manifest.py
python
| 1 | """Hierarchical chunk manifests for the Muse MIDI plugin. |
| 2 | |
| 3 | Evolves the flat ``{"files": {"track.mid": "<sha256>"}}`` snapshot beyond |
| 4 | a single content hash per file to a rich, per-bar, per-track manifest that |
| 5 | enables: |
| 6 | |
| 7 | - **Partial diff** — compare only the bars that changed. |
| 8 | - **Query acceleration** — answer note queries without reading full MIDI blobs. |
| 9 | - **Targeted merge** — attempt to merge only the bars with conflicts. |
| 10 | - **Historical analytics** — aggregate statistics over commit history without |
| 11 | re-parsing MIDI bytes on every query. |
| 12 | |
| 13 | Backward compatibility |
| 14 | ---------------------- |
| 15 | The ``MusicManifest.files`` field is identical to the standard |
| 16 | ``SnapshotManifest.files`` so that the core Muse engine and all existing |
| 17 | commands continue to work unchanged. The ``tracks`` field is additive |
| 18 | metadata stored as a sidecar under ``.muse/music_manifests/`` — never |
| 19 | replacing the canonical flat manifest. |
| 20 | |
| 21 | Storage layout:: |
| 22 | |
| 23 | .muse/music_manifests/ |
| 24 | <snapshot_id>.json — full MusicManifest for this snapshot |
| 25 | (rebuildable from history; add to .museignore [domain.midi] in CI) |
| 26 | |
| 27 | Public API |
| 28 | ---------- |
| 29 | - :class:`BarChunk` — per-bar chunk descriptor. |
| 30 | - :class:`TrackManifest` — rich metadata for one MIDI track. |
| 31 | - :class:`MusicManifest` — top-level sidecar manifest. |
| 32 | - :func:`build_bar_chunk` — build a :class:`BarChunk` from a bar's notes. |
| 33 | - :func:`build_track_manifest` — build a :class:`TrackManifest`. |
| 34 | - :func:`build_music_manifest` — build the full :class:`MusicManifest`. |
| 35 | - :func:`write_music_manifest` — persist to ``.muse/music_manifests/``. |
| 36 | - :func:`read_music_manifest` — load from the sidecar store. |
| 37 | """ |
| 38 | |
| 39 | from __future__ import annotations |
| 40 | |
| 41 | import hashlib |
| 42 | import json |
| 43 | import logging |
| 44 | import pathlib |
| 45 | from typing import Literal, TypedDict |
| 46 | |
| 47 | from muse.plugins.midi._query import ( |
| 48 | NoteInfo, |
| 49 | detect_chord, |
| 50 | key_signature_guess, |
| 51 | notes_by_bar, |
| 52 | ) |
| 53 | from muse.plugins.midi.midi_diff import extract_notes |
| 54 | |
| 55 | logger = logging.getLogger(__name__) |
| 56 | |
| 57 | _MANIFEST_DIR = ".muse/music_manifests" |
| 58 | |
| 59 | |
| 60 | # --------------------------------------------------------------------------- |
| 61 | # Types |
| 62 | # --------------------------------------------------------------------------- |
| 63 | |
| 64 | |
| 65 | class BarChunk(TypedDict): |
| 66 | """Descriptor for one bar's worth of note events in a MIDI track. |
| 67 | |
| 68 | ``bar`` 1-indexed bar number (assumes 4/4 time). |
| 69 | ``chunk_hash`` SHA-256 of the canonical JSON of all notes in this bar. |
| 70 | Used for per-bar change detection without re-parsing MIDI. |
| 71 | ``note_count`` Number of notes in this bar. |
| 72 | ``chord`` Best-guess chord name for this bar (e.g. ``"Cmaj"``). |
| 73 | ``pitch_range`` ``[min_pitch, max_pitch]`` MIDI pitch values in this bar. |
| 74 | """ |
| 75 | |
| 76 | bar: int |
| 77 | chunk_hash: str |
| 78 | note_count: int |
| 79 | chord: str |
| 80 | pitch_range: list[int] |
| 81 | |
| 82 | |
| 83 | class TrackManifest(TypedDict): |
| 84 | """Rich metadata descriptor for one MIDI track at a specific snapshot. |
| 85 | |
| 86 | ``track_id`` Stable identifier for this track (SHA-256 of file path). |
| 87 | Stable across renames if you track by content; changes |
| 88 | on rename. Use entity IDs in the entity index for |
| 89 | true cross-rename continuity. |
| 90 | ``file_path`` Workspace-relative MIDI file path. |
| 91 | ``content_hash`` SHA-256 of the full MIDI file bytes (same as the flat |
| 92 | manifest entry — the canonical content address). |
| 93 | ``bars`` Mapping from ``str(bar_number)`` → :class:`BarChunk`. |
| 94 | JSON keys are always strings; callers convert to int. |
| 95 | ``ticks_per_beat`` MIDI ticks per quarter note for this file. |
| 96 | ``note_count`` Total note count across all bars. |
| 97 | ``key_guess`` Krumhansl-Schmuckler key estimate (e.g. ``"G major"``). |
| 98 | ``bar_count`` Number of bars with at least one note. |
| 99 | """ |
| 100 | |
| 101 | track_id: str |
| 102 | file_path: str |
| 103 | content_hash: str |
| 104 | bars: dict[str, BarChunk] |
| 105 | ticks_per_beat: int |
| 106 | note_count: int |
| 107 | key_guess: str |
| 108 | bar_count: int |
| 109 | |
| 110 | |
| 111 | class MusicManifest(TypedDict): |
| 112 | """Top-level hierarchical manifest for a music snapshot. |
| 113 | |
| 114 | This is the sidecar companion to the standard :class:`~muse.domain.SnapshotManifest`. |
| 115 | The ``files`` field is identical to the flat manifest — the core engine |
| 116 | reads only ``files`` for content addressing. The ``tracks`` field is |
| 117 | additive richness for music-domain queries, diff, and merge. |
| 118 | |
| 119 | ``schema_version`` Always ``2`` for this format. |
| 120 | ``snapshot_id`` The snapshot this manifest belongs to. |
| 121 | ``files`` Standard flat ``{path: sha256}`` manifest (compat layer). |
| 122 | ``tracks`` ``{path: TrackManifest}`` for each MIDI file. |
| 123 | """ |
| 124 | |
| 125 | domain: Literal["midi"] |
| 126 | schema_version: Literal[2] |
| 127 | snapshot_id: str |
| 128 | files: dict[str, str] |
| 129 | tracks: dict[str, TrackManifest] |
| 130 | |
| 131 | |
| 132 | # --------------------------------------------------------------------------- |
| 133 | # Builders |
| 134 | # --------------------------------------------------------------------------- |
| 135 | |
| 136 | |
| 137 | def _bar_chunk_hash(notes: list[NoteInfo]) -> str: |
| 138 | """Return a SHA-256 of the canonical JSON of a bar's notes.""" |
| 139 | payload = json.dumps( |
| 140 | [ |
| 141 | { |
| 142 | "pitch": n.pitch, |
| 143 | "velocity": n.velocity, |
| 144 | "start_tick": n.start_tick, |
| 145 | "duration_ticks": n.duration_ticks, |
| 146 | "channel": n.channel, |
| 147 | } |
| 148 | for n in sorted(notes, key=lambda n: (n.start_tick, n.pitch)) |
| 149 | ], |
| 150 | sort_keys=True, |
| 151 | separators=(",", ":"), |
| 152 | ).encode() |
| 153 | return hashlib.sha256(payload).hexdigest() |
| 154 | |
| 155 | |
| 156 | def build_bar_chunk(bar_num: int, notes: list[NoteInfo]) -> BarChunk: |
| 157 | """Build a :class:`BarChunk` descriptor for *bar_num*. |
| 158 | |
| 159 | Args: |
| 160 | bar_num: 1-indexed bar number. |
| 161 | notes: All notes in this bar. |
| 162 | |
| 163 | Returns: |
| 164 | A populated :class:`BarChunk`. |
| 165 | """ |
| 166 | pcs = frozenset(n.pitch_class for n in notes) |
| 167 | chord = detect_chord(pcs) |
| 168 | pitches = [n.pitch for n in notes] |
| 169 | pitch_range: list[int] = [min(pitches), max(pitches)] if pitches else [0, 0] |
| 170 | return BarChunk( |
| 171 | bar=bar_num, |
| 172 | chunk_hash=_bar_chunk_hash(notes), |
| 173 | note_count=len(notes), |
| 174 | chord=chord, |
| 175 | pitch_range=pitch_range, |
| 176 | ) |
| 177 | |
| 178 | |
| 179 | def build_track_manifest( |
| 180 | notes: list[NoteInfo], |
| 181 | file_path: str, |
| 182 | content_hash: str, |
| 183 | ticks_per_beat: int, |
| 184 | ) -> TrackManifest: |
| 185 | """Build a :class:`TrackManifest` from a parsed note list. |
| 186 | |
| 187 | Args: |
| 188 | notes: All notes extracted from the MIDI file. |
| 189 | file_path: Workspace-relative MIDI file path. |
| 190 | content_hash: SHA-256 of the MIDI file bytes (from the flat manifest). |
| 191 | ticks_per_beat: MIDI timing resolution. |
| 192 | |
| 193 | Returns: |
| 194 | A populated :class:`TrackManifest`. |
| 195 | """ |
| 196 | track_id = hashlib.sha256(file_path.encode()).hexdigest() |
| 197 | bars_map = notes_by_bar(notes) |
| 198 | bars: dict[str, BarChunk] = {} |
| 199 | for bar_num, bar_notes in sorted(bars_map.items()): |
| 200 | bars[str(bar_num)] = build_bar_chunk(bar_num, bar_notes) |
| 201 | |
| 202 | key_guess = key_signature_guess(notes) |
| 203 | |
| 204 | return TrackManifest( |
| 205 | track_id=track_id, |
| 206 | file_path=file_path, |
| 207 | content_hash=content_hash, |
| 208 | bars=bars, |
| 209 | ticks_per_beat=ticks_per_beat, |
| 210 | note_count=len(notes), |
| 211 | key_guess=key_guess, |
| 212 | bar_count=len(bars), |
| 213 | ) |
| 214 | |
| 215 | |
| 216 | def build_music_manifest( |
| 217 | file_manifest: dict[str, str], |
| 218 | root: "pathlib.Path", |
| 219 | snapshot_id: str = "", |
| 220 | ) -> MusicManifest: |
| 221 | """Build a :class:`MusicManifest` from a flat file manifest. |
| 222 | |
| 223 | For each ``.mid`` file in *file_manifest*, loads the MIDI bytes from the |
| 224 | object store, parses notes, and builds a :class:`TrackManifest`. |
| 225 | |
| 226 | Non-MIDI files appear only in ``files`` — they have no ``TrackManifest``. |
| 227 | |
| 228 | Args: |
| 229 | file_manifest: Standard ``{path: sha256}`` manifest. |
| 230 | root: Repository root for object store access. |
| 231 | snapshot_id: The snapshot ID this manifest belongs to. |
| 232 | |
| 233 | Returns: |
| 234 | A populated :class:`MusicManifest`. Tracks whose MIDI bytes are |
| 235 | unreadable or unparseable are silently omitted from ``tracks`` but |
| 236 | still appear in ``files``. |
| 237 | """ |
| 238 | import pathlib as _pathlib |
| 239 | from muse.core.object_store import read_object |
| 240 | |
| 241 | tracks: dict[str, TrackManifest] = {} |
| 242 | |
| 243 | for path, content_hash in sorted(file_manifest.items()): |
| 244 | if not path.lower().endswith(".mid"): |
| 245 | continue |
| 246 | raw = read_object(root, content_hash) |
| 247 | if raw is None: |
| 248 | logger.debug("⚠️ Object %s for %r not in store — skipping manifest", content_hash[:8], path) |
| 249 | continue |
| 250 | try: |
| 251 | keys, tpb = extract_notes(raw) |
| 252 | except ValueError as exc: |
| 253 | logger.debug("⚠️ Cannot parse MIDI %r: %s — skipping manifest", path, exc) |
| 254 | continue |
| 255 | notes = [NoteInfo.from_note_key(k, tpb) for k in keys] |
| 256 | tracks[path] = build_track_manifest(notes, path, content_hash, tpb) |
| 257 | |
| 258 | return MusicManifest( |
| 259 | domain="midi", |
| 260 | schema_version=2, |
| 261 | snapshot_id=snapshot_id, |
| 262 | files=dict(file_manifest), |
| 263 | tracks=tracks, |
| 264 | ) |
| 265 | |
| 266 | |
| 267 | # --------------------------------------------------------------------------- |
| 268 | # Persistence |
| 269 | # --------------------------------------------------------------------------- |
| 270 | |
| 271 | |
| 272 | def _manifest_path(repo_root: pathlib.Path, snapshot_id: str) -> pathlib.Path: |
| 273 | return repo_root / _MANIFEST_DIR / f"{snapshot_id}.json" |
| 274 | |
| 275 | |
| 276 | def write_music_manifest( |
| 277 | repo_root: pathlib.Path, |
| 278 | manifest: MusicManifest, |
| 279 | ) -> pathlib.Path: |
| 280 | """Persist *manifest* to ``.muse/music_manifests/<snapshot_id>.json``. |
| 281 | |
| 282 | Args: |
| 283 | repo_root: Repository root. |
| 284 | manifest: The manifest to write. |
| 285 | |
| 286 | Returns: |
| 287 | Path to the written file. |
| 288 | """ |
| 289 | snapshot_id = manifest.get("snapshot_id", "") |
| 290 | if not snapshot_id: |
| 291 | raise ValueError("MusicManifest.snapshot_id must be non-empty") |
| 292 | path = _manifest_path(repo_root, snapshot_id) |
| 293 | path.parent.mkdir(parents=True, exist_ok=True) |
| 294 | path.write_text(json.dumps(manifest, indent=2) + "\n") |
| 295 | logger.debug( |
| 296 | "✅ Music manifest written: %s (%d tracks)", |
| 297 | snapshot_id[:8], |
| 298 | len(manifest["tracks"]), |
| 299 | ) |
| 300 | return path |
| 301 | |
| 302 | |
| 303 | def read_music_manifest( |
| 304 | repo_root: pathlib.Path, |
| 305 | snapshot_id: str, |
| 306 | ) -> MusicManifest | None: |
| 307 | """Load the music manifest for *snapshot_id*, or ``None`` if absent. |
| 308 | |
| 309 | Args: |
| 310 | repo_root: Repository root. |
| 311 | snapshot_id: Snapshot ID. |
| 312 | |
| 313 | Returns: |
| 314 | The :class:`MusicManifest`, or ``None`` when the sidecar file does |
| 315 | not exist. |
| 316 | """ |
| 317 | path = _manifest_path(repo_root, snapshot_id) |
| 318 | if not path.exists(): |
| 319 | return None |
| 320 | try: |
| 321 | raw: MusicManifest = json.loads(path.read_text()) |
| 322 | return raw |
| 323 | except (json.JSONDecodeError, KeyError) as exc: |
| 324 | logger.warning("⚠️ Corrupt music manifest %s: %s", path, exc) |
| 325 | return None |
| 326 | |
| 327 | |
| 328 | # --------------------------------------------------------------------------- |
| 329 | # Partial diff helper |
| 330 | # --------------------------------------------------------------------------- |
| 331 | |
| 332 | |
| 333 | def diff_manifests_by_bar( |
| 334 | base: MusicManifest, |
| 335 | target: MusicManifest, |
| 336 | ) -> dict[str, list[int]]: |
| 337 | """Return a per-track list of bars that changed between two manifests. |
| 338 | |
| 339 | Uses the per-bar ``chunk_hash`` values to detect changes without |
| 340 | loading any MIDI bytes. |
| 341 | |
| 342 | Args: |
| 343 | base: Ancestor manifest. |
| 344 | target: Newer manifest. |
| 345 | |
| 346 | Returns: |
| 347 | ``{track_path: [changed_bar_numbers]}`` for all tracks where at |
| 348 | least one bar differs. Tracks added or removed appear with ``[-1]`` |
| 349 | as a sentinel indicating the whole track changed. |
| 350 | """ |
| 351 | changed: dict[str, list[int]] = {} |
| 352 | |
| 353 | all_tracks = set(base["tracks"]) | set(target["tracks"]) |
| 354 | |
| 355 | for track in sorted(all_tracks): |
| 356 | base_track = base["tracks"].get(track) |
| 357 | target_track = target["tracks"].get(track) |
| 358 | |
| 359 | if base_track is None or target_track is None: |
| 360 | changed[track] = [-1] |
| 361 | continue |
| 362 | |
| 363 | if base_track["content_hash"] == target_track["content_hash"]: |
| 364 | continue |
| 365 | |
| 366 | # Content changed — find which bars. |
| 367 | base_bars = base_track["bars"] |
| 368 | target_bars = target_track["bars"] |
| 369 | all_bar_keys = set(base_bars) | set(target_bars) |
| 370 | |
| 371 | changed_bars: list[int] = [] |
| 372 | for bar_key in sorted(all_bar_keys, key=lambda k: int(k)): |
| 373 | base_chunk = base_bars.get(bar_key) |
| 374 | target_chunk = target_bars.get(bar_key) |
| 375 | if base_chunk is None or target_chunk is None: |
| 376 | changed_bars.append(int(bar_key)) |
| 377 | elif base_chunk["chunk_hash"] != target_chunk["chunk_hash"]: |
| 378 | changed_bars.append(int(bar_key)) |
| 379 | |
| 380 | if changed_bars: |
| 381 | changed[track] = sorted(changed_bars) |
| 382 | |
| 383 | return changed |