commit.py
python
sha256:99451767674c70e97323b61d5ef248ebe91530a91c2ab5902c2bb3e4acf250dd
Run in a fresh repo so no stale MERGE_STATE.json can bleed in
Human
patch
35 days ago
| 1 | """``muse commit`` — record the current workspace state as a new version. |
| 2 | |
| 3 | Algorithm |
| 4 | --------- |
| 5 | 1. Resolve repo root (walk up for ``.muse/``). |
| 6 | 2. Read the current branch from ``.muse/HEAD``. |
| 7 | 3. Invoke ``plugin.snapshot(root)`` to collect the workspace manifest |
| 8 | (domain-specific; the code plugin walks tracked source files). |
| 9 | 4. If the computed ``snapshot_id`` matches HEAD → "nothing to commit". |
| 10 | 5. Compute a deterministic ``commit_id`` = SHA-256 of (parents | snapshot | |
| 11 | message | timestamp). |
| 12 | 6. Write content-addressed blob objects to ``.muse/objects/sha256/``. |
| 13 | 7. Write snapshot record to ``.muse/objects/sha256/`` (unified store). |
| 14 | 8. Write commit record to ``.muse/objects/sha256/`` (unified store). |
| 15 | 9. Advance ``.muse/refs/heads/<branch>`` to the new ``commit_id``. |
| 16 | |
| 17 | ``--dry-run`` |
| 18 | Perform steps 1–5 (compute snapshot and commit_id) without writing |
| 19 | anything. Exits 0 when changes are present, 1 when the tree is clean. |
| 20 | Combine with ``--json`` for structured preflight output in agent pipelines. |
| 21 | |
| 22 | Exit codes:: |
| 23 | |
| 24 | 0 — commit created, OR nothing to commit (clean tree) |
| 25 | 1 — validation error (no message, unresolved conflicts, clean tree with --dry-run) |
| 26 | 3 — I/O error |
| 27 | """ |
| 28 | |
| 29 | import argparse |
| 30 | import datetime |
| 31 | import json |
| 32 | import logging |
| 33 | import os |
| 34 | import pathlib |
| 35 | import re |
| 36 | import sys |
| 37 | |
| 38 | from muse.cli.config import get_config_value, get_protected_branches, is_branch_protected |
| 39 | from muse.core.types import long_id, short_id, split_id |
| 40 | from muse.core.errors import ExitCode |
| 41 | from muse.core.merge_engine import clear_merge_state, read_merge_state |
| 42 | from muse.core.object_store import has_object, write_object_from_path |
| 43 | from muse.core.provenance import ( |
| 44 | encode_public_key, |
| 45 | make_agent_identity, |
| 46 | provenance_payload, |
| 47 | sign_commit_record, |
| 48 | ) |
| 49 | from muse.core.reflog import append_reflog |
| 50 | from muse.core.symlog import _write_symlogs |
| 51 | from muse.core.repo import require_repo |
| 52 | from muse.core.harmony import record_resolutions as harmony_record_resolutions |
| 53 | from muse.core.ids import hash_commit, hash_snapshot |
| 54 | from muse.core.types import ( |
| 55 | Manifest, |
| 56 | Metadata, |
| 57 | ) |
| 58 | from muse.core.refs import ( |
| 59 | RefConflictError, |
| 60 | get_head_commit_id, |
| 61 | read_current_branch, |
| 62 | write_branch_ref, |
| 63 | ) |
| 64 | from muse.core.commits import ( |
| 65 | CommitRecord, |
| 66 | MissingParentError, |
| 67 | get_head_snapshot_id, |
| 68 | read_commit, |
| 69 | write_commit, |
| 70 | ) |
| 71 | from muse.core.snapshots import ( |
| 72 | SnapshotRecord, |
| 73 | read_snapshot, |
| 74 | write_snapshot, |
| 75 | ) |
| 76 | from muse.core.validation import sanitize_display, sanitize_provenance, validate_branch_name |
| 77 | from muse.core.semver_classifier import classify_delta |
| 78 | from muse.domain import SemVerBump, SnapshotManifest, StagePlugin, StructuredDelta |
| 79 | from muse.plugins.code.stage import read_stage |
| 80 | from muse.plugins.registry import read_domain, resolve_plugin |
| 81 | from muse.core.timing import start_timer |
| 82 | from muse.core.envelope import EnvelopeJson, make_envelope |
| 83 | from typing import TypedDict |
| 84 | |
| 85 | logger = logging.getLogger(__name__) |
| 86 | |
| 87 | class _CommitErrorJson(EnvelopeJson): |
| 88 | """JSON output for commit error paths.""" |
| 89 | |
| 90 | error: str |
| 91 | message: str |
| 92 | |
| 93 | class _CommitConflictErrorJson(_CommitErrorJson): |
| 94 | """JSON output when there are unresolved merge conflicts.""" |
| 95 | |
| 96 | conflict_paths: list[str] |
| 97 | |
| 98 | class _CommitCleanJson(EnvelopeJson): |
| 99 | """JSON output when working tree is clean (nothing to commit).""" |
| 100 | |
| 101 | dry_run: bool |
| 102 | clean: bool |
| 103 | message: str |
| 104 | |
| 105 | class _CommitFilesChangedJson(TypedDict): |
| 106 | """File-change counts embedded in commit output.""" |
| 107 | |
| 108 | added: int |
| 109 | modified: int |
| 110 | deleted: int |
| 111 | total: int |
| 112 | |
| 113 | class _CommitJson(EnvelopeJson): |
| 114 | """JSON output for ``muse commit --json`` success and dry-run paths.""" |
| 115 | |
| 116 | dry_run: bool |
| 117 | clean: bool |
| 118 | commit_id: str |
| 119 | branch: str |
| 120 | snapshot_id: str |
| 121 | message: str |
| 122 | parent_commit_id: str | None |
| 123 | parent2_commit_id: str | None |
| 124 | committed_at: str |
| 125 | author: str |
| 126 | agent_id: str |
| 127 | model_id: str |
| 128 | toolchain_id: str |
| 129 | sem_ver_bump: str |
| 130 | breaking_changes: list[str] |
| 131 | signer_public_key: str |
| 132 | files_changed: _CommitFilesChangedJson |
| 133 | |
| 134 | # Maximum length for author and agent-provenance fields. |
| 135 | # Prevents DoS via arbitrarily long values and keeps commit records bounded. |
| 136 | _MAX_FIELD_LEN = 256 |
| 137 | def _normalize_prompt_hash(value: str) -> str: |
| 138 | """Canonicalize prompt_hash to sha256:<64-hex> or empty string. |
| 139 | |
| 140 | Accepts a bare 64-char hex digest or an already-prefixed sha256: value. |
| 141 | Any other input is rejected and returns "" — the prompt_hash field must |
| 142 | be self-describing or absent. |
| 143 | """ |
| 144 | if not value: |
| 145 | return "" |
| 146 | try: |
| 147 | _, hex_id = split_id(value) |
| 148 | return long_id(hex_id) |
| 149 | except ValueError: |
| 150 | return "" |
| 151 | |
| 152 | def register(subparsers: "argparse._SubParsersAction[argparse.ArgumentParser]") -> None: |
| 153 | """Register the ``muse commit`` subcommand and its flags.""" |
| 154 | parser = subparsers.add_parser( |
| 155 | "commit", |
| 156 | help="Record the current state as a new version.", |
| 157 | description=__doc__, |
| 158 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 159 | ) |
| 160 | parser.add_argument( |
| 161 | "-m", "--message", default=None, |
| 162 | help="Commit message (required unless --allow-empty is set).", |
| 163 | ) |
| 164 | parser.add_argument( |
| 165 | "--allow-empty", action="store_true", |
| 166 | help="Allow committing with no changes (empty-message commits still warn).", |
| 167 | ) |
| 168 | parser.add_argument( |
| 169 | "--dry-run", "-n", action="store_true", dest="dry_run", |
| 170 | help=( |
| 171 | "Compute snapshot and commit_id without writing anything. " |
| 172 | "Exits 0 when changes exist, 1 when the working tree is clean. " |
| 173 | "Combine with --json for structured preflight output in agent pipelines." |
| 174 | ), |
| 175 | ) |
| 176 | parser.add_argument( |
| 177 | "--section", default=None, |
| 178 | help="Tag this commit with a section label (verse, chorus, bridge…).", |
| 179 | ) |
| 180 | parser.add_argument( |
| 181 | "--track", default=None, |
| 182 | help="Tag this commit with an instrument track (drums, bass, keys…).", |
| 183 | ) |
| 184 | parser.add_argument( |
| 185 | "--emotion", default=None, |
| 186 | help="Attach an emotion label (joyful, melancholic, tense…).", |
| 187 | ) |
| 188 | parser.add_argument( |
| 189 | "--author", default=None, |
| 190 | help="Override the commit author.", |
| 191 | ) |
| 192 | parser.add_argument( |
| 193 | "--agent-id", default=None, dest="agent_id", |
| 194 | help="Agent identity string (overrides MUSE_AGENT_ID env var).", |
| 195 | ) |
| 196 | parser.add_argument( |
| 197 | "--model-id", default=None, dest="model_id", |
| 198 | help="Model identifier for AI agents (overrides MUSE_MODEL_ID env var).", |
| 199 | ) |
| 200 | parser.add_argument( |
| 201 | "--toolchain-id", default=None, dest="toolchain_id", |
| 202 | help="Toolchain string (overrides MUSE_TOOLCHAIN_ID env var).", |
| 203 | ) |
| 204 | parser.add_argument( |
| 205 | "--sign", action="store_true", |
| 206 | help="HMAC-sign the commit using the agent's stored key (requires --agent-id or MUSE_AGENT_ID).", |
| 207 | ) |
| 208 | parser.add_argument( |
| 209 | "--json", "-j", action="store_true", dest="json_out", |
| 210 | help="Emit machine-readable JSON.", |
| 211 | ) |
| 212 | parser.set_defaults(func=run) |
| 213 | |
| 214 | def run(args: argparse.Namespace) -> None: |
| 215 | """Record the current staged state as a new version. |
| 216 | |
| 217 | Snapshots the stage, writes the commit record, and advances the branch |
| 218 | pointer. Agent commits must include ``--agent-id``, ``--model-id``, and |
| 219 | ``--sign`` for full provenance. Use ``--dry-run`` to preview without |
| 220 | writing anything. |
| 221 | |
| 222 | Agent quickstart |
| 223 | ---------------- |
| 224 | :: |
| 225 | |
| 226 | muse commit -m "feat: add X" --agent-id claude-code --model-id claude-sonnet-4-6 --sign --json |
| 227 | muse commit -m "feat: add X" --dry-run --json |
| 228 | |
| 229 | JSON fields |
| 230 | ----------- |
| 231 | commit_id Full ``sha256:…`` commit ID (deterministic). |
| 232 | branch Branch the commit was written to. |
| 233 | snapshot_id Full ``sha256:…`` snapshot ID. |
| 234 | message Commit message. |
| 235 | parent_commit_id Parent commit ID; ``null`` for first commit. |
| 236 | parent2_commit_id Second parent; ``null`` for non-merge commits. |
| 237 | committed_at ISO-8601 timestamp. |
| 238 | author Author handle. |
| 239 | agent_id Agent identifier (empty string for human commits). |
| 240 | sem_ver_bump Inferred bump: ``"major"``, ``"minor"``, ``"patch"``, |
| 241 | or ``"none"``. |
| 242 | breaking_changes List of addresses of removed public symbols. |
| 243 | files_changed ``{added, modified, deleted}`` counts. |
| 244 | dry_run ``true`` when ``--dry-run`` was passed. |
| 245 | |
| 246 | Exit codes |
| 247 | ---------- |
| 248 | 0 Commit created; or nothing to commit (clean tree, no ``--dry-run``). |
| 249 | 1 Dry-run with clean tree; or validation error (missing message, conflicts). |
| 250 | 3 I/O error or repository not found. |
| 251 | """ |
| 252 | elapsed = start_timer() |
| 253 | message: str | None = args.message |
| 254 | allow_empty: bool = args.allow_empty |
| 255 | dry_run: bool = args.dry_run |
| 256 | section: str | None = args.section |
| 257 | track: str | None = args.track |
| 258 | emotion: str | None = args.emotion |
| 259 | raw_author: str | None = args.author |
| 260 | agent_id: str | None = args.agent_id |
| 261 | model_id: str | None = args.model_id |
| 262 | toolchain_id: str | None = args.toolchain_id |
| 263 | sign: bool = args.sign |
| 264 | json_out: bool = args.json_out |
| 265 | |
| 266 | if message is None and not allow_empty: |
| 267 | if json_out: |
| 268 | print(json.dumps(_CommitErrorJson( |
| 269 | **make_envelope(elapsed, exit_code=ExitCode.USER_ERROR), |
| 270 | error="no_message", |
| 271 | message="provide a commit message with -m MESSAGE", |
| 272 | ))) |
| 273 | print("❌ Provide a commit message with -m MESSAGE.", file=sys.stderr) |
| 274 | raise SystemExit(ExitCode.USER_ERROR) |
| 275 | |
| 276 | if message is None and allow_empty: |
| 277 | logger.warning( |
| 278 | "⚠️ --allow-empty used without -m: commit will have an empty message." |
| 279 | ) |
| 280 | |
| 281 | # Sanitize and cap the author field. An explicit --author override is a |
| 282 | # potential impersonation vector (an agent could supply a human's name). |
| 283 | # We strip all C0/DEL/C1 control chars and cap at _MAX_FIELD_LEN. |
| 284 | # A warning is emitted when the caller explicitly passes --author so the |
| 285 | # act is always visible in logs. |
| 286 | author: str | None = ( |
| 287 | sanitize_provenance(raw_author[:_MAX_FIELD_LEN]) if raw_author else None |
| 288 | ) |
| 289 | if raw_author is not None: |
| 290 | logger.warning( |
| 291 | "⚠️ --author override supplied: %r — this is not verified against " |
| 292 | "the stored identity and may allow impersonation.", |
| 293 | author, |
| 294 | ) |
| 295 | |
| 296 | root = require_repo() |
| 297 | |
| 298 | # config-based auto-sign: commit.sign = true → behave as if --sign was passed |
| 299 | if not sign and get_config_value("commit.sign", root) == "true": |
| 300 | sign = True |
| 301 | |
| 302 | # When no explicit --author is provided, resolve user.handle from |
| 303 | # identity.toml for the active hub (via get_config_value delegation). |
| 304 | if author is None: |
| 305 | identity_handle = get_config_value("user.handle", root) |
| 306 | if identity_handle: |
| 307 | author = sanitize_provenance(identity_handle[:_MAX_FIELD_LEN]) |
| 308 | |
| 309 | # Read merge state before any writes — needed for conflict check and |
| 310 | # harmony recording later. |
| 311 | merge_state = read_merge_state(root) |
| 312 | if merge_state is not None and merge_state.conflict_paths: |
| 313 | conflict_paths = sorted(merge_state.conflict_paths) |
| 314 | if json_out: |
| 315 | print(json.dumps(_CommitConflictErrorJson( |
| 316 | **make_envelope(elapsed, exit_code=ExitCode.USER_ERROR), |
| 317 | error="unresolved_conflicts", |
| 318 | conflict_paths=conflict_paths, |
| 319 | message="you have unresolved merge conflicts — resolve them before committing", |
| 320 | ))) |
| 321 | print( |
| 322 | "❌ You have unresolved merge conflicts. Resolve them before committing.", |
| 323 | file=sys.stderr, |
| 324 | ) |
| 325 | for p in conflict_paths: |
| 326 | print(f" both modified: {sanitize_display(p)}", file=sys.stderr) |
| 327 | raise SystemExit(ExitCode.USER_ERROR) |
| 328 | |
| 329 | branch = read_current_branch(root) |
| 330 | |
| 331 | protected = get_protected_branches(root) |
| 332 | if is_branch_protected(branch, protected): |
| 333 | msg = f"Branch '{branch}' is protected — commit directly to it is not allowed. Use a feature branch and merge." |
| 334 | if json_out: |
| 335 | print(json.dumps({ |
| 336 | **make_envelope(elapsed, exit_code=ExitCode.USER_ERROR), |
| 337 | "error": "protected_branch", |
| 338 | "message": msg, |
| 339 | })) |
| 340 | print(f"❌ {msg}", file=sys.stderr) |
| 341 | raise SystemExit(ExitCode.USER_ERROR) |
| 342 | |
| 343 | parent_id = get_head_commit_id(root, branch) |
| 344 | |
| 345 | # ── Guard: refuse when only unstaged changes exist (no staged entries) ──── |
| 346 | # Matches git behaviour: unstaged tracked modifications are not committed. |
| 347 | # Exception: no parent commit yet (first commit) — stage not required then. |
| 348 | # Staging is a code-domain concept; non-code domains commit the full workdir. |
| 349 | if parent_id and not allow_empty and read_domain(root) == "code": |
| 350 | stage = read_stage(root) |
| 351 | if not stage: |
| 352 | from muse.core.snapshot import diff_workdir_vs_snapshot |
| 353 | from muse.core.snapshots import get_head_snapshot_manifest |
| 354 | head_manifest = get_head_snapshot_manifest(root, branch) or {} |
| 355 | if head_manifest: |
| 356 | added, modified, deleted, _, _, _ = diff_workdir_vs_snapshot(root, head_manifest) |
| 357 | if added or modified or deleted: |
| 358 | msg = ( |
| 359 | "No changes staged for commit.\n" |
| 360 | " Use 'muse code add <file>' to stage changes before committing." |
| 361 | ) |
| 362 | if json_out: |
| 363 | print(json.dumps(_CommitErrorJson( |
| 364 | **make_envelope(elapsed, exit_code=ExitCode.USER_ERROR), |
| 365 | error="nothing_staged", |
| 366 | message=msg, |
| 367 | ))) |
| 368 | print(f"⚠️ {msg}", file=sys.stderr) |
| 369 | raise SystemExit(ExitCode.USER_ERROR) |
| 370 | |
| 371 | plugin = resolve_plugin(root) |
| 372 | snap = plugin.snapshot(root) |
| 373 | manifest = snap["files"] |
| 374 | directories = list(snap.get("directories") or []) |
| 375 | if not manifest and not allow_empty: |
| 376 | # An empty snapshot is valid when staged deletions produced it — |
| 377 | # e.g. `muse rm` removed the last tracked file(s). Only reject if |
| 378 | # there are no staged changes at all (truly virgin working tree). |
| 379 | stage = read_stage(root) |
| 380 | has_staged_deletions = any(e["mode"] == "D" for e in stage.values()) |
| 381 | if not has_staged_deletions: |
| 382 | if json_out: |
| 383 | print(json.dumps(_CommitErrorJson( |
| 384 | **make_envelope(elapsed, exit_code=ExitCode.USER_ERROR), |
| 385 | error="empty_workdir", |
| 386 | message="nothing tracked — working tree is empty", |
| 387 | ))) |
| 388 | print("⚠️ Nothing tracked — working tree is empty.", file=sys.stderr) |
| 389 | raise SystemExit(ExitCode.USER_ERROR) |
| 390 | |
| 391 | snapshot_id = hash_snapshot(manifest, directories) |
| 392 | |
| 393 | if not allow_empty: |
| 394 | head_snapshot = get_head_snapshot_id(root, branch) |
| 395 | if head_snapshot == snapshot_id: |
| 396 | if dry_run: |
| 397 | if json_out: |
| 398 | print(json.dumps(_CommitCleanJson( |
| 399 | **make_envelope(elapsed, exit_code=1), |
| 400 | dry_run=True, |
| 401 | clean=True, |
| 402 | message="Nothing to commit, working tree clean", |
| 403 | ))) |
| 404 | else: |
| 405 | print("Nothing to commit, working tree clean") |
| 406 | raise SystemExit(1) |
| 407 | if json_out: |
| 408 | print(json.dumps(_CommitCleanJson( |
| 409 | **make_envelope(elapsed), |
| 410 | dry_run=False, |
| 411 | clean=True, |
| 412 | message="Nothing to commit, working tree clean", |
| 413 | ))) |
| 414 | else: |
| 415 | print("Nothing to commit, working tree clean") |
| 416 | raise SystemExit(ExitCode.SUCCESS) |
| 417 | |
| 418 | committed_at = datetime.datetime.now(datetime.timezone.utc) |
| 419 | parent_ids = [parent_id] if parent_id else [] |
| 420 | |
| 421 | # When completing a conflicted merge, include the second parent so that |
| 422 | # the merge is recorded as a true two-parent merge commit. This ensures |
| 423 | # subsequent merge --dry-run correctly computes the LCA as the resolved |
| 424 | # merge commit rather than the pre-merge common ancestor, preventing the |
| 425 | # same conflicts from re-appearing on the next merge attempt. |
| 426 | merge_parent2: str | None = None |
| 427 | if merge_state is not None and merge_state.theirs_commit: |
| 428 | merge_parent2 = merge_state.theirs_commit |
| 429 | if merge_parent2 not in parent_ids: |
| 430 | parent_ids = parent_ids + [merge_parent2] |
| 431 | |
| 432 | # Resolve agent provenance: CLI flags take priority over environment vars. |
| 433 | # 1. Truncate to _MAX_FIELD_LEN chars — prevents DoS via arbitrarily long values. |
| 434 | # 2. Strip all C0/DEL/C1 control characters — prevents terminal injection |
| 435 | # when provenance fields are rendered in display paths (muse log, muse read, |
| 436 | # agent dashboards), log-line splitting, and visual spoofing. |
| 437 | resolved_agent_id = sanitize_provenance( |
| 438 | (agent_id or os.environ.get("MUSE_AGENT_ID", ""))[:_MAX_FIELD_LEN] |
| 439 | ) |
| 440 | resolved_model_id = sanitize_provenance( |
| 441 | (model_id or os.environ.get("MUSE_MODEL_ID", ""))[:_MAX_FIELD_LEN] |
| 442 | ) |
| 443 | resolved_toolchain_id = sanitize_provenance( |
| 444 | (toolchain_id or os.environ.get("MUSE_TOOLCHAIN_ID", ""))[:_MAX_FIELD_LEN] |
| 445 | ) |
| 446 | _raw_prompt_hash = sanitize_provenance( |
| 447 | os.environ.get("MUSE_PROMPT_HASH", "")[:_MAX_FIELD_LEN] |
| 448 | ) |
| 449 | resolved_prompt_hash = _normalize_prompt_hash(_raw_prompt_hash) |
| 450 | |
| 451 | # Resolve signing identity early so signer_public_key is bound into the |
| 452 | # commit ID (v2 formula). The public key is deterministic from the private |
| 453 | # key — no side effects from resolving it here. |
| 454 | signing = None |
| 455 | pre_signer_public_key = "" |
| 456 | if sign and resolved_agent_id: |
| 457 | from muse.cli.config import get_signing_identity |
| 458 | signing = get_signing_identity(root, agent_id=resolved_agent_id) |
| 459 | if signing is not None: |
| 460 | _, pre_signer_public_key = encode_public_key(signing.private_key) |
| 461 | else: |
| 462 | logger.warning( |
| 463 | "No signing identity found for agent %r — commit will be unsigned. " |
| 464 | "Run `muse auth keygen && muse auth register` to set up a keypair.", |
| 465 | resolved_agent_id, |
| 466 | ) |
| 467 | |
| 468 | commit_id = hash_commit( |
| 469 | parent_ids=parent_ids, |
| 470 | snapshot_id=snapshot_id, |
| 471 | message=message or "", |
| 472 | committed_at_iso=committed_at.isoformat(), |
| 473 | author=author or "", |
| 474 | signer_public_key=pre_signer_public_key, |
| 475 | ) |
| 476 | |
| 477 | metadata: Metadata = {} |
| 478 | if section: |
| 479 | metadata["section"] = section |
| 480 | if track: |
| 481 | metadata["track"] = track |
| 482 | if emotion: |
| 483 | metadata["emotion"] = emotion |
| 484 | |
| 485 | # Load the parent snapshot manifest once and reuse it for both |
| 486 | # structured_delta computation and file-count output. Previously the |
| 487 | # manifest was loaded independently in each section — two separate |
| 488 | # read_snapshot() calls per commit. |
| 489 | parent_manifest: Manifest = {} |
| 490 | parent_directories: list[str] = [] |
| 491 | if parent_id is not None: |
| 492 | parent_commit_rec = read_commit(root, parent_id) |
| 493 | if parent_commit_rec is not None: |
| 494 | parent_snap_record = read_snapshot(root, parent_commit_rec.snapshot_id) |
| 495 | if parent_snap_record is not None: |
| 496 | parent_manifest = dict(parent_snap_record.manifest) |
| 497 | parent_directories = list(parent_snap_record.directories) |
| 498 | |
| 499 | # Compute a structured delta against the parent snapshot so muse read |
| 500 | # can display note-level changes without reloading blobs. |
| 501 | # For the genesis commit (no parent) diff against an empty snapshot so |
| 502 | # every tracked symbol appears as op=insert — indexers depend on this to |
| 503 | # record symbol births as op=add rather than op=modify. |
| 504 | structured_delta: StructuredDelta | None = None |
| 505 | sem_ver_bump: SemVerBump = "none" |
| 506 | breaking_changes: list[str] = [] |
| 507 | domain = read_domain(root) |
| 508 | base_snap = SnapshotManifest( |
| 509 | files=parent_manifest, |
| 510 | domain=domain, |
| 511 | directories=parent_directories, |
| 512 | ) |
| 513 | try: |
| 514 | structured_delta = plugin.diff(base_snap, snap, repo_root=root) |
| 515 | except Exception as exc: |
| 516 | # plugin.diff() is domain-specific and may fail on unsupported |
| 517 | # file types. The commit proceeds without a structured delta; |
| 518 | # sem_ver_bump defaults to "none". |
| 519 | logger.debug("plugin.diff() failed — structured delta omitted: %s", exc) |
| 520 | structured_delta = None |
| 521 | |
| 522 | # Classify the structured delta into a semver bump and breaking-change list. |
| 523 | if structured_delta is not None: |
| 524 | classification = classify_delta(structured_delta, repo_root=root) |
| 525 | sem_ver_bump = classification.bump |
| 526 | breaking_changes = classification.breaking_addresses |
| 527 | structured_delta["sem_ver_bump"] = sem_ver_bump |
| 528 | structured_delta["breaking_changes"] = breaking_changes |
| 529 | |
| 530 | # Compute file-level change counts from the (now single-read) parent manifest. |
| 531 | files_added = len(set(manifest) - set(parent_manifest)) |
| 532 | files_deleted = len(set(parent_manifest) - set(manifest)) |
| 533 | files_modified = sum( |
| 534 | 1 for p in set(manifest) & set(parent_manifest) |
| 535 | if manifest[p] != parent_manifest[p] |
| 536 | ) |
| 537 | |
| 538 | # ── Dry-run path — no writes beyond this point ──────────────────────────── |
| 539 | if dry_run: |
| 540 | if json_out: |
| 541 | print(json.dumps(_CommitJson( |
| 542 | **make_envelope(elapsed), |
| 543 | dry_run=True, |
| 544 | clean=False, |
| 545 | commit_id=commit_id, |
| 546 | branch=branch, |
| 547 | snapshot_id=snapshot_id, |
| 548 | message=message or "", |
| 549 | parent_commit_id=parent_id, |
| 550 | parent2_commit_id=merge_parent2, |
| 551 | committed_at=committed_at.isoformat(), |
| 552 | author=author or "", |
| 553 | agent_id=resolved_agent_id, |
| 554 | model_id=resolved_model_id, |
| 555 | toolchain_id=resolved_toolchain_id, |
| 556 | sem_ver_bump=sem_ver_bump, |
| 557 | breaking_changes=breaking_changes, |
| 558 | signer_public_key=pre_signer_public_key, |
| 559 | files_changed=_CommitFilesChangedJson( |
| 560 | added=files_added, |
| 561 | modified=files_modified, |
| 562 | deleted=files_deleted, |
| 563 | total=files_added + files_modified + files_deleted, |
| 564 | ), |
| 565 | ))) |
| 566 | else: |
| 567 | total = files_added + files_deleted + files_modified |
| 568 | print(f"[dry-run] [{sanitize_display(branch)} {commit_id}] {sanitize_display(message or '')}") |
| 569 | if total: |
| 570 | parts: list[str] = [] |
| 571 | if files_modified: |
| 572 | parts.append(f"{files_modified} modified") |
| 573 | if files_added: |
| 574 | parts.append(f"{files_added} added") |
| 575 | if files_deleted: |
| 576 | parts.append(f"{files_deleted} removed") |
| 577 | print(f" {total} file{'s' if total != 1 else ''} changed ({', '.join(parts)})") |
| 578 | print(" (dry-run: nothing written)") |
| 579 | return |
| 580 | |
| 581 | # ── Actual writes ───────────────────────────────────────────────────────── |
| 582 | # Write objects for every file whose object is not yet in the store. |
| 583 | # We skip only when the parent manifest has the same ID *and* the object |
| 584 | # is actually present — parent objects may be absent after a clone without |
| 585 | # blobs, a gc run, or a prior commit that failed to write them. |
| 586 | for rel_path, object_id in manifest.items(): |
| 587 | if parent_manifest.get(rel_path) == object_id and has_object(root, object_id): |
| 588 | continue |
| 589 | write_object_from_path(root, object_id, root / rel_path) |
| 590 | |
| 591 | write_snapshot(root, SnapshotRecord(snapshot_id=snapshot_id, manifest=manifest, directories=directories)) |
| 592 | |
| 593 | signature = "" |
| 594 | signer_public_key = pre_signer_public_key |
| 595 | signer_key_id = "" |
| 596 | if signing is not None: |
| 597 | result = sign_commit_record( |
| 598 | commit_id, |
| 599 | resolved_agent_id, |
| 600 | signing.private_key, |
| 601 | author=author or "", |
| 602 | model_id=resolved_model_id, |
| 603 | toolchain_id=resolved_toolchain_id, |
| 604 | prompt_hash=resolved_prompt_hash, |
| 605 | committed_at=committed_at.isoformat(), |
| 606 | ) |
| 607 | if result is not None: |
| 608 | signature, signer_public_key, signer_key_id = result |
| 609 | |
| 610 | _commit_record = CommitRecord( |
| 611 | commit_id=commit_id, |
| 612 | branch=branch, |
| 613 | snapshot_id=snapshot_id, |
| 614 | message=message or "", |
| 615 | committed_at=committed_at, |
| 616 | parent_commit_id=parent_id, |
| 617 | parent2_commit_id=merge_parent2, |
| 618 | author=author or "", |
| 619 | metadata=metadata, |
| 620 | structured_delta=structured_delta, |
| 621 | sem_ver_bump=sem_ver_bump, |
| 622 | breaking_changes=breaking_changes, |
| 623 | agent_id=resolved_agent_id, |
| 624 | model_id=resolved_model_id, |
| 625 | toolchain_id=resolved_toolchain_id, |
| 626 | prompt_hash=resolved_prompt_hash, |
| 627 | signature=signature, |
| 628 | signer_public_key=signer_public_key, |
| 629 | signer_key_id=signer_key_id, |
| 630 | ) |
| 631 | try: |
| 632 | write_commit(root, _commit_record) |
| 633 | except MissingParentError: |
| 634 | write_commit(root, _commit_record, skip_parent_check=True) |
| 635 | |
| 636 | try: |
| 637 | write_branch_ref(root, branch, commit_id, expected_id=parent_id) |
| 638 | except RefConflictError as exc: |
| 639 | msg = str(exc) |
| 640 | if json_out: |
| 641 | print(json.dumps(_CommitErrorJson( |
| 642 | **make_envelope(elapsed, exit_code=ExitCode.USER_ERROR), |
| 643 | error="branch_conflict", |
| 644 | message=msg, |
| 645 | ))) |
| 646 | print(f"❌ {msg}", file=sys.stderr) |
| 647 | raise SystemExit(ExitCode.USER_ERROR) |
| 648 | |
| 649 | # Clear the stage after a successful commit so the next muse commit |
| 650 | # returns to full-snapshot mode unless the user runs muse code add again. |
| 651 | # Wrapped in try/except: clear_stage failure must not hide a successful |
| 652 | # commit. A failure here leaves staged index entries for already-committed |
| 653 | # content — the next muse status would show them as staged, and a naive |
| 654 | # re-commit would produce a duplicate commit with identical snapshot_id. |
| 655 | # Logging the error is sufficient; the commit is already durable. |
| 656 | if isinstance(plugin, StagePlugin): |
| 657 | try: |
| 658 | plugin.clear_stage(root) |
| 659 | except Exception as _clear_exc: |
| 660 | logger.warning( |
| 661 | "⚠️ clear_stage failed after successful commit %s — " |
| 662 | "stage index may still list committed files: %s", |
| 663 | commit_id, _clear_exc, |
| 664 | ) |
| 665 | |
| 666 | append_reflog( |
| 667 | root, |
| 668 | branch, |
| 669 | old_id=parent_id, |
| 670 | new_id=commit_id, |
| 671 | author=author or "unknown", |
| 672 | operation=f"commit: {sanitize_display(message or '(no message)')}", |
| 673 | ) |
| 674 | |
| 675 | try: |
| 676 | _write_symlogs( |
| 677 | root, |
| 678 | parent_commit_id=parent_id, |
| 679 | new_snapshot_id=snapshot_id, |
| 680 | new_commit_id=commit_id, |
| 681 | author=author or "unknown", |
| 682 | commit_message=message or "", |
| 683 | ) |
| 684 | except Exception as _symlog_exc: |
| 685 | logger.warning( |
| 686 | "⚠️ symlog write failed for commit %s — symbol journal may be incomplete: %s", |
| 687 | commit_id, _symlog_exc, |
| 688 | ) |
| 689 | |
| 690 | # If this commit completed a conflicted merge, record how each conflict |
| 691 | # was resolved so harmony can replay it on future identical conflicts. |
| 692 | # clear_merge_state is unconditional — harmony recording is optional |
| 693 | # bookkeeping, but cleanup must always happen after a successful commit. |
| 694 | if merge_state is not None: |
| 695 | if merge_state.ours_commit and merge_state.theirs_commit: |
| 696 | def _manifest_for(cid: str) -> Manifest: |
| 697 | cr = read_commit(root, cid) |
| 698 | if cr is None: |
| 699 | return {} |
| 700 | snap_rec = read_snapshot(root, cr.snapshot_id) |
| 701 | return snap_rec.manifest if snap_rec else {} |
| 702 | |
| 703 | ours_manifest = _manifest_for(merge_state.ours_commit) |
| 704 | theirs_manifest = _manifest_for(merge_state.theirs_commit) |
| 705 | domain = read_domain(root) |
| 706 | # Use original_conflict_paths so harmony learns even when all conflicts |
| 707 | # were resolved via `muse checkout --ours/--theirs` before commit |
| 708 | # (which clears conflict_paths but preserves original_conflict_paths). |
| 709 | all_conflict_paths = ( |
| 710 | merge_state.original_conflict_paths or merge_state.conflict_paths |
| 711 | ) |
| 712 | harmony_record_resolutions( |
| 713 | root, |
| 714 | list(all_conflict_paths), |
| 715 | ours_manifest, |
| 716 | theirs_manifest, |
| 717 | manifest, |
| 718 | domain, |
| 719 | plugin, |
| 720 | manually_resolved=set(merge_state.manually_resolved) if merge_state.manually_resolved else None, |
| 721 | ) |
| 722 | clear_merge_state(root) |
| 723 | |
| 724 | # ── Output ──────────────────────────────────────────────────────────────── |
| 725 | if json_out: |
| 726 | print(json.dumps(_CommitJson( |
| 727 | **make_envelope(elapsed), |
| 728 | dry_run=False, |
| 729 | clean=False, |
| 730 | commit_id=commit_id, |
| 731 | branch=branch, |
| 732 | snapshot_id=snapshot_id, |
| 733 | message=message or "", |
| 734 | parent_commit_id=parent_id, |
| 735 | parent2_commit_id=merge_parent2, |
| 736 | committed_at=committed_at.isoformat(), |
| 737 | author=author or "", |
| 738 | agent_id=resolved_agent_id, |
| 739 | model_id=resolved_model_id, |
| 740 | toolchain_id=resolved_toolchain_id, |
| 741 | sem_ver_bump=sem_ver_bump, |
| 742 | breaking_changes=breaking_changes, |
| 743 | signer_public_key=signer_public_key, |
| 744 | files_changed=_CommitFilesChangedJson( |
| 745 | added=files_added, |
| 746 | modified=files_modified, |
| 747 | deleted=files_deleted, |
| 748 | total=files_added + files_modified + files_deleted, |
| 749 | ), |
| 750 | ))) |
| 751 | else: |
| 752 | print(f"[{sanitize_display(branch)} {commit_id}] {sanitize_display(message or '')}") |
| 753 | total_files = files_added + files_deleted + files_modified |
| 754 | if total_files: |
| 755 | stat_parts: list[str] = [] |
| 756 | if files_modified: |
| 757 | stat_parts.append(f"{files_modified} modified") |
| 758 | if files_added: |
| 759 | stat_parts.append(f"{files_added} added") |
| 760 | if files_deleted: |
| 761 | stat_parts.append(f"{files_deleted} removed") |
| 762 | print(f" {total_files} file{'s' if total_files != 1 else ''} changed ({', '.join(stat_parts)})") |
File History
3 commits
sha256:99451767674c70e97323b61d5ef248ebe91530a91c2ab5902c2bb3e4acf250dd
Run in a fresh repo so no stale MERGE_STATE.json can bleed in
Human
patch
35 days ago
sha256:1d3f5470f45db58e32047678debc9438fdded1b2c7332cc743d2b8be32fdafc8
fixing more broken tests
Human
patch
41 days ago
sha256:2a1cf861048b753a21d6ca853a83cdfc2a46f15dcbb561ee79ebb9dc40c03af6
switch same-commit fix, agent-config user-global config, an…
Human
patch
42 days ago