekly",
+        PLAYED_CTE + SELECT_TRACK + f"""
+          LEFT JOIN played p ON p.track_id = t.id
+         WHERE {PLAYABLE} AND p.track_id IS NULL
+         ORDER BY {SHUFFLE}
+         LIMIT :limit
+        """,
+    ),
+)
+
+
+def library_root_of(store):
+    """Return the directory Lidarr's artist folders sit under.
+
+    Derived rather than configured, because it has to agree with what Lidarr
+    reports and no one wants to keep a second copy of that in step by hand.
+    """
+    paths = [
+        row["path"]
+        for row in store.connection.execute(
+            "SELECT path FROM lidarr_artist WHERE path IS NOT NULL AND path <> ''"
+        )
+    ]
+    if not paths:
+        return None
+    if len(paths) == 1:
+        return str(Path(paths[0]).parent)
+    try:
+        return os.path.commonpath(paths)
+    except ValueError:
+        return None
+
+
+def mirror_path_for(source, library_root, mirror_root):
+    """Return where music-mirror would have put the MP3 for a source file."""
+    try:
+        relative = Path(source).relative_to(library_root)
+    except ValueError:
+        return None
+    return (Path(mirror_root) / relative).with_suffix(MIRROR_SUFFIX)
+
+
+def write_playlist(path, entries):
+    """Write one extended M3U, atomically.
+
+    Paths are relative to the playlist file, so the same playlist works from the
+    NAS, from a Mac over SMB and from Linux without rewriting.
+    """
+    lines = ["#EXTM3U"]
+    for entry in entries:
+        seconds = round((entry["duration"] or 0) / 1000)
+        lines.append(f"#EXTINF:{seconds},{entry['artist']} - {entry['title']}")
+        lines.append(os.path.relpath(entry["mirror"], path.parent))
+
+    path.parent.mkdir(parents=True, exist_ok=True)
+    handle, temporary = tempfile.mkstemp(dir=path.parent, suffix=".m3u.part")
+    os.close(handle)
+    temporary = Path(temporary)
+    try:
+        temporary.write_text("\n".join(lines) + "\n", encoding="utf-8")
+        # The mirror is read back by something else; see music-mirror, which had
+        # to learn this the hard way.
+        mode = temporary.stat().st_mode
+        if not mode & GROUP_READ:
+            temporary.chmod(mode | GROUP_READ)
+        os.replace(temporary, path)
+    finally:
+        temporary.unlink(missing_ok=True)
+
+
+def build_playlists(store, mirror_root, library_root, limit, now):
+    """Write every playlist into the mirror. Returns how many tracks were listed.
+
+    A track is only listed once its mirror file has been confirmed to exist. The
+    index knows what Lidarr holds, which is the lossless source; whether the MP3
+    beside it has been encoded yet is music-mirror's business and is checked
+    rather than assumed.
+    """
+    directory = Path(mirror_root) / PLAYLIST_DIRECTORY
+    parameters = {
+        "limit": limit,
+        "year_ago": now - 365 * 86400,
+        "week": now // ROTATION_PERIOD_SECONDS,
+    }
+
+    total = 0
+    missing = 0
+    for name, description, sql in PLAYLISTS:
+        entries = []
+        for row in store.connection.execute(sql, parameters):
+            mirror = mirror_path_for(row["path"], library_root, mirror_root)
+            if mirror is None or not mirror.is_file():
+                missing += 1
+                continue
+            entries.append({**dict(row), "mirror": mirror})
+        write_playlist(directory / f"{name}.m3u", entries)
+        total += len(entries)
+        logger.info("playlist %-20s %4d tracks -- %s", name, len(entries), description)
+
+    if missing:
+        logger.warning(
+            "%d selected tracks had no file in the mirror and were left out."
+            " A handful means music-mirror has not encoded them yet; a large"
+            " number means --library-root or --mirror is pointing at the wrong"
+            " place, since the paths are then being mapped to nothing.",
+            missing,
+        )
+    return total
+
+
 def coverage_report(store):
     """Log how much of the listening history could be tied to the library.
 
@@ -1078,18 +1323,107 @@ def coverage_report(store):
             row["plays"],
         )
 
+    # Why the mbid tier performs the way it does. A recording id on both sides
+    # that still fails to join means the two disagree about which recording the
+    # song is -- MusicBrainz holds a separate recording per release, and Last.fm
+    # and Lidarr need not have picked the same one. That is not a fault to fix
+    # in the matcher; it is the reason the name tier has to carry the load.
+    library_with_mbid = store.scalar(
+        "SELECT COUNT(*) FROM lidarr_track WHERE recording_mbid IS NOT NULL"
+    )
+    logger.info(
+        "library tracks carrying a recording MBID: %d of %d (%.1f%%)",
+        library_with_mbid,
+        tracks,
+        100 * library_with_mbid / tracks,
+    )
+    disagreed = store.connection.execute(
+        "SELECT COUNT(*) AS pairs, COALESCE(SUM(plays), 0) AS plays FROM scrobble_key"
+        " WHERE track_mbid IS NOT NULL AND method = 'name'"
+    ).fetchone()
+    logger.info(
+        "carried a recording MBID, joined on name instead: %d pairs, %d plays"
+        " -- both sides know the song, they disagree on which recording it is",
+        disagreed["pairs"],
+        disagreed["plays"],
+    )
+
     suspect = store.connection.execute(
         "SELECT COUNT(*) AS pairs, COALESCE(SUM(plays), 0) AS plays FROM scrobble_key k"
         " WHERE k.track_id IS NULL"
         "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
     ).fetchone()
     logger.info(
-        "unmatched by an artist the library holds: %d pairs, %d plays -- these are the"
+        "unmatched by an artist the library holds: %d pairs, %d plays",
-        " matcher's misses, not music you do not own",
         suspect["pairs"],
         suspect["plays"],
     )
 
+    # Owning an artist is a weak proxy for owning a track, so that figure alone
+    # overstates the matcher's failings. Split it -- but not on the title alone.
+    # Across fifty thousand tracks, titles collide constantly: "Everyday" is
+    # Rusko and also Def Leppard, "Kaleidoscope" is Delta Heavy and also
+    # Chappell Roan. Matching those would be worse than missing them.
+    #
+    # The signal for a genuine attribution miss is that the library's own title
+    # credits the artist the scrobble is filed under -- "Voodoo People (Pendulum
+    # Remix)" against a play credited to Pendulum. The normalised title has that
+    # suffix stripped, which is exactly what let them meet, so the raw one has to
+    # be searched for the name.
+    attribution = store.connection.execute(
+        "SELECT COUNT(*) AS pairs, COALESCE(SUM(plays), 0) AS plays FROM scrobble_key k"
+        " WHERE k.track_id IS NULL"
+        "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
+        "   AND EXISTS (SELECT 1 FROM lidarr_track t"
+        "                WHERE t.norm_title = k.norm_track"
+        "                  AND instr(lower(t.title), lower(k.artist)) > 0)"
+    ).fetchone()
+    collision = store.connection.execute(
+        "SELECT COUNT(*) AS pairs, COALESCE(SUM(plays), 0) AS plays FROM scrobble_key k"
+        " WHERE k.track_id IS NULL"
+        "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
+        "   AND EXISTS (SELECT 1 FROM lidarr_track t WHERE t.norm_title = k.norm_track)"
+    ).fetchone()
+    logger.info(
+        "  the library's title credits the scrobbled artist: %d pairs, %d plays"
+        " -- remixes and guest spots, and the genuine misses",
+        attribution["pairs"],
+        attribution["plays"],
+    )
+    logger.info(
+        "  same title under an unrelated artist: %d pairs, %d plays"
+        " -- title collisions, not misses; matching these would be a mistake",
+        collision["pairs"] - attribution["pairs"],
+        collision["plays"] - attribution["plays"],
+    )
+    logger.info(
+        "  the rest, %d pairs, %d plays: you own the artist but not the track",
+        suspect["pairs"] - collision["pairs"],
+        suspect["plays"] - collision["plays"],
+    )
+
+    mismatched = store.connection.execute(
+        "SELECT k.artist, k.track, k.plays,"
+        "       (SELECT t.title FROM lidarr_track t"
+        "         WHERE t.norm_title = k.norm_track"
+        "           AND instr(lower(t.title), lower(k.artist)) > 0 LIMIT 1) AS library_title"
+        "  FROM scrobble_key k"
+        " WHERE k.track_id IS NULL"
+        "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
+        "   AND EXISTS (SELECT 1 FROM lidarr_track t"
+        "                WHERE t.norm_title = k.norm_track"
+        "                  AND instr(lower(t.title), lower(k.artist)) > 0)"
+        " ORDER BY k.plays DESC, k.artist LIMIT 10"
+    ).fetchall()
+    for position, row in enumerate(mismatched, start=1):
+        logger.info(
+            "  attribution %2d: %-45s %4d plays, library has %r",
+            position,
+            f"{row['artist']} - {row['track']}"[:45],
+            row["plays"],
+            row["library_title"],
+        )
+
     with_files = store.scalar("SELECT COUNT(*) FROM lidarr_track WHERE has_file = 1")
     played = store.scalar(
         "SELECT COUNT(DISTINCT k.track_id) FROM scrobble_key k"
@@ -1158,7 +1492,10 @@ def report(store, now):
         logger.info("  top track 90d %2d: %-50s %d", position, label[:50], row["plays"])
 
 
-def run_once(client, store, user, now, backfill_limit, lidarr=None):
+def run_once(
+    client, store, user, now, backfill_limit, lidarr=None, mirror=None,
+    library_root=None, playlist_limit=100,
+):
     """Run a single pass. Returns the number of scrobbles added."""
     started = time.monotonic()
     added = catch_up(client, store, user, now)
@@ -1171,6 +1508,15 @@ def run_once(client, store, user, now, backfill_limit, lidarr=None):
     # arrived, and they need a verdict too.
     if store.scalar("SELECT COUNT(*) FROM lidarr_track"):
         match_library(store)
+        if mirror is not None:
+            root = library_root or library_root_of(store)
+            if root is None:
+                logger.warning(
+                    "cannot work out where Lidarr's music lives, so no playlists were"
+                    " written; set --library-root"
+                )
+            else:
+                build_playlists(store, mirror, root, playlist_limit, now)
 
     logger.info(
         "pass complete in %.1fs: %d scrobbles added, %d loved",
@@ -1243,6 +1589,25 @@ def build_parser():
         default=os.getenv("MUSIC_CURATOR_LIDARR_API_KEY"),
         help="Lidarr API key (env MUSIC_CURATOR_LIDARR_API_KEY)",
     )
+    parser.add_argument(
+        "--mirror",
+        default=os.getenv("MUSIC_CURATOR_MIRROR"),
+        help="root of the MP3 mirror; playlists are written into it"
+        " (env MUSIC_CURATOR_MIRROR)",
+    )
+    parser.add_argument(
+        "--library-root",
+        default=os.getenv("MUSIC_CURATOR_LIBRARY_ROOT"),
+        help="prefix to strip from Lidarr's track paths when mapping them into the"
+        " mirror; derived from the indexed artist folders when unset"
+        " (env MUSIC_CURATOR_LIBRARY_ROOT)",
+    )
+    parser.add_argument(
+        "--playlist-limit",
+        type=int,
+        default=int(os.getenv("MUSIC_CURATOR_PLAYLIST_LIMIT", "100")),
+        help="most tracks to put in any one playlist (env MUSIC_CURATOR_PLAYLIST_LIMIT)",
+    )
     parser.add_argument(
         "--skip-index",
         action="store_true",
@@ -1307,7 +1672,17 @@ def main(argv=None, clock=time.time):
         while True:
             now = int(clock())
             try:
-                run_once(client, store, args.user, now, args.backfill_limit, lidarr)
+                run_once(
+                    client,
+                    store,
+                    args.user,
+                    now,
+                    args.backfill_limit,
+                    lidarr,
+                    args.mirror,
+                    args.library_root,
+                    args.playlist_limit,
+                )
             except (LastfmError, LidarrError) as error:
                 logger.error("%s", error)
                 if interval is None:

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "music-curator"
-version = "0.2.3"
+version = "0.4.0"
 description = "Ingest a Last.fm listening history and curate a music library from it"
 readme = "README.md"
 requires-python = ">=3.11"

tests/test_music_curator.py

@@ -1,5 +1,7 @@
 import json
+import stat
 import urllib.error
+from pathlib import Path
 
 import pytest
 from conftest import FakeLastfm, FakeLidarr, make_loved, make_tracks
@@ -55,8 +57,22 @@ LIBRARY = [
     {
         "name": "The Prodigy",
         "albums": [
-            {"title": "The Fat of the Land", "tracks": [{"title": "Breathe (Remastered)"}]}
+            {
+                "title": "The Fat of the Land",
+                "tracks": [
+                    {"title": "Breathe (Remastered)"},
+                    # Credits its remixer in the title, which is the only signal
+                    # separating a real attribution miss from a title collision.
+                    {"title": "Voodoo People (Pendulum Remix)"},
                 ],
+            }
+        ],
+    },
+    # Held by the library in its own right, which is what puts its scrobbles
+    # inside the "artist the library holds" filter at all.
+    {
+        "name": "Pendulum",
+        "albums": [{"title": "Immersion", "tracks": [{"title": "Watercolour"}]}],
     },
 ]
 
@@ -393,7 +409,7 @@ def test_keys_carry_the_play_count_and_the_span(tmp_path):
 def test_re_indexing_drops_what_lidarr_no_longer_has(tmp_path):
     """The index is Lidarr's mirror, not an accumulation of everything ever seen."""
     store = indexed(tmp_path, [scrobble_of("AC/DC", "Hells Bells")])
-    assert store.scalar("SELECT COUNT(*) FROM lidarr_artist") == 3
+    assert store.scalar("SELECT COUNT(*) FROM lidarr_artist") == 4
 
     music_curator.index_library(
         music_curator.Lidarr("http://lidarr", "key", transport=FakeLidarr(LIBRARY[:1])), store
@@ -512,7 +528,7 @@ def test_albums_come_from_the_unfiltered_endpoint(tmp_path):
 
     album_calls = [query for path, query in api.calls if path == "album"]
     assert album_calls == [{}]
-    assert store.scalar("SELECT COUNT(*) FROM lidarr_album") == 3
+    assert store.scalar("SELECT COUNT(*) FROM lidarr_album") == 4
 
 
 def test_a_bad_album_falls_back_to_asking_per_artist(tmp_path):
@@ -526,7 +542,7 @@ def test_a_bad_album_falls_back_to_asking_per_artist(tmp_path):
 
     assert store.scalar("SELECT COUNT(*) FROM lidarr_album WHERE artist_id = 2") == 0
     # The other two artists keep their albums.
-    assert store.scalar("SELECT COUNT(*) FROM lidarr_album") == 2
+    assert store.scalar("SELECT COUNT(*) FROM lidarr_album") == 3
     assert store.get_state("index_albums_skipped") == "1"
 
 
@@ -549,10 +565,10 @@ def test_an_artist_lidarr_cannot_serve_does_not_kill_the_index(tmp_path):
 
     music_curator.index_library(music_curator.Lidarr("http://lidarr", "key", transport=api), store)
 
-    assert store.scalar("SELECT COUNT(*) FROM lidarr_artist") == 3
+    assert store.scalar("SELECT COUNT(*) FROM lidarr_artist") == 4
     assert store.scalar("SELECT COUNT(*) FROM lidarr_track WHERE artist_id = 2") == 0
     # The other two artists are indexed in full.
-    assert store.scalar("SELECT COUNT(*) FROM lidarr_track") == 3
+    assert store.scalar("SELECT COUNT(*) FROM lidarr_track") == 5
     assert store.get_state("index_skipped") == "1"
 
 
@@ -687,3 +703,283 @@ def test_lidarr_talks_to_a_real_server_through_keep_alive(http_server):
         client.transport.close()
 
     assert server.connections == 1
+
+
+# Titles taken verbatim from a real coverage report's unmatched list. Each one
+# was a genuine miss before the normaliser handled it.
+@pytest.mark.parametrize(
+    ("scrobbled", "tagged"),
+    [
+        ("Self vs Self (feat. In Flames)", "Self vs Self"),
+        ("Grime Battle of Hastings (feat. The Town Crier)", "Grime Battle of Hastings"),
+        ("Gold Dust - Shy FX Re-Edit", "Gold Dust"),
+        ("Back To Your Roots - Friction & K-Tee Remix", "Back To Your Roots"),
+        ("Constellations - Forza Horizon 3 VIP", "Constellations"),
+        ("Everyday (Netsky Remix)", "Everyday"),
+        ("Voodoo People [Pendulum Remix] [Live At Brixton Academy]", "Voodoo People"),
+    ],
+)
+def test_real_unmatched_titles_now_agree_with_their_tags(scrobbled, tagged):
+    assert music_curator.normalise(scrobbled) == music_curator.normalise(tagged)
+
+
+@pytest.mark.parametrize(
+    "title",
+    [
+        "Dancing with Myself",
+        "(Don't Fear) The Reaper",
+        "Live and Let Die",
+        "Radio Ga Ga",
+        "Editors",
+        "Mixed Emotions",
+        "Vipassana",
+    ],
+)
+def test_the_version_words_do_not_eat_ordinary_titles(title):
+    """Every one of these contains a version word and must survive intact."""
+    assert music_curator.normalise(title) == music_curator.normalise(title.lower())
+    assert len(music_curator.normalise(title).split()) == len(title.split())
+
+
+def test_a_bracketed_guest_credit_matches_the_bare_tag(tmp_path):
+    """Whitespace-then-feat misses the bracketed form, which is most of them."""
+    store = indexed(tmp_path, [scrobble_of("Yellowcard", "Here I Am Alive (feat. Someone)")])
+
+    method, track_id = verdict(store, "Yellowcard", "Here I Am Alive (feat. Someone)")
+    assert method == "name"
+    assert track_id is not None
+
+
+def test_misses_are_split_by_whether_the_library_holds_the_title(tmp_path):
+    """Owning an artist is a weak proxy for owning a track. Counting both as
+    matcher failures overstates the problem and would over-block the cull."""
+    store = indexed(
+        tmp_path,
+        [
+            # The library holds "Hells Bells", but under AC/DC, not Yellowcard:
+            # an attribution disagreement, and a real miss.
+            scrobble_of("Yellowcard", "Hells Bells"),
+            # Yellowcard is in the library; this track is not, under any artist.
+            scrobble_of("Yellowcard", "A Single She Never Bought"),
+        ],
+    )
+
+    def count(extra):
+        return store.connection.execute(
+            "SELECT COUNT(*) FROM scrobble_key k WHERE k.track_id IS NULL"
+            "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
+            f"  {extra}"
+        ).fetchone()[0]
+
+    assert count("") == 2
+    assert count("AND EXISTS (SELECT 1 FROM lidarr_track t WHERE t.norm_title = k.norm_track)") == 1
+
+
+def test_the_report_survives_the_attribution_split(tmp_path):
+    store = indexed(tmp_path, [scrobble_of("Yellowcard", "Hells Bells")])
+
+    music_curator.report(store, NOW)
+
+
+def attribution_pairs(store):
+    """Unmatched pairs where the library's own title credits the scrobbled artist."""
+    return [
+        row["artist"]
+        for row in store.connection.execute(
+            "SELECT k.artist FROM scrobble_key k WHERE k.track_id IS NULL"
+            "   AND EXISTS (SELECT 1 FROM lidarr_artist a WHERE a.norm_name = k.norm_artist)"
+            "   AND EXISTS (SELECT 1 FROM lidarr_track t"
+            "                WHERE t.norm_title = k.norm_track"
+            "                  AND instr(lower(t.title), lower(k.artist)) > 0)"
+        )
+    ]
+
+
+def test_a_shared_title_is_not_an_attribution_miss(tmp_path):
+    """Across fifty thousand tracks, titles collide constantly: "Everyday" is
+    Rusko and also Def Leppard. Matching those would be worse than missing."""
+    store = indexed(tmp_path, [scrobble_of("Yellowcard", "Hells Bells")])
+
+    # The library holds "Hells Bells", by AC/DC, and its title says nothing
+    # about Yellowcard. A collision, not a miss.
+    assert verdict(store, "Yellowcard", "Hells Bells") == ("none", None)
+    assert attribution_pairs(store) == []
+
+
+def test_a_remix_credited_in_the_library_title_is_an_attribution_miss(tmp_path):
+    """The library has "Voodoo People (Pendulum Remix)" under The Prodigy; the
+    scrobble credits Pendulum. Same song, different filing."""
+    store = indexed(tmp_path, [scrobble_of("Pendulum", "Voodoo People")])
+
+    assert attribution_pairs(store) == ["Pendulum"]
+
+
+def test_the_title_index_is_used_for_the_report_lookup(tmp_path):
+    """Without it the report scans every track for every unmatched key: forty-
+    seven seconds on a real library."""
+    store = indexed(tmp_path, [])
+
+    plan = "\n".join(
+        row[-1]
+        for row in store.connection.execute(
+            "EXPLAIN QUERY PLAN SELECT 1 FROM scrobble_key k WHERE k.track_id IS NULL"
+            " AND EXISTS (SELECT 1 FROM lidarr_track t WHERE t.norm_title = k.norm_track)"
+        )
+    )
+    assert "lidarr_track_title" in plan, plan
+
+
+def playlist_library(tmp_path):
+    """A library on disk, with mirror MP3s beside the Lidarr source paths."""
+    source = tmp_path / "music"
+    mirror = tmp_path / "mirror"
+    library = [
+        {
+            "name": "Played Band",
+            "albums": [
+                {
+                    "title": "Known",
+                    "tracks": [{"title": "Hit"}, {"title": "Album Track"}],
+                }
+            ],
+        },
+        {
+            "name": "Silent Band",
+            "albums": [{"title": "Unknown", "tracks": [{"title": "Never Heard"}]}],
+        },
+    ]
+    api = FakeLidarr(library)
+    # FakeLidarr invents /music/<artist>/<album>/<title>.flac; put the mirror
+    # MP3s at the paths music-mirror would have produced from those.
+    for handle in api.files:
+        relative = Path(handle["path"]).relative_to("/music")
+        handle["path"] = str(source / relative)
+        mp3 = (mirror / relative).with_suffix(".mp3")
+        mp3.parent.mkdir(parents=True, exist_ok=True)
+        mp3.write_bytes(b"not really an mp3")
+    for artist in api.artists:
+        artist["path"] = str(source / Path(artist["path"]).name)
+    return api, source, mirror
+
+
+def test_playlists_are_written_into_the_mirror(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    ingest(store, FakeLastfm([scrobble_of("Played Band", "Hit")] * 1))
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    music_curator.build_playlists(store, mirror, str(source), 100, NOW)
+
+    written = sorted(p.name for p in (mirror / "_playlists").glob("*.m3u"))
+    assert written == [
+        "all-time.m3u",
+        "deep-cuts.m3u",
+        "heavy-rotation.m3u",
+        "neglected.m3u",
+        "unheard-favourites.m3u",
+        "unheard.m3u",
+    ]
+
+    played = (mirror / "_playlists" / "all-time.m3u").read_text().splitlines()
+    assert played[0] == "#EXTM3U"
+    assert played[1].startswith("#EXTINF:")
+    assert "Played Band - Hit" in played[1]
+    # Relative to the playlist file, so the same file works from any mount.
+    assert played[2] == "../Played Band/Known/Hit.mp3"
+    assert (mirror / "_playlists" / "all-time.m3u").parent.joinpath(played[2]).resolve().is_file()
+
+
+def test_an_unplayed_track_lands_in_the_unheard_playlist(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    ingest(store, FakeLastfm([scrobble_of("Played Band", "Hit")]))
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    music_curator.build_playlists(store, mirror, str(source), 100, NOW)
+
+    unheard = (mirror / "_playlists" / "unheard.m3u").read_text()
+    assert "Never Heard" in unheard
+    assert "Album Track" in unheard
+    # The one thing that was played must not be in it.
+    assert "Played Band - Hit\n" not in unheard
+
+
+def test_a_track_with_no_mirror_file_is_left_out(tmp_path):
+    """The index knows what Lidarr holds; whether music-mirror has encoded the
+    MP3 yet is a different question, and is checked rather than assumed."""
+    api, source, mirror = playlist_library(tmp_path)
+    for mp3 in mirror.rglob("*.mp3"):
+        mp3.unlink()
+    store = store_at(tmp_path)
+    ingest(store, FakeLastfm([scrobble_of("Played Band", "Hit")]))
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    total = music_curator.build_playlists(store, mirror, str(source), 100, NOW)
+
+    assert total == 0
+    assert (mirror / "_playlists" / "all-time.m3u").read_text() == "#EXTM3U\n"
+
+
+def test_the_library_root_is_derived_from_the_artist_folders(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+
+    assert music_curator.library_root_of(store) == str(source)
+
+
+def test_the_playlist_limit_is_honoured(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    music_curator.build_playlists(store, mirror, str(source), 1, NOW)
+
+    unheard = (mirror / "_playlists" / "unheard.m3u").read_text().splitlines()
+    assert len([line for line in unheard if line.startswith("#EXTINF")]) == 1
+
+
+def test_the_rotation_moves_weekly_not_every_pass(tmp_path):
+    """A playlist that reorders on every pass is one that has to be re-imported
+    on every pass; the Music app imports a snapshot, it does not track a file."""
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    def unheard_at(when):
+        music_curator.build_playlists(store, mirror, str(source), 100, when)
+        return (mirror / "_playlists" / "unheard.m3u").read_text()
+
+    same_week = unheard_at(NOW), unheard_at(NOW + 3600)
+    assert same_week[0] == same_week[1]
+
+
+def test_playlists_are_group_readable(tmp_path):
+    api, source, mirror = playlist_library(tmp_path)
+    store = store_at(tmp_path)
+    music_curator.index_library(
+        music_curator.Lidarr("http://lidarr", "key", transport=api), store
+    )
+    music_curator.match_library(store)
+
+    music_curator.build_playlists(store, mirror, str(source), 100, NOW)
+
+    for playlist in (mirror / "_playlists").glob("*.m3u"):
+        assert playlist.stat().st_mode & stat.S_IRGRP, playlist