feat(music): transcode WMA/Musepack to AAC so they actually play

Tagging and covers for these two formats landed already, but neither one
decodes in any mainstream browser's <audio> element at all — a real
library scan turned up 273 such files that would show up correctly in the
Music app and then simply fail on click. This closes that gap: the node
transcodes to AAC/M4A on request (a one-shot whole-file conversion, not
live-piped like video's fMP4 segments — an audio file is small enough that
streaming it buys nothing), caches the result under its own content hash
the same way a TMDB poster or a MusicBrainz cover is cached, and serves it
back through the ordinary file_req/chunk path. That path used to assume
anything in the media cache was thumbnail-sized (single chunk, always);
generalized it to slice a cached blob the same way a real file on disk
gets sliced, since a transcoded track can be several MB.

New MNP pair (`audio_transcode_req`/`_resp`, version bump to 0.9), shares
its concurrency cap with video's transcode pool rather than getting its
own — both are real ffmpeg processes on the same node. Every other audio
format is untouched: this only fires for .wma/.mpc, the two extensions
that need it.

1 files changed, 193 insertions, 0 deletions

diff --git a/packages/meshbay-node/tests/test_audio_transcode.py b/packages/meshbay-node/tests/test_audio_transcode.py
new file mode 100644
index 0000000..725839b
--- /dev/null
+++ b/packages/meshbay-node/tests/test_audio_transcode.py
@@ -0,0 +1,193 @@
+"""
+Tests for the Music app's one exception to "no node-side transcode pool"
+(docs/musicbay.md §2.2): WMA and Musepack tag/cover fine (enrich_audio.py)
+but decode in no mainstream browser's <audio> element at all, so
+`_do_audio_transcode_request` converts to AAC/M4A on request and caches the
+result — served back through the ordinary file_req/chunk path, generalized
+in `_try_serve_thumbnail` to handle more than one chunk.
+"""
+
+import shutil
+import subprocess
+from pathlib import Path
+
+import blake3
+import pytest
+from cryptography.hazmat.primitives.asymmetric.ed25519 import Ed25519PrivateKey
+from meshbay_common.crypto import generate_gek
+from meshbay_common.protocol import MNP, IndexEntry
+from meshbay_common.webcrypto import chunk_key_aes, decrypt_chunk_aes
+from meshbay_node.indexer.group_index import GroupIndex
+from meshbay_node.media_cache import MediaCache
+from meshbay_node.transport import webrtc_server as wrs
+from meshbay_node.transport.webrtc_server import WebRTCPeerSession
+
+from conftest import one_root
+
+_HAVE_FFMPEG = shutil.which("ffmpeg") and shutil.which("ffprobe")
+pytestmark = pytest.mark.asyncio
+
+
+@pytest.fixture
+async def media_cache(tmp_path):
+    c = MediaCache(db_path=tmp_path / "media_cache.db")
+    await c.open()
+    yield c
+    await c.close()
+
+
+def _make_wma_clip(path: Path) -> None:
+    subprocess.run(
+        ["ffmpeg", "-hide_banner", "-loglevel", "error", "-y",
+         "-f", "lavfi", "-i", "sine=frequency=440:duration=1",
+         "-c:a", "wmav2", "-b:a", "64k", str(path)],
+        check=True, capture_output=True,
+    )
+
+
+def _session(tmp_path: Path, file_path: Path, gek: bytes, media_cache: MediaCache):
+    file_bytes = file_path.read_bytes()
+    file_id = blake3.blake3(file_bytes).hexdigest()
+
+    sk_node = Ed25519PrivateKey.generate()
+    index = GroupIndex(group_id="g" * 32, sk_node=sk_node, gek=gek)
+    index.add_entry(IndexEntry(
+        id=file_id, name=file_path.name, path=file_path.parent.name,
+        size=len(file_bytes), type="audio", added_at=0))
+
+    session = WebRTCPeerSession.__new__(WebRTCPeerSession)
+    session._ctx = {
+        "roots": one_root(file_path.parent),
+        "index": index,
+        "gek": gek,
+        "sk_node": sk_node,
+        "media_cache": media_cache,
+        "max_concurrent_streams": 4,
+    }
+    session._group_id = None
+    session._user_id = "tester"
+    session.sent = []
+    session._send = session.sent.append
+    session._audit = lambda *a, **k: None
+    return session, file_id
+
+
+def _reassemble_file_chunks(sent: list[dict], gek: bytes, hash_hex: str) -> bytes:
+    file_hash = bytes.fromhex(hash_hex)
+    chunks = sorted(
+        (m for m in sent if m.get("type") == "file_chunk" and m.get("file_id") == hash_hex),
+        key=lambda m: m["chunk_index"])
+    out = b""
+    for m in chunks:
+        key = chunk_key_aes(gek, file_hash, m["chunk_index"])
+        out += decrypt_chunk_aes(key, m["nonce"], m["ct"])
+    return out
+
+
+@pytest.mark.skipif(not _HAVE_FFMPEG, reason="ffmpeg/ffprobe not installed")
+async def test_wma_transcodes_to_a_browser_playable_aac_file(tmp_path, media_cache):
+    clip = tmp_path / "clip.wma"
+    _make_wma_clip(clip)
+    gek = generate_gek()
+    session, file_id = _session(tmp_path, clip, gek, media_cache)
+
+    await session._do_audio_transcode_request({"file_id": file_id})
+
+    resp = next(m for m in session.sent if m.get("type") == MNP.AUDIO_TRANSCODE_RESP)
+    assert resp["mime"] == "audio/mp4"
+    assert resp["size"] > 0
+    transcode_hash = resp["hash"]
+
+    # Pull it back exactly the way a client would: file_req/chunk, resolved
+    # against the media cache since this hash is not a real index entry.
+    await session._do_file_request({"file_id": transcode_hash, "chunk_index": 0})
+    out_bytes = _reassemble_file_chunks(session.sent, gek, transcode_hash)
+    assert len(out_bytes) == resp["size"]
+
+    out_path = tmp_path / "out.m4a"
+    out_path.write_bytes(out_bytes)
+    probe = subprocess.run(
+        ["ffprobe", "-v", "error", "-show_entries", "stream=codec_name,codec_type",
+         "-of", "csv=p=0", str(out_path)],
+        check=True, capture_output=True, text=True)
+    rows = [line.split(",") for line in probe.stdout.strip().splitlines()]
+    codecs = {r[1]: r[0] for r in rows}
+    assert codecs.get("audio") == "aac", f"must be AAC, playable in a browser: {codecs}"
+
+
+@pytest.mark.skipif(not _HAVE_FFMPEG, reason="ffmpeg/ffprobe not installed")
+async def test_second_request_reuses_the_cached_transcode(tmp_path, media_cache, monkeypatch):
+    clip = tmp_path / "clip.wma"
+    _make_wma_clip(clip)
+    gek = generate_gek()
+    session, file_id = _session(tmp_path, clip, gek, media_cache)
+
+    calls = {"n": 0}
+    real = wrs._transcode_audio_to_aac
+
+    async def counting(path):
+        calls["n"] += 1
+        return await real(path)
+
+    monkeypatch.setattr(wrs, "_transcode_audio_to_aac", counting)
+
+    await session._do_audio_transcode_request({"file_id": file_id})
+    session.sent.clear()
+    await session._do_audio_transcode_request({"file_id": file_id})
+
+    resp = next(m for m in session.sent if m.get("type") == MNP.AUDIO_TRANSCODE_RESP)
+    assert resp["hash"]
+    assert calls["n"] == 1, "a second request for the same file must not re-run ffmpeg"
+
+
+async def test_missing_file_id_is_an_error(tmp_path, media_cache):
+    gek = generate_gek()
+    sk_node = Ed25519PrivateKey.generate()
+    session = WebRTCPeerSession.__new__(WebRTCPeerSession)
+    session._ctx = {
+        "roots": one_root(tmp_path), "index": GroupIndex(group_id="g" * 32, sk_node=sk_node, gek=gek),
+        "gek": gek, "sk_node": sk_node, "media_cache": media_cache,
+    }
+    session._group_id = None
+    session.sent = []
+    session._send = session.sent.append
+
+    await session._do_audio_transcode_request({"file_id": "nonexistent"})
+
+    assert session.sent[-1]["type"] == "error"
+
+
+async def test_multi_chunk_cached_blob_reassembles_correctly(tmp_path, media_cache):
+    """
+    `_try_serve_thumbnail` used to assume a cached blob never exceeds one
+    chunk (true for a thumbnail, not true for a multi-MB audio transcode) —
+    this exercises the slicing directly, without needing ffmpeg at all: a
+    blob a little over two chunks, fetched chunk by chunk, must reassemble
+    to exactly the original bytes.
+    """
+    import os
+    gek = generate_gek()
+    sk_node = Ed25519PrivateKey.generate()
+    session = WebRTCPeerSession.__new__(WebRTCPeerSession)
+    session._ctx = {
+        "roots": one_root(tmp_path), "index": GroupIndex(group_id="g" * 32, sk_node=sk_node, gek=gek),
+        "gek": gek, "sk_node": sk_node, "media_cache": media_cache,
+    }
+    session._group_id = None
+    session.sent = []
+    session._send = session.sent.append
+
+    blob = os.urandom(int(wrs.CHUNK_SIZE * 2.3))
+    blob_hash = blake3.blake3(blob).hexdigest()
+    await media_cache.put_thumb(blob_hash, "synthetic:test", blob)
+
+    total_chunks = -(-len(blob) // wrs.CHUNK_SIZE)
+    for i in range(total_chunks):
+        await session._do_file_request({"file_id": blob_hash, "chunk_index": i})
+    # One past the end must be a clean miss, not a partial/garbage chunk.
+    await session._do_file_request({"file_id": blob_hash, "chunk_index": total_chunks})
+
+    chunk_msgs = [m for m in session.sent if m.get("type") == "file_chunk"]
+    assert len(chunk_msgs) == total_chunks
+    reassembled = _reassemble_file_chunks(session.sent, gek, blob_hash)
+    assert reassembled == blob