migrate.py
python
sha256:707b4461365efd52a6360948db35151dd01ccdfbb8e5c62d6ed14afca7a0dc35
migrate_blob_ids now correctly skips snapshot and commit ob…
Human
patch
43 days ago
| 1 | """Full layout and commit-ID migration engine. |
| 2 | |
| 3 | Migrates any Muse repository from legacy on-disk layouts to the canonical |
| 4 | current layout: |
| 5 | |
| 6 | Old layout Canonical layout |
| 7 | -------------------------------------- --------------------------------------- |
| 8 | objects/<shard>/<rest> objects/sha256/<shard>/<rest> |
| 9 | commits/<hex>.msgpack commits/sha256/<hex>.msgpack |
| 10 | snapshots/<hex>.msgpack snapshots/sha256/<hex>.msgpack |
| 11 | branch ref: bare hex branch ref: sha256:<hex> |
| 12 | remote ref: bare hex or stale ID remote ref: sha256:<current-hex> |
| 13 | repo_id: plain string (pre-sha256) repo_id: sha256:<hex> |
| 14 | commit field: created_on_branch commit field: branch |
| 15 | commit field: format_version < 8 commit field: format_version = 8 |
| 16 | commit signature: bare base64url commit signature: ed25519:<base64url> |
| 17 | commit_id: legacy formula commit_id: hash_commit(...) |
| 18 | |
| 19 | Entry points |
| 20 | ------------ |
| 21 | - :func:`migrate` — full execution or dry-run |
| 22 | - :class:`MigrateResult` — returned by :func:`migrate` |
| 23 | """ |
| 24 | |
| 25 | import collections |
| 26 | import datetime |
| 27 | import json |
| 28 | import logging |
| 29 | import pathlib |
| 30 | import shutil |
| 31 | from dataclasses import dataclass, field |
| 32 | from typing import TYPE_CHECKING |
| 33 | |
| 34 | import msgpack |
| 35 | |
| 36 | from muse.core.paths import ( |
| 37 | commits_dir, heads_dir, logs_dir, muse_dir, |
| 38 | objects_dir, rebase_merge_dir, remotes_dir, |
| 39 | repo_json_path, snapshots_dir, |
| 40 | ) |
| 41 | from muse.core.ids import hash_blob, hash_commit, hash_snapshot |
| 42 | from muse.core.types import b64url_decode, blob_id, encode_sig, long_id |
| 43 | from muse.core.refs import get_all_branch_heads, write_branch_ref |
| 44 | |
| 45 | if TYPE_CHECKING: |
| 46 | from muse.core.transport import SigningIdentity |
| 47 | |
| 48 | logger = logging.getLogger(__name__) |
| 49 | |
| 50 | type _IdMap = dict[str, str] |
| 51 | type _RawCommit = dict[str, str | int | float | bytes | datetime.datetime | None] |
| 52 | |
| 53 | |
| 54 | # --------------------------------------------------------------------------- |
| 55 | # Phase 7 — blob ID migration result |
| 56 | # --------------------------------------------------------------------------- |
| 57 | |
| 58 | @dataclass |
| 59 | class BlobMigrateResult: |
| 60 | """Result of :func:`migrate_blob_ids`.""" |
| 61 | id_map: dict[str, str] = field(default_factory=dict) |
| 62 | blobs_written: int = 0 |
| 63 | blobs_skipped: int = 0 |
| 64 | dry_run: bool = False |
| 65 | type _CommitEntry = tuple[_RawCommit, pathlib.Path] |
| 66 | type _CommitIndex = dict[str, _CommitEntry] |
| 67 | |
| 68 | # --------------------------------------------------------------------------- |
| 69 | # Result type |
| 70 | # --------------------------------------------------------------------------- |
| 71 | |
| 72 | @dataclass |
| 73 | class MigrateResult: |
| 74 | id_map: dict[str, str] = field(default_factory=dict) |
| 75 | commits_rewritten: int = 0 |
| 76 | blobs_migrated: int = 0 |
| 77 | legacy_dirs_removed: int = 0 |
| 78 | dry_run: bool = False |
| 79 | commits_relocated: int = 0 |
| 80 | snapshots_relocated: int = 0 |
| 81 | refs_updated: int = 0 |
| 82 | remote_refs_updated: int = 0 |
| 83 | repo_id_updated: bool = False |
| 84 | branch_fields_renamed: int = 0 |
| 85 | signatures_normalised: int = 0 |
| 86 | format_versions_bumped: int = 0 |
| 87 | reflogs_updated: int = 0 |
| 88 | commits_signed: int = 0 |
| 89 | unsigned_commits_skipped: int = 0 |
| 90 | # Phase 7 — unified object store backfill |
| 91 | blobs_rewritten: int = 0 |
| 92 | snapshots_rewritten: int = 0 |
| 93 | commits_unified: int = 0 |
| 94 | |
| 95 | # --------------------------------------------------------------------------- |
| 96 | # Internal helpers |
| 97 | # --------------------------------------------------------------------------- |
| 98 | |
| 99 | def _normalise_sig(sig: str) -> str: |
| 100 | if not sig or sig.startswith("ed25519:"): |
| 101 | return sig |
| 102 | raw = b64url_decode(sig) |
| 103 | return encode_sig("ed25519", raw) |
| 104 | |
| 105 | def _committed_at_iso(raw: _RawCommit) -> str: |
| 106 | val = raw.get("committed_at", "") |
| 107 | if isinstance(val, datetime.datetime): |
| 108 | return val.isoformat() |
| 109 | return str(val) |
| 110 | |
| 111 | def _collect_all_commits(root: pathlib.Path) -> _CommitIndex: |
| 112 | """Read every commit file in the store. |
| 113 | |
| 114 | Returns a mapping of commit_id → (raw_dict, file_path). Covers: |
| 115 | - Legacy msgpack files in ``.muse/commits/sha256/`` |
| 116 | - Unified object store entries in ``.muse/objects/sha256/`` with a |
| 117 | ``commit <size>\\0<json>`` header |
| 118 | """ |
| 119 | result: _CommitIndex = {} |
| 120 | |
| 121 | # Legacy msgpack commits |
| 122 | _dir = commits_dir(root) |
| 123 | if _dir.exists(): |
| 124 | for path in _dir.rglob("*.msgpack"): |
| 125 | try: |
| 126 | raw = msgpack.unpackb(path.read_bytes(), raw=False) |
| 127 | cid = raw.get("commit_id", "") |
| 128 | if cid: |
| 129 | result[cid] = (raw, path) |
| 130 | except Exception as exc: |
| 131 | logger.warning("Skipping unreadable commit %s: %s", path, exc) |
| 132 | |
| 133 | # Unified object store commits (commit <size>\0<json>) |
| 134 | obj_dir = objects_dir(root) / "sha256" |
| 135 | if obj_dir.exists(): |
| 136 | for path in obj_dir.glob("*/*"): |
| 137 | if not path.is_file(): |
| 138 | continue |
| 139 | try: |
| 140 | data = path.read_bytes() |
| 141 | if not data.startswith(b"commit "): |
| 142 | continue |
| 143 | nl = data.index(b"\x00") |
| 144 | raw = json.loads(data[nl + 1:]) |
| 145 | cid = raw.get("commit_id", "") |
| 146 | if cid and cid not in result: |
| 147 | result[cid] = (raw, path) |
| 148 | except Exception as exc: |
| 149 | logger.warning("Skipping unreadable object %s: %s", path, exc) |
| 150 | |
| 151 | return result |
| 152 | |
| 153 | def _topological_sort(commits: _CommitIndex) -> list[str]: |
| 154 | """Return commit IDs in topological order (parents before children).""" |
| 155 | in_degree: dict[str, int] = {cid: 0 for cid in commits} |
| 156 | children: dict[str, list[str]] = {cid: [] for cid in commits} |
| 157 | |
| 158 | for cid, (raw, _) in commits.items(): |
| 159 | for key in ("parent_commit_id", "parent2_commit_id"): |
| 160 | pid = raw.get(key) |
| 161 | if pid and pid in commits: |
| 162 | in_degree[cid] = in_degree.get(cid, 0) + 1 |
| 163 | children[pid].append(cid) |
| 164 | |
| 165 | queue: collections.deque[str] = collections.deque( |
| 166 | sorted(cid for cid, deg in in_degree.items() if deg == 0) |
| 167 | ) |
| 168 | order: list[str] = [] |
| 169 | while queue: |
| 170 | cid = queue.popleft() |
| 171 | order.append(cid) |
| 172 | for child in sorted(children.get(cid, [])): |
| 173 | in_degree[child] -= 1 |
| 174 | if in_degree[child] == 0: |
| 175 | queue.append(child) |
| 176 | return order |
| 177 | |
| 178 | def _make_canonical_dict( |
| 179 | raw: _RawCommit, |
| 180 | id_map: _IdMap, |
| 181 | signing_identity: SigningIdentity | None = None, |
| 182 | force_resign: bool = False, |
| 183 | ) -> tuple[_RawCommit, str, bool]: |
| 184 | """Build the canonical version of a raw commit dict. |
| 185 | |
| 186 | Returns ``(canonical_dict, new_commit_id, was_signed)``. |
| 187 | ``was_signed`` is True when a new signature was created (unsigned commit, |
| 188 | parent cascade, or ``force_resign``). |
| 189 | """ |
| 190 | p1 = raw.get("parent_commit_id") |
| 191 | p2 = raw.get("parent2_commit_id") |
| 192 | new_p1 = id_map.get(p1, p1) if p1 else None |
| 193 | new_p2 = id_map.get(p2, p2) if p2 else None |
| 194 | |
| 195 | parent_ids: list[str] = [] |
| 196 | if new_p1: |
| 197 | parent_ids.append(new_p1) |
| 198 | if new_p2: |
| 199 | parent_ids.append(new_p2) |
| 200 | |
| 201 | snapshot_id: str = raw.get("snapshot_id", "") |
| 202 | message: str = raw.get("message", "") |
| 203 | committed_at_iso = _committed_at_iso(raw) |
| 204 | author: str = raw.get("author", "") |
| 205 | |
| 206 | parents_changed = (new_p1 != p1) or (new_p2 != p2) |
| 207 | is_unsigned = not raw.get("signature") and not raw.get("signer_public_key") |
| 208 | # Re-sign when: (a) commit was unsigned, (b) parent IDs changed so the |
| 209 | # old signature covers the wrong commit_id, or (c) force_resign is set. |
| 210 | should_sign = signing_identity is not None and (is_unsigned or parents_changed or force_resign) |
| 211 | |
| 212 | # Normalise empty author to the signing identity's handle so every |
| 213 | # resigned commit carries proper attribution baked into its content hash. |
| 214 | if should_sign and signing_identity is not None and not author: |
| 215 | author = signing_identity.handle |
| 216 | |
| 217 | if should_sign: |
| 218 | from muse.core.provenance import encode_public_key, provenance_payload, sign_commit_ed25519 |
| 219 | _, pub_str = encode_public_key(signing_identity.private_key) # type: ignore[arg-type] |
| 220 | new_id = hash_commit( |
| 221 | parent_ids=parent_ids, |
| 222 | snapshot_id=snapshot_id, |
| 223 | message=message, |
| 224 | committed_at_iso=committed_at_iso, |
| 225 | author=author, |
| 226 | signer_public_key=pub_str, |
| 227 | ) |
| 228 | payload = provenance_payload( |
| 229 | commit_id=new_id, |
| 230 | author=author, |
| 231 | agent_id=raw.get("agent_id", ""), |
| 232 | model_id=raw.get("model_id", ""), |
| 233 | toolchain_id=raw.get("toolchain_id", ""), |
| 234 | prompt_hash=raw.get("prompt_hash", ""), |
| 235 | committed_at=committed_at_iso, |
| 236 | ) |
| 237 | sig = sign_commit_ed25519(payload, signing_identity.private_key) # type: ignore[arg-type] |
| 238 | else: |
| 239 | pub_str = raw.get("signer_public_key", "") |
| 240 | new_id = hash_commit( |
| 241 | parent_ids=parent_ids, |
| 242 | snapshot_id=snapshot_id, |
| 243 | message=message, |
| 244 | committed_at_iso=committed_at_iso, |
| 245 | author=author, |
| 246 | signer_public_key=pub_str, |
| 247 | ) |
| 248 | sig = _normalise_sig(raw.get("signature", "")) |
| 249 | |
| 250 | branch_val = raw.get("branch") or raw.get("created_on_branch") or "main" |
| 251 | |
| 252 | out = dict(raw) |
| 253 | out["commit_id"] = new_id |
| 254 | out["parent_commit_id"] = new_p1 |
| 255 | out["parent2_commit_id"] = new_p2 |
| 256 | out["author"] = author |
| 257 | out["signature"] = sig |
| 258 | out["branch"] = branch_val |
| 259 | out["format_version"] = 8 |
| 260 | out["committed_at"] = committed_at_iso |
| 261 | if should_sign: |
| 262 | out["signer_public_key"] = pub_str |
| 263 | out.pop("created_on_branch", None) |
| 264 | out.pop("repo_id", None) |
| 265 | |
| 266 | return out, new_id, should_sign |
| 267 | |
| 268 | def _canonical_commit_path(root: pathlib.Path, hex_id: str) -> pathlib.Path: |
| 269 | return commits_dir(root) / "sha256" / f"{hex_id}.msgpack" |
| 270 | |
| 271 | def _write_raw_commit(root: pathlib.Path, canonical: _RawCommit) -> None: |
| 272 | hex_id = long_id(canonical["commit_id"], strip=True) |
| 273 | path = _canonical_commit_path(root, hex_id) |
| 274 | path.parent.mkdir(parents=True, exist_ok=True) |
| 275 | path.write_bytes(msgpack.packb(canonical, use_bin_type=True)) |
| 276 | |
| 277 | def _is_flat_commit_path(root: pathlib.Path, path: pathlib.Path) -> bool: |
| 278 | _dir = commits_dir(root) |
| 279 | try: |
| 280 | rel = path.relative_to(_dir) |
| 281 | except ValueError: |
| 282 | return False |
| 283 | return rel.parts[0] != "sha256" |
| 284 | |
| 285 | def _migrate_objects(root: pathlib.Path, dry_run: bool) -> tuple[int, int]: |
| 286 | _dir = objects_dir(root) |
| 287 | if not _dir.exists(): |
| 288 | return 0, 0 |
| 289 | |
| 290 | blobs_moved = 0 |
| 291 | dirs_removed = 0 |
| 292 | |
| 293 | for entry in sorted(_dir.iterdir()): |
| 294 | if not entry.is_dir() or entry.name == "sha256": |
| 295 | continue |
| 296 | prefix = entry.name |
| 297 | for blob_path in sorted(entry.iterdir()): |
| 298 | if not blob_path.is_file(): |
| 299 | continue |
| 300 | rest = blob_path.name |
| 301 | canonical = _dir / "sha256" / prefix / rest |
| 302 | if canonical.exists(): |
| 303 | if not dry_run: |
| 304 | blob_path.unlink() |
| 305 | else: |
| 306 | if not dry_run: |
| 307 | canonical.parent.mkdir(parents=True, exist_ok=True) |
| 308 | blob_path.rename(canonical) |
| 309 | blobs_moved += 1 |
| 310 | if not dry_run and entry.is_dir() and not any(entry.iterdir()): |
| 311 | entry.rmdir() |
| 312 | dirs_removed += 1 |
| 313 | |
| 314 | return blobs_moved, dirs_removed |
| 315 | |
| 316 | def _migrate_snapshots(root: pathlib.Path, dry_run: bool) -> int: |
| 317 | snaps_dir = snapshots_dir(root) |
| 318 | if not snaps_dir.exists(): |
| 319 | return 0 |
| 320 | count = 0 |
| 321 | for path in sorted(snaps_dir.iterdir()): |
| 322 | if not path.is_file() or not path.name.endswith(".msgpack"): |
| 323 | continue |
| 324 | hex_id = path.stem |
| 325 | canonical = snaps_dir / "sha256" / f"{hex_id}.msgpack" |
| 326 | if canonical.exists(): |
| 327 | if not dry_run: |
| 328 | path.unlink() |
| 329 | else: |
| 330 | if not dry_run: |
| 331 | canonical.parent.mkdir(parents=True, exist_ok=True) |
| 332 | path.rename(canonical) |
| 333 | count += 1 |
| 334 | return count |
| 335 | |
| 336 | def _normalise_ref_value(val: str, id_map: _IdMap) -> str: |
| 337 | val = val.strip() |
| 338 | val = long_id(val) |
| 339 | return id_map.get(val, val) |
| 340 | |
| 341 | def _migrate_refs(root: pathlib.Path, id_map: _IdMap, dry_run: bool) -> int: |
| 342 | refs_dir = heads_dir(root) |
| 343 | if not refs_dir.exists(): |
| 344 | return 0 |
| 345 | count = 0 |
| 346 | for path in sorted(refs_dir.rglob("*")): |
| 347 | if not path.is_file(): |
| 348 | continue |
| 349 | val = path.read_text(encoding="utf-8").strip() |
| 350 | new_val = _normalise_ref_value(val, id_map) |
| 351 | if new_val != val: |
| 352 | count += 1 |
| 353 | if not dry_run: |
| 354 | path.write_text(f"{new_val}\n", encoding="utf-8") |
| 355 | return count |
| 356 | |
| 357 | def _migrate_remote_refs( |
| 358 | root: pathlib.Path, id_map: dict[str, str], dry_run: bool |
| 359 | ) -> int: |
| 360 | _dir = remotes_dir(root) |
| 361 | if not _dir.exists(): |
| 362 | return 0 |
| 363 | count = 0 |
| 364 | for path in sorted(_dir.rglob("*")): |
| 365 | if not path.is_file(): |
| 366 | continue |
| 367 | val = path.read_text(encoding="utf-8").strip() |
| 368 | new_val = _normalise_ref_value(val, id_map) |
| 369 | if new_val != val: |
| 370 | count += 1 |
| 371 | if not dry_run: |
| 372 | path.write_text(f"{new_val}\n", encoding="utf-8") |
| 373 | return count |
| 374 | |
| 375 | def _migrate_repo_id(root: pathlib.Path, dry_run: bool) -> bool: |
| 376 | repo_json = repo_json_path(root) |
| 377 | if not repo_json.exists(): |
| 378 | return False |
| 379 | data = json.loads(repo_json.read_text(encoding="utf-8")) |
| 380 | rid = data.get("repo_id", "") |
| 381 | if rid.startswith("sha256:"): |
| 382 | return False |
| 383 | new_id = blob_id(rid.encode()) |
| 384 | if not dry_run: |
| 385 | data["repo_id"] = new_id |
| 386 | repo_json.write_text(json.dumps(data), encoding="utf-8") |
| 387 | return True |
| 388 | |
| 389 | def _migrate_reflogs( |
| 390 | root: pathlib.Path, id_map: dict[str, str], dry_run: bool |
| 391 | ) -> int: |
| 392 | _dir = logs_dir(root) |
| 393 | if not _dir.exists(): |
| 394 | return 0 |
| 395 | count = 0 |
| 396 | for path in sorted(_dir.rglob("*")): |
| 397 | if not path.is_file(): |
| 398 | continue |
| 399 | content = path.read_text(encoding="utf-8") |
| 400 | new_content = content |
| 401 | for old_id, new_id in id_map.items(): |
| 402 | new_content = new_content.replace(old_id, new_id) |
| 403 | if new_content != content: |
| 404 | count += 1 |
| 405 | if not dry_run: |
| 406 | path.write_text(new_content, encoding="utf-8") |
| 407 | return count |
| 408 | |
| 409 | # --------------------------------------------------------------------------- |
| 410 | # Phase 7 — blob ID migration |
| 411 | # --------------------------------------------------------------------------- |
| 412 | |
| 413 | def migrate_blob_ids( |
| 414 | repo_root: pathlib.Path, |
| 415 | dry_run: bool = False, |
| 416 | ) -> BlobMigrateResult: |
| 417 | """Migrate raw (old-formula) blobs to muse-format at new hash_blob paths. |
| 418 | |
| 419 | For each blob stored with the pre-Phase-2 formula ``sha256(content)``, |
| 420 | this function: |
| 421 | |
| 422 | 1. Reads the raw bytes from the old object path. |
| 423 | 2. Detects whether the file is already muse-format (has a valid |
| 424 | ``"blob <size>\\0"`` header whose declared size matches the payload). |
| 425 | 3. If old-format: writes ``"blob <size>\\0<content>"`` to the new object |
| 426 | path derived from ``hash_blob(content)``. |
| 427 | 4. Builds and returns an old_id → new_id map for use by the snapshot |
| 428 | migration pass. |
| 429 | |
| 430 | Non-destructive: old files are never deleted. |
| 431 | |
| 432 | Args: |
| 433 | repo_root: Repository root (contains ``.muse/``). |
| 434 | dry_run: When True, compute the mapping but write nothing. |
| 435 | |
| 436 | Returns: |
| 437 | :class:`BlobMigrateResult` with the ID map and write counts. |
| 438 | """ |
| 439 | from muse.core.object_store import iter_stored_objects, object_path as _object_path |
| 440 | |
| 441 | result = BlobMigrateResult(dry_run=dry_run) |
| 442 | |
| 443 | for old_id, path in iter_stored_objects(repo_root): |
| 444 | data = path.read_bytes() |
| 445 | |
| 446 | # Detect typed objects: "blob|snapshot|commit <size>\0<payload>". |
| 447 | # Snapshot and commit objects must never be treated as blobs. |
| 448 | null_idx = data.find(b"\0") |
| 449 | if null_idx > 0: |
| 450 | header = data[:null_idx].decode(errors="replace") |
| 451 | parts = header.split(" ", 1) |
| 452 | if len(parts) == 2 and parts[1].isdigit() and len(data) - null_idx - 1 == int(parts[1]): |
| 453 | if parts[0] in ("snapshot", "commit"): |
| 454 | # Not a blob — skip entirely, never rewrite. |
| 455 | result.id_map[old_id] = old_id |
| 456 | result.blobs_skipped += 1 |
| 457 | continue |
| 458 | if parts[0] == "blob": |
| 459 | # Already muse-format blob — map to itself and skip. |
| 460 | result.id_map[old_id] = old_id |
| 461 | result.blobs_skipped += 1 |
| 462 | continue |
| 463 | |
| 464 | # Old-format raw blob: compute the new muse-idiomatic ID. |
| 465 | new_id = hash_blob(data) |
| 466 | result.id_map[old_id] = new_id |
| 467 | |
| 468 | if new_id == old_id: |
| 469 | result.blobs_skipped += 1 |
| 470 | continue |
| 471 | |
| 472 | new_path = _object_path(repo_root, new_id) |
| 473 | if new_path.exists(): |
| 474 | # New blob already written — delete the stale old-format file. |
| 475 | if not dry_run: |
| 476 | path.unlink(missing_ok=True) |
| 477 | result.blobs_skipped += 1 |
| 478 | continue |
| 479 | |
| 480 | if not dry_run: |
| 481 | header_bytes = f"blob {len(data)}\0".encode() |
| 482 | new_path.parent.mkdir(parents=True, exist_ok=True) |
| 483 | new_path.write_bytes(header_bytes + data) |
| 484 | # Remove the old-format file now that the new one is safely written. |
| 485 | path.unlink(missing_ok=True) |
| 486 | |
| 487 | result.blobs_written += 1 |
| 488 | |
| 489 | return result |
| 490 | |
| 491 | |
| 492 | # --------------------------------------------------------------------------- |
| 493 | # Phase 7 — snapshot ID migration |
| 494 | # --------------------------------------------------------------------------- |
| 495 | |
| 496 | @dataclass |
| 497 | class SnapshotMigrateResult: |
| 498 | """Result of :func:`migrate_snapshot_ids`.""" |
| 499 | id_map: dict[str, str] = field(default_factory=dict) |
| 500 | snapshots_written: int = 0 |
| 501 | snapshots_skipped: int = 0 |
| 502 | dry_run: bool = False |
| 503 | |
| 504 | |
| 505 | def migrate_snapshot_ids( |
| 506 | repo_root: pathlib.Path, |
| 507 | blob_id_map: dict[str, str], |
| 508 | dry_run: bool = False, |
| 509 | ) -> SnapshotMigrateResult: |
| 510 | """Migrate old-formula snapshots to muse-format in the unified object store. |
| 511 | |
| 512 | For each snapshot msgpack in ``.muse/snapshots/sha256/``: |
| 513 | |
| 514 | 1. Reads the record and updates every manifest entry using *blob_id_map* |
| 515 | (old blob ID → new blob ID). |
| 516 | 2. Recomputes the snapshot ID with the current ``hash_snapshot`` formula. |
| 517 | 3. Writes the updated record as JSON to ``.muse/objects/sha256/<2>/<62>`` |
| 518 | in ``"snapshot <size>\\0<json>"`` format. |
| 519 | |
| 520 | Non-destructive: old msgpack files are never deleted. |
| 521 | |
| 522 | Args: |
| 523 | repo_root: Repository root (contains ``.muse/``). |
| 524 | blob_id_map: Old blob ID → new blob ID map from :func:`migrate_blob_ids`. |
| 525 | dry_run: When True, compute the mapping but write nothing. |
| 526 | |
| 527 | Returns: |
| 528 | :class:`SnapshotMigrateResult` with the ID map and write counts. |
| 529 | """ |
| 530 | from muse.core.object_store import object_path as _object_path |
| 531 | |
| 532 | result = SnapshotMigrateResult(dry_run=dry_run) |
| 533 | snaps = snapshots_dir(repo_root) / "sha256" |
| 534 | if not snaps.exists(): |
| 535 | return result |
| 536 | |
| 537 | from muse.core.io import zstd_decompress_if_needed |
| 538 | |
| 539 | for path in sorted(snaps.glob("*.msgpack")): |
| 540 | try: |
| 541 | raw = msgpack.unpackb(zstd_decompress_if_needed(path.read_bytes()), raw=False) |
| 542 | except Exception as exc: |
| 543 | logger.warning("Skipping unreadable snapshot %s: %s", path, exc) |
| 544 | continue |
| 545 | |
| 546 | old_id: str = raw.get("snapshot_id", "") |
| 547 | if not old_id: |
| 548 | continue |
| 549 | |
| 550 | old_manifest: dict[str, str] = dict(raw.get("manifest") or {}) |
| 551 | new_manifest = { |
| 552 | file_path: blob_id_map.get(oid, oid) |
| 553 | for file_path, oid in old_manifest.items() |
| 554 | } |
| 555 | directories = raw.get("directories") or [] |
| 556 | new_snap_id = hash_snapshot(new_manifest, directories or None) |
| 557 | result.id_map[old_id] = new_snap_id |
| 558 | |
| 559 | new_path = _object_path(repo_root, new_snap_id) |
| 560 | if new_path.exists(): |
| 561 | result.snapshots_skipped += 1 |
| 562 | continue |
| 563 | |
| 564 | if not dry_run: |
| 565 | updated = dict(raw) |
| 566 | updated["snapshot_id"] = new_snap_id |
| 567 | updated["manifest"] = new_manifest |
| 568 | # created_at may be a datetime — normalise to ISO string for JSON. |
| 569 | if hasattr(updated.get("created_at"), "isoformat"): |
| 570 | updated["created_at"] = updated["created_at"].isoformat() |
| 571 | payload = json.dumps(updated, separators=(",", ":")).encode() |
| 572 | header = f"snapshot {len(payload)}\0".encode() |
| 573 | new_path.parent.mkdir(parents=True, exist_ok=True) |
| 574 | new_path.write_bytes(header + payload) |
| 575 | |
| 576 | result.snapshots_written += 1 |
| 577 | |
| 578 | return result |
| 579 | |
| 580 | |
| 581 | # --------------------------------------------------------------------------- |
| 582 | # Phase 7 — commit ID migration |
| 583 | # --------------------------------------------------------------------------- |
| 584 | |
| 585 | @dataclass |
| 586 | class CommitMigrateResult: |
| 587 | """Result of :func:`migrate_commit_ids`.""" |
| 588 | id_map: dict[str, str] = field(default_factory=dict) |
| 589 | commits_written: int = 0 |
| 590 | commits_skipped: int = 0 |
| 591 | dry_run: bool = False |
| 592 | |
| 593 | |
| 594 | def migrate_commit_ids( |
| 595 | repo_root: pathlib.Path, |
| 596 | snapshot_id_map: dict[str, str], |
| 597 | dry_run: bool = False, |
| 598 | ) -> CommitMigrateResult: |
| 599 | """Migrate old-formula commits to muse-format in the unified object store. |
| 600 | |
| 601 | For each commit msgpack in ``.muse/commits/sha256/``: |
| 602 | |
| 603 | 1. Reads the record and updates ``snapshot_id`` using *snapshot_id_map*. |
| 604 | 2. Updates parent commit IDs from the result's own ``id_map`` (two-pass: |
| 605 | all new IDs are computed first, then files are written). |
| 606 | 3. Recomputes the commit ID with the current ``hash_commit`` formula. |
| 607 | 4. Writes the updated record as JSON to ``.muse/objects/sha256/<2>/<62>`` |
| 608 | in ``"commit <size>\\0<json>"`` format. |
| 609 | |
| 610 | Non-destructive: old msgpack files are never deleted. |
| 611 | |
| 612 | Args: |
| 613 | repo_root: Repository root (contains ``.muse/``). |
| 614 | snapshot_id_map: Old snapshot ID → new snapshot ID map from |
| 615 | :func:`migrate_snapshot_ids`. |
| 616 | dry_run: When True, compute the mapping but write nothing. |
| 617 | |
| 618 | Returns: |
| 619 | :class:`CommitMigrateResult` with the ID map and write counts. |
| 620 | """ |
| 621 | from muse.core.object_store import object_path as _object_path |
| 622 | |
| 623 | result = CommitMigrateResult(dry_run=dry_run) |
| 624 | all_commits = _collect_all_commits(repo_root) |
| 625 | order = _topological_sort(all_commits) |
| 626 | |
| 627 | # Pass 1: build the commit id_map (parents-before-children order ensures |
| 628 | # parent IDs are already in id_map when processing each child). |
| 629 | commit_id_map: dict[str, str] = {} |
| 630 | for old_id in order: |
| 631 | raw, _ = all_commits[old_id] |
| 632 | p1 = raw.get("parent_commit_id") |
| 633 | p2 = raw.get("parent2_commit_id") |
| 634 | new_p1 = commit_id_map.get(p1, p1) if p1 else None |
| 635 | new_p2 = commit_id_map.get(p2, p2) if p2 else None |
| 636 | |
| 637 | parent_ids: list[str] = [x for x in (new_p1, new_p2) if x] |
| 638 | old_snap = raw.get("snapshot_id", "") |
| 639 | new_snap = snapshot_id_map.get(old_snap, old_snap) |
| 640 | committed_at_iso = _committed_at_iso(raw) |
| 641 | |
| 642 | new_id = hash_commit( |
| 643 | parent_ids=parent_ids, |
| 644 | snapshot_id=new_snap, |
| 645 | message=raw.get("message", ""), |
| 646 | committed_at_iso=committed_at_iso, |
| 647 | author=raw.get("author", ""), |
| 648 | signer_public_key=raw.get("signer_public_key", "") or "", |
| 649 | ) |
| 650 | if new_id != old_id: |
| 651 | commit_id_map[old_id] = new_id |
| 652 | |
| 653 | result.id_map = {**commit_id_map} |
| 654 | |
| 655 | # Pass 2: write updated records. |
| 656 | for old_id in order: |
| 657 | raw, _ = all_commits[old_id] |
| 658 | p1 = raw.get("parent_commit_id") |
| 659 | p2 = raw.get("parent2_commit_id") |
| 660 | new_p1 = commit_id_map.get(p1, p1) if p1 else None |
| 661 | new_p2 = commit_id_map.get(p2, p2) if p2 else None |
| 662 | |
| 663 | parent_ids = [x for x in (new_p1, new_p2) if x] |
| 664 | old_snap = raw.get("snapshot_id", "") |
| 665 | new_snap = snapshot_id_map.get(old_snap, old_snap) |
| 666 | new_id = commit_id_map.get(old_id, old_id) |
| 667 | committed_at_iso = _committed_at_iso(raw) |
| 668 | |
| 669 | obj_path = _object_path(repo_root, new_id) |
| 670 | if obj_path.exists(): |
| 671 | result.commits_skipped += 1 |
| 672 | continue |
| 673 | |
| 674 | if not dry_run: |
| 675 | updated = dict(raw) |
| 676 | updated["commit_id"] = new_id |
| 677 | updated["snapshot_id"] = new_snap |
| 678 | updated["parent_commit_id"] = new_p1 |
| 679 | updated["parent2_commit_id"] = new_p2 |
| 680 | updated["committed_at"] = committed_at_iso |
| 681 | updated.pop("created_on_branch", None) |
| 682 | |
| 683 | payload = json.dumps(updated, separators=(",", ":"), default=str).encode() |
| 684 | header = f"commit {len(payload)}\0".encode() |
| 685 | obj_path.parent.mkdir(parents=True, exist_ok=True) |
| 686 | obj_path.write_bytes(header + payload) |
| 687 | |
| 688 | |
| 689 | result.commits_written += 1 |
| 690 | |
| 691 | return result |
| 692 | |
| 693 | |
| 694 | # --------------------------------------------------------------------------- |
| 695 | # Public API |
| 696 | # --------------------------------------------------------------------------- |
| 697 | |
| 698 | def migrate( |
| 699 | root: pathlib.Path, |
| 700 | dry_run: bool = False, |
| 701 | signing_identity: SigningIdentity | None = None, |
| 702 | force_resign: bool = False, |
| 703 | ) -> MigrateResult: |
| 704 | """Migrate a Muse repository to the canonical current layout. |
| 705 | |
| 706 | Args: |
| 707 | root: Repo root (contains ``.muse/``). |
| 708 | dry_run: When True, compute all changes and report them but |
| 709 | write nothing to disk. |
| 710 | signing_identity: When provided, unsigned commits are signed with this |
| 711 | identity during migration. Already-signed commits |
| 712 | are never re-signed. |
| 713 | |
| 714 | Returns: |
| 715 | :class:`MigrateResult` with full migration summary. |
| 716 | |
| 717 | Raises: |
| 718 | RuntimeError: If a merge or rebase is in progress. |
| 719 | """ |
| 720 | if force_resign and signing_identity is None: |
| 721 | raise ValueError( |
| 722 | "force_resign=True requires a signing identity. " |
| 723 | "Run `muse auth keygen && muse auth register`, or connect a hub " |
| 724 | "with `muse hub connect` so credentials can be resolved." |
| 725 | ) |
| 726 | if (muse_dir(root) / "MERGE_STATE").exists(): |
| 727 | raise RuntimeError( |
| 728 | "A merge is in progress — finish or abort it before running migrate." |
| 729 | ) |
| 730 | if rebase_merge_dir(root).exists(): |
| 731 | raise RuntimeError( |
| 732 | "A rebase is in progress — finish or abort it before running migrate." |
| 733 | ) |
| 734 | |
| 735 | # --- repo.json --- |
| 736 | repo_id_updated = _migrate_repo_id(root, dry_run=dry_run) |
| 737 | |
| 738 | # --- snapshots --- |
| 739 | snapshots_relocated = _migrate_snapshots(root, dry_run=dry_run) |
| 740 | |
| 741 | # --- collect all commits (flat + canonical) --- |
| 742 | all_commits = _collect_all_commits(root) |
| 743 | order = _topological_sort(all_commits) |
| 744 | |
| 745 | id_map: dict[str, str] = {} |
| 746 | commits_rewritten = 0 |
| 747 | commits_relocated = 0 |
| 748 | branch_fields_renamed = 0 |
| 749 | signatures_normalised = 0 |
| 750 | format_versions_bumped = 0 |
| 751 | commits_signed = 0 |
| 752 | unsigned_commits_skipped = 0 |
| 753 | |
| 754 | # First pass: build id_map (must be complete before second pass). |
| 755 | # Signing-aware: if signing_identity is provided, unsigned commits will |
| 756 | # get a new signer_public_key baked into their commit_id. |
| 757 | for old_id in order: |
| 758 | raw, _ = all_commits[old_id] |
| 759 | _, new_id, _ = _make_canonical_dict(raw, id_map, signing_identity, force_resign) |
| 760 | if new_id != old_id: |
| 761 | id_map[old_id] = new_id |
| 762 | |
| 763 | # Second pass: write canonical files, collect stats |
| 764 | for old_id in order: |
| 765 | raw, old_path = all_commits[old_id] |
| 766 | canonical, new_id, was_signed = _make_canonical_dict(raw, id_map, signing_identity, force_resign) |
| 767 | |
| 768 | is_flat = _is_flat_commit_path(root, old_path) |
| 769 | id_changed = new_id != old_id |
| 770 | branch_renamed = "created_on_branch" in raw and "branch" not in raw |
| 771 | sig_normalised = bool(raw.get("signature")) and not raw["signature"].startswith("ed25519:") |
| 772 | fv_bumped = raw.get("format_version", 1) != 8 |
| 773 | is_unsigned = not raw.get("signature") and not raw.get("signer_public_key") |
| 774 | has_repo_id = "repo_id" in raw |
| 775 | |
| 776 | if id_changed: |
| 777 | commits_rewritten += 1 |
| 778 | if is_flat: |
| 779 | commits_relocated += 1 |
| 780 | if branch_renamed: |
| 781 | branch_fields_renamed += 1 |
| 782 | if sig_normalised: |
| 783 | signatures_normalised += 1 |
| 784 | if fv_bumped: |
| 785 | format_versions_bumped += 1 |
| 786 | if was_signed: |
| 787 | commits_signed += 1 |
| 788 | elif is_unsigned and signing_identity is None: |
| 789 | unsigned_commits_skipped += 1 |
| 790 | |
| 791 | needs_write = id_changed or is_flat or branch_renamed or sig_normalised or fv_bumped or was_signed or has_repo_id |
| 792 | if not dry_run and needs_write: |
| 793 | _write_raw_commit(root, canonical) |
| 794 | # Delete old file if it's at a different location than the new canonical path |
| 795 | new_hex = long_id(new_id, strip=True) |
| 796 | new_path = _canonical_commit_path(root, new_hex) |
| 797 | if old_path != new_path and old_path.exists(): |
| 798 | old_path.unlink() |
| 799 | |
| 800 | # --- objects (legacy layout: objects/<shard> → objects/sha256/<shard>) --- |
| 801 | blobs_migrated, legacy_dirs_removed = _migrate_objects(root, dry_run=dry_run) |
| 802 | |
| 803 | # --- Phase 7: ID chain migration (blobs → snapshots → commits) --- |
| 804 | # Pass 1: rewrite raw blobs to muse-format at new hash_blob paths. |
| 805 | blob_result = migrate_blob_ids(root, dry_run=dry_run) |
| 806 | # Pass 2: update snapshot manifests with new blob IDs, write to object store. |
| 807 | snap_result = migrate_snapshot_ids(root, blob_result.id_map, dry_run=dry_run) |
| 808 | # Pass 3: update commit snapshot_id fields, recompute commit IDs, write to |
| 809 | # object store and new msgpack paths. |
| 810 | commit_result = migrate_commit_ids(root, snap_result.id_map, dry_run=dry_run) |
| 811 | |
| 812 | # Build the combined old→new ID map for ref and reflog updates. |
| 813 | # |
| 814 | # The existing loop maps old_id → legacy_id (parent cascade, signing). |
| 815 | # Phase 7 maps legacy_id → phase7_id (full ID chain). |
| 816 | # |
| 817 | # A naïve merge ({**id_map, **commit_result.id_map}) only resolves one hop: |
| 818 | # refs containing old_id would land at legacy_id, not phase7_id. We need |
| 819 | # the transitive closure: old_id → phase7_id in one step. |
| 820 | combined_id_map: dict[str, str] = {} |
| 821 | for old_id, legacy_id in id_map.items(): |
| 822 | final_id = commit_result.id_map.get(legacy_id, legacy_id) |
| 823 | combined_id_map[old_id] = final_id |
| 824 | # Also carry through any Phase 7 entries not covered by the legacy map |
| 825 | # (commits that were already at their legacy form before this run). |
| 826 | for legacy_id, phase7_id in commit_result.id_map.items(): |
| 827 | if legacy_id not in combined_id_map: |
| 828 | combined_id_map[legacy_id] = phase7_id |
| 829 | |
| 830 | # --- branch refs --- |
| 831 | refs_updated = _migrate_refs(root, combined_id_map, dry_run=dry_run) |
| 832 | |
| 833 | # --- remote refs --- |
| 834 | remote_refs_updated = _migrate_remote_refs(root, combined_id_map, dry_run=dry_run) |
| 835 | |
| 836 | # --- reflogs --- |
| 837 | reflogs_updated = _migrate_reflogs(root, combined_id_map, dry_run=dry_run) |
| 838 | |
| 839 | # --- remove legacy directories --- |
| 840 | # All content has been written to the unified object store by this point. |
| 841 | # Orphaned objects unreferenced by any branch will be collected by gc. |
| 842 | if not dry_run: |
| 843 | for legacy_dir in (commits_dir(root), snapshots_dir(root)): |
| 844 | if legacy_dir.exists(): |
| 845 | shutil.rmtree(legacy_dir) |
| 846 | |
| 847 | return MigrateResult( |
| 848 | id_map=combined_id_map, |
| 849 | commits_rewritten=commits_rewritten, |
| 850 | blobs_migrated=blobs_migrated, |
| 851 | legacy_dirs_removed=legacy_dirs_removed, |
| 852 | dry_run=dry_run, |
| 853 | commits_relocated=commits_relocated, |
| 854 | snapshots_relocated=snapshots_relocated, |
| 855 | refs_updated=refs_updated, |
| 856 | remote_refs_updated=remote_refs_updated, |
| 857 | repo_id_updated=repo_id_updated, |
| 858 | branch_fields_renamed=branch_fields_renamed, |
| 859 | signatures_normalised=signatures_normalised, |
| 860 | format_versions_bumped=format_versions_bumped, |
| 861 | reflogs_updated=reflogs_updated, |
| 862 | commits_signed=commits_signed, |
| 863 | unsigned_commits_skipped=unsigned_commits_skipped, |
| 864 | blobs_rewritten=blob_result.blobs_written, |
| 865 | snapshots_rewritten=snap_result.snapshots_written, |
| 866 | commits_unified=commit_result.commits_written, |
| 867 | ) |
File History
1 commit
sha256:707b4461365efd52a6360948db35151dd01ccdfbb8e5c62d6ed14afca7a0dc35
migrate_blob_ids now correctly skips snapshot and commit ob…
Human
patch
43 days ago