muse/app.py

"""
muse — request a song, get it in Navidrome.

The gap this fills: Lidarr thinks in MusicBrainz *releases*, so the smallest
thing it can want is an album. Soulseek trades individual files. Nothing in
between existed, so requesting one song meant either hoovering a whole
discography through Lidarr or hand-filing downloads out of slskd's UI.

The trick that makes this simple: the user PICKS a specific track before
anything is downloaded. That means we hold ground-truth metadata — artist,
album, year, track number — before the file exists. Taggers like beets have to
infer all of it afterwards from whatever a stranger named the file; we never
guess. We tag what was asked for.

Flow:
    /api/search   Deezer track search -> the user picks one
    /api/request  /track/{id} for exact metadata, then slskd search ->
                  score candidates -> enqueue best
    worker        poll slskd -> on completion: tag, then hardlink into
                  /data/media/music/<Artist>/<Album> (<Year>)/NN - Title.ext

Hardlink, not copy: the download stays in slskd's completed dir so it keeps
being shared, while the library gets its own name for the same bytes. One
file on disk, two names — the same trick the *arrs use.
"""

import asyncio
import difflib
import logging
import os
import re
import sqlite3
import uuid
from contextlib import asynccontextmanager
from pathlib import Path

import httpx
from fastapi import FastAPI, HTTPException
from fastapi.responses import FileResponse
from fastapi.staticfiles import StaticFiles
from pydantic import BaseModel

log = logging.getLogger("muse")
logging.basicConfig(
    level=os.getenv("MUSE_LOG_LEVEL", "INFO"),
    format="[%(levelname)s|%(name)s] %(asctime)s: %(message)s",
    datefmt="%Y-%m-%dT%H:%M:%S%z",
)

SLSKD_URL = os.getenv("SLSKD_URL", "http://192.168.1.87:5030").rstrip("/")
SLSKD_API_KEY = os.getenv("SLSKD_API_KEY", "")
DOWNLOAD_DIR = Path(os.getenv("MUSE_DOWNLOAD_DIR", "/data/torrents/soulseek/complete"))
LIBRARY_DIR = Path(os.getenv("MUSE_LIBRARY_DIR", "/data/media/music"))
DB_PATH = Path(os.getenv("MUSE_DB", "/config/muse.db"))

# Ordered best-first. A candidate's extension must appear here or it is
# rejected outright — this is what keeps .m4a rips and stray .cue files out.
FORMATS = [f.strip() for f in os.getenv("MUSE_FORMATS", "flac,mp3").split(",") if f.strip()]

# Below this, the remote filename doesn't plausibly contain the track title.
# Soulseek filenames are chaotic ("08-radiohead-creep_(accoustic)-ent.mp3"),
# so this compares against a normalised form, not the raw string.
MIN_MATCH = float(os.getenv("MUSE_MIN_MATCH", "0.45"))

# How many Soulseek searches may run at once.
#
# This is a hard cap, not a suggestion. Each search takes 10-40s of polling
# and real load on slskd and the Soulseek server, so firing one task per
# request — which is what an earlier version did — means pasting a 50-track
# list launches 50 concurrent searches and drowns the client. Everything goes
# through a bounded queue instead: requests return instantly, work drains at a
# civil rate.
WORKERS = int(os.getenv("MUSE_WORKERS", "3"))

# Search uses Deezer's public API: no key, no OAuth, and vastly better
# relevance than MusicBrainz for the way people actually type.
#
# MusicBrainz was tried first and rejected on measurements, not taste. It is a
# superb database but a poor search engine: "radiohead creep" returned eight
# covers and mashups before the Radiohead recording, and "bohemian rhapsody"
# never surfaced Queen at all across 90 results — because its text index
# matches "radiohead" inside a cover's *title* as readily as in the artist
# field. Client-side re-ranking can only reorder what came back, so it could
# not fix a result set that lacked the right row. Deezer returns
# "Creep / Radiohead / Pablo Honey" first for the same query.
#
# Deezer search omits track number and release date, so those are fetched from
# /track/{id} at request time — one extra call per request, not per result.
DEEZER = "https://api.deezer.com"

# Derivative versions to push down when the user did not ask for them.
# Matched on WORD BOUNDARIES, not as substrings: a bare "live" is needed to
# catch "Come As You Are (Live)", but substring matching would also penalise
# "Deliver Me" and "Olive". Same trap for "demo" vs "Demolition".
#
# A term only counts against a result when the user did NOT type it, so asking
# for "oasis live forever" or "nirvana unplugged demo" still works.
DERIVATIVE = ("cover", "covers", "karaoke", "tribute", "remix", "remixed",
              "instrumental", "acapella", "originally performed",
              "made popular by", "in the style of", "demo", "rehearsal",
              "live", "unplugged", "re recorded", "rerecorded",
              "sped up", "slowed", "reverb", "8d", "nightcore")
DERIVATIVE_RE = re.compile(r"\b(" + "|".join(DERIVATIVE) + r")\b")

SAFE = re.compile(r'[<>:"/\\|?*\x00-\x1f]')
NOISE = re.compile(r"[^a-z0-9]+")


def safe(name: str) -> str:
    """Filesystem-safe path component. Trailing dots/spaces break on some FS."""
    return SAFE.sub("_", name).strip(". ") or "Unknown"


def norm(s: str) -> str:
    return NOISE.sub(" ", s.lower()).strip()


# ─────────────────────────── storage ───────────────────────────

def db():
    c = sqlite3.connect(DB_PATH, timeout=30)
    c.row_factory = sqlite3.Row
    return c


def init_db():
    DB_PATH.parent.mkdir(parents=True, exist_ok=True)
    with db() as c:
        c.execute("""
            CREATE TABLE IF NOT EXISTS jobs (
                id TEXT PRIMARY KEY,
                state TEXT NOT NULL,          -- searching|queued|downloading|imported|failed
                detail TEXT,
                artist TEXT, title TEXT, album TEXT,
                year TEXT, track INTEGER,
                recording_mbid TEXT, release_mbid TEXT,
                username TEXT, remote_file TEXT, size INTEGER,
                imported_path TEXT,
                created_at TEXT DEFAULT (datetime('now'))
            )
        """)


def set_state(job_id: str, state: str, detail: str = "", **cols):
    sets = ", ".join(f"{k}=?" for k in cols)
    sql = f"UPDATE jobs SET state=?, detail=?{', ' + sets if sets else ''} WHERE id=?"
    with db() as c:
        c.execute(sql, [state, detail, *cols.values(), job_id])
    log.info("job %s -> %s %s", job_id[:8], state, detail)


# ─────────────────────────── slskd ───────────────────────────

def slskd_client() -> httpx.AsyncClient:
    return httpx.AsyncClient(
        base_url=SLSKD_URL,
        headers={"X-API-Key": SLSKD_API_KEY},
        timeout=httpx.Timeout(30.0),
    )


def ext_of(remote_path: str) -> str:
    # Soulseek paths are Windows-style; the API's `extension` field comes back
    # empty in practice, so derive it ourselves.
    return remote_path.rsplit(".", 1)[-1].lower() if "." in remote_path else ""


def score(candidate: dict, peer: dict, want: str) -> float | None:
    """Rank one remote file. None = reject.

    Ordering intent: correct format first, then actually-gettable (a peer with
    a free slot and a short queue beats a marginally better file behind 200
    people), then filename plausibility, then bitrate.
    """
    ext = ext_of(candidate["filename"])
    if ext not in FORMATS:
        return None

    base = candidate["filename"].replace("\\", "/").rsplit("/", 1)[-1]
    ratio = difflib.SequenceMatcher(None, norm(base), want).ratio()
    # A short title inside a long scene filename scores badly on raw ratio, so
    # give credit for simple containment too.
    if want in norm(base):
        ratio = max(ratio, 0.9)
    if ratio < MIN_MATCH:
        return None

    s = 0.0
    s += (len(FORMATS) - FORMATS.index(ext)) * 100      # format preference
    s += 60 if peer.get("hasFreeUploadSlot") else 0
    s += max(0, 40 - peer.get("queueLength", 0))        # short queues
    s += min(peer.get("uploadSpeed", 0) / 50_000, 20)   # capped speed bonus
    s += ratio * 50
    s += min(candidate.get("bitRate") or 0, 320) / 32   # 0-10
    return s


async def pick_and_enqueue(job_id: str, artist: str, title: str):
    """Search slskd for one track and queue the best candidate."""
    want = norm(f"{artist} {title}")
    search_id = str(uuid.uuid4())
    async with slskd_client() as sc:
        try:
            r = await sc.post("/api/v0/searches",
                              json={"id": search_id, "searchText": f"{artist} {title}"})
            r.raise_for_status()
        except Exception as e:
            set_state(job_id, "failed", f"slskd search failed: {e}")
            return

        # Poll until slskd calls it done, or we run out of patience. Soulseek
        # searches trickle in; results at t=5s are much worse than at t=15s.
        results = {}
        for _ in range(20):
            await asyncio.sleep(2)
            try:
                rr = await sc.get(f"/api/v0/searches/{search_id}",
                                  params={"includeResponses": "true"})
                rr.raise_for_status()
                results = rr.json()
            except Exception:
                continue
            if str(results.get("state", "")).startswith("Completed"):
                break

        best, best_score = None, -1.0
        for peer in results.get("responses", []):
            for f in peer.get("files", []):
                if f.get("isLocked"):
                    continue
                sc_ = score(f, peer, want)
                if sc_ is not None and sc_ > best_score:
                    best, best_score = (peer, f), sc_

        # Tidy up so slskd's search history doesn't grow without bound.
        try:
            await sc.delete(f"/api/v0/searches/{search_id}")
        except Exception:
            pass

        if not best:
            set_state(job_id, "failed",
                      f"no {'/'.join(FORMATS)} candidate matched (searched {results.get('fileCount', 0)} files)")
            return

        peer, f = best
        try:
            er = await sc.post(f"/api/v0/transfers/downloads/{peer['username']}",
                               json=[{"filename": f["filename"], "size": f["size"]}])
            er.raise_for_status()
        except Exception as e:
            set_state(job_id, "failed", f"enqueue failed: {e}")
            return

        set_state(job_id, "downloading",
                  f"from {peer['username']} (score {best_score:.0f})",
                  username=peer["username"], remote_file=f["filename"], size=f["size"])


# ─────────────────────────── import ───────────────────────────

def tag_and_file(job: sqlite3.Row, src: Path) -> Path:
    """Write known-good tags, then hardlink into the library."""
    from mutagen import File as MFile
    from mutagen.easyid3 import EasyID3
    from mutagen.flac import FLAC
    from mutagen.mp3 import MP3

    try:
        if src.suffix.lower() == ".flac":
            audio = FLAC(src)
        else:
            try:
                audio = EasyID3(src)
            except Exception:
                m = MP3(src)
                m.add_tags()
                m.save()
                audio = EasyID3(src)

        audio["title"] = job["title"]
        audio["artist"] = job["artist"]
        audio["albumartist"] = job["artist"]
        if job["album"]:
            audio["album"] = job["album"]
        if job["year"]:
            audio["date"] = str(job["year"])
        if job["track"]:
            audio["tracknumber"] = str(job["track"])
        # Deliberately NOT writing the Deezer id into musicbrainz_trackid --
        # that field means a specific thing and a wrong value there would
        # mislead Picard, beets, or anything else that later reads this file.
        audio.save()
    except Exception as e:
        # A tagging failure must not strand the file — Navidrome can still
        # read whatever tags it shipped with.
        log.warning("tagging failed for %s: %s", src.name, e)

    album = f"{safe(job['album'] or 'Singles')}"
    if job["year"]:
        album += f" ({job['year']})"
    dest_dir = LIBRARY_DIR / safe(job["artist"]) / album
    dest_dir.mkdir(parents=True, exist_ok=True)

    num = f"{int(job['track']):02d} - " if job["track"] else ""
    dest = dest_dir / f"{num}{safe(job['title'])}{src.suffix.lower()}"

    if dest.exists():
        return dest
    try:
        os.link(src, dest)          # same filesystem -> free, and keeps sharing
    except OSError:
        import shutil
        shutil.copy2(src, dest)     # cross-device fallback
    return dest


def find_downloaded(remote_file: str) -> Path | None:
    """Locate the completed file. slskd nests it under the remote's parent
    directory name, so match on basename rather than guessing the path."""
    base = remote_file.replace("\\", "/").rsplit("/", 1)[-1]
    for p in DOWNLOAD_DIR.rglob("*"):
        if p.is_file() and p.name == base:
            return p
    return None


# ─────────────────────────── search queue ───────────────────────────

work_q: "asyncio.Queue[str]" = asyncio.Queue()


async def search_worker(n: int):
    """Drain the search queue, one job at a time, WORKERS of these running."""
    while True:
        job_id = await work_q.get()
        try:
            with db() as c:
                job = c.execute("SELECT * FROM jobs WHERE id=?", (job_id,)).fetchone()
            if job and job["state"] in ("queued", "searching"):
                set_state(job_id, "searching", "")
                await pick_and_enqueue(job_id, job["artist"], job["title"])
        except Exception as e:
            log.warning("worker %d: %s", n, e)
            try:
                set_state(job_id, "failed", str(e)[:200])
            except Exception:
                pass
        finally:
            work_q.task_done()


async def import_worker():
    """Poll slskd for in-flight downloads and import the finished ones."""
    while True:
        await asyncio.sleep(10)
        try:
            with db() as c:
                jobs = c.execute(
                    "SELECT * FROM jobs WHERE state='downloading'").fetchall()
            if not jobs:
                continue

            async with slskd_client() as sc:
                r = await sc.get("/api/v0/transfers/downloads")
                r.raise_for_status()
                payload = r.json()

            # Flatten slskd's user -> directories -> files nesting.
            states: dict[str, str] = {}
            for user in payload:
                for d in user.get("directories", []):
                    for f in d.get("files", []):
                        states[f.get("filename", "")] = str(f.get("state", ""))

            for job in jobs:
                st = states.get(job["remote_file"], "")
                if "Completed, Succeeded" in st:
                    src = find_downloaded(job["remote_file"])
                    if not src:
                        continue  # still being moved out of incomplete
                    dest = tag_and_file(job, src)
                    set_state(job["id"], "imported", str(dest), imported_path=str(dest))
                elif "Completed" in st and "Succeeded" not in st:
                    set_state(job["id"], "failed", f"transfer {st}")
        except Exception as e:
            log.warning("worker: %s", e)


# ─────────────────────────── api ───────────────────────────

@asynccontextmanager
async def lifespan(_: FastAPI):
    init_db()

    # The queue lives in memory, so a restart mid-import would otherwise
    # strand every unstarted job in 'queued' forever with nothing to pick it
    # up. Anything not yet handed to slskd goes back on the queue.
    with db() as c:
        orphans = [r["id"] for r in c.execute(
            "SELECT id FROM jobs WHERE state IN ('queued','searching')").fetchall()]
    for jid in orphans:
        work_q.put_nowait(jid)
    if orphans:
        log.info("requeued %d job(s) left over from the last run", len(orphans))

    tasks = [asyncio.create_task(import_worker())]
    tasks += [asyncio.create_task(search_worker(i)) for i in range(WORKERS)]
    log.info("muse up — slskd=%s library=%s formats=%s workers=%d",
             SLSKD_URL, LIBRARY_DIR, ",".join(FORMATS), WORKERS)
    yield
    for t in tasks:
        t.cancel()


app = FastAPI(title="muse", lifespan=lifespan)


class RequestBody(BaseModel):
    artist: str
    title: str
    album: str | None = None
    year: str | None = None
    track: int | None = None
    deezer_id: int | None = None


async def deezer_search(q: str, limit: int = 25) -> list[dict]:
    """Ranked Deezer track search. Shared by /api/search and /api/import."""
    if not q.strip():
        return []
    async with httpx.AsyncClient(timeout=20) as c:
        r = await c.get(f"{DEEZER}/search", params={"q": q, "limit": limit * 2})
        if r.status_code != 200:
            raise HTTPException(502, f"deezer returned {r.status_code}")
        data = r.json()

    qn = norm(q)
    scored = []
    for t in data.get("data", []):
        title = t.get("title") or ""
        artist = (t.get("artist") or {}).get("name") or ""
        album = (t.get("album") or {}).get("title") or ""

        s = float(t.get("rank") or 0) / 10_000       # Deezer popularity, 0-~100

        # An artist match separates the real track from a covers compilation.
        at = set(norm(artist).split())
        if norm(artist) and norm(artist) in qn:
            s += 300

        # Title relevance. This has to dominate everything else: without it a
        # completely different song by the right artist competes on equal
        # footing, and "Oasis - Live Forever" resolved to "Cloudburst" because
        # a variant penalty was enough to tip between two unrelated tracks.
        # Whatever query words the artist name did not explain should show up
        # in the title.
        rest = set(qn.split()) - at
        if rest:
            s += 400 * len(rest & set(norm(title).split())) / len(rest)
        for w in set(DERIVATIVE_RE.findall(norm(title))):
            if not re.search(rf"\b{re.escape(w)}\b", qn):
                s -= 150
        # Catch-all for variants the keyword list will never cover in full
        # ("Devonshire Mix", "Radio Edit", "2011 Version"): a plain title beats
        # a parenthesised one. Deliberately small -- it should break ties
        # between versions of the SAME song, never outweigh title relevance
        # above and pick a different song.
        if ("(" in title or "[" in title) and "(" not in q:
            s -= 40
        scored.append((s, {
            "deezer_id": t.get("id"),
            "title": title,
            "artist": artist,
            "album": album,
            "length": (t.get("duration") or 0) * 1000,
            "cover": (t.get("album") or {}).get("cover_medium"),
            # filled in at request time from /track/{id}
            "year": None, "track": None,
        }))

    seen, out = set(), []
    for _, item in sorted(scored, key=lambda x: -x[0]):
        key = (norm(item["artist"]), norm(item["title"]))
        if key in seen:
            continue
        seen.add(key)
        out.append(item)
    return out[:limit]


@app.get("/api/search")
async def search(q: str, limit: int = 25):
    """Track search. Returns individual songs, not albums — the whole point."""
    return {"results": await deezer_search(q, limit)}


async def enqueue_job(artist: str, title: str, album=None, year=None,
                      track=None, deezer_id=None) -> str:
    """Create a job and hand it to the search queue."""
    job_id = str(uuid.uuid4())
    with db() as c:
        c.execute(
            "INSERT INTO jobs (id,state,detail,artist,title,album,year,track,"
            "recording_mbid,release_mbid) VALUES (?,?,?,?,?,?,?,?,?,?)",
            (job_id, "queued", "", artist, title, album, year, track,
             str(deezer_id or ""), None))
    await work_q.put(job_id)
    return job_id


async def enrich(deezer_id) -> dict:
    """Deezer's search payload has no track number or release date, and both
    go into the filename and tags. One lookup per request, not per result."""
    if not deezer_id:
        return {}
    try:
        async with httpx.AsyncClient(timeout=15) as c:
            r = await c.get(f"{DEEZER}/track/{deezer_id}")
            if r.status_code != 200:
                return {}
            d = r.json()
            rd = d.get("release_date") or (d.get("album") or {}).get("release_date")
            return {
                "track": d.get("track_position"),
                "year": (rd or "")[:4] or None,
                "album": (d.get("album") or {}).get("title"),
                "artist": (d.get("artist") or {}).get("name"),
            }
    except Exception as e:
        log.warning("deezer track lookup failed: %s", e)
        return {}


@app.post("/api/request")
async def request_track(body: RequestBody):
    e = await enrich(body.deezer_id)
    return {"id": await enqueue_job(
        e.get("artist") or body.artist, body.title,
        e.get("album") or body.album, e.get("year") or body.year,
        e.get("track") or body.track, body.deezer_id)}


class ImportBody(BaseModel):
    text: str


# "Artist - Title", "Artist – Title" (en dash), "Title by Artist". Anything
# that doesn't split just goes to Deezer whole — its search copes fine with
# free text, and guessing wrong is worse than not guessing.
SPLIT = re.compile(r"\s+[-–—|]\s+|\s+by\s+", re.I)
DEEZER_URL = re.compile(r"deezer\.com/(?:[a-z]{2}/)?(album|playlist)/(\d+)")


async def expand_url(kind: str, ident: str) -> list[str]:
    """Turn a Deezer album/playlist URL into 'Artist - Title' lines."""
    out = []
    try:
        async with httpx.AsyncClient(timeout=25) as c:
            r = await c.get(f"{DEEZER}/{kind}/{ident}")
            if r.status_code != 200:
                return []
            for t in (r.json().get("tracks") or {}).get("data", []):
                a = (t.get("artist") or {}).get("name") or ""
                out.append(f"{a} - {t.get('title')}" if a else t.get("title", ""))
    except Exception as e:
        log.warning("expanding %s/%s failed: %s", kind, ident, e)
    return [x for x in out if x]


@app.post("/api/import")
async def import_list(body: ImportBody):
    """Paste a list, get it all queued.

    Each line is resolved against Deezer and the top-ranked hit is taken
    automatically — there is no picker in bulk mode. Resolved artist/album are
    recorded on every job so a wrong auto-pick is visible in the list rather
    than silently landing the wrong song in the library.
    """
    lines, queued, skipped = [], [], []
    for raw in body.text.splitlines():
        line = raw.strip().lstrip("-•*0123456789. )").strip()
        if not line:
            continue
        m = DEEZER_URL.search(line)
        lines.extend(await expand_url(*m.groups()) if m else [line])

    for line in lines:
        # Search the line as typed; the ranker already weights an artist-name
        # match heavily, so "Artist - Title" needs no special handling.
        try:
            # Ask for a real candidate pool, not one row. deezer_search ranks
            # what it is given, so limit=1 fetched two results and handed back
            # the better of a demo and a rehearsal take while the studio
            # version -- 6x the popularity -- sat further down the list unseen.
            hits = await deezer_search(SPLIT.sub(" ", line), limit=10)
        except Exception as e:
            skipped.append({"line": line, "why": str(e)[:120]})
            continue
        if not hits:
            skipped.append({"line": line, "why": "no match on Deezer"})
            continue
        h = hits[0]
        e = await enrich(h["deezer_id"])
        jid = await enqueue_job(
            e.get("artist") or h["artist"], h["title"],
            e.get("album") or h["album"], e.get("year"), e.get("track"),
            h["deezer_id"])
        queued.append({"id": jid, "line": line,
                       "matched": f"{h['artist']} — {h['title']}"})

    log.info("import: %d queued, %d skipped", len(queued), len(skipped))
    return {"queued": queued, "skipped": skipped, "pending": work_q.qsize()}


@app.get("/api/jobs")
async def jobs(limit: int = 200):
    with db() as c:
        rows = c.execute(
            "SELECT * FROM jobs ORDER BY created_at DESC, rowid DESC LIMIT ?",
            (limit,)).fetchall()
        counts = {r["state"]: r["n"] for r in c.execute(
            "SELECT state, COUNT(*) n FROM jobs GROUP BY state").fetchall()}
    return {"jobs": [dict(r) for r in rows], "counts": counts,
            "pending": work_q.qsize()}


@app.post("/api/retry/{job_id}")
async def retry(job_id: str):
    """Put a failed job back on the queue. Soulseek failures are usually the
    peer going offline, not the request being wrong, so retrying later often
    just works."""
    with db() as c:
        row = c.execute("SELECT id FROM jobs WHERE id=?", (job_id,)).fetchone()
    if not row:
        raise HTTPException(404, "no such job")
    set_state(job_id, "queued", "")
    await work_q.put(job_id)
    return {"ok": True}


@app.get("/api/health")
async def health():
    ok, detail = True, {}
    async with slskd_client() as sc:
        try:
            r = await sc.get("/api/v0/application")
            detail["slskd"] = r.status_code
            ok = ok and r.status_code == 200
        except Exception as e:
            detail["slskd"] = str(e)
            ok = False
    detail["library_writable"] = os.access(LIBRARY_DIR, os.W_OK)
    detail["downloads_readable"] = os.access(DOWNLOAD_DIR, os.R_OK)
    return {"ok": ok and detail["library_writable"], **detail}


STATIC = Path(__file__).parent / "static"


@app.get("/")
async def index():
    return FileResponse(STATIC / "index.html")


app.mount("/static", StaticFiles(directory=STATIC), name="static")