bisect.py
python
sha256:51ce277f663e01a43eaffbe77509b1de7ac2d4251b55d23306304bcdeb92c90d
feat(pack): delta-encode snapshots in MPackBundle wire format
Sonnet 4.6
minor
⚠ breaking
128 days ago
| 1 | """``muse bisect`` — binary search through commit history to find regressions. |
| 2 | |
| 3 | ``muse bisect`` is Muse's power-tool for regression hunting. Given a |
| 4 | known-bad commit and a known-good commit it performs a binary search through |
| 5 | the history between them, asking at each midpoint: *"does the bug exist |
| 6 | here?"* until the first bad commit is isolated. |
| 7 | |
| 8 | It is fully agent-safe: ``muse bisect run <cmd>`` automates the search by |
| 9 | running an arbitrary command at each step and interpreting the exit code: |
| 10 | |
| 11 | 0 → good (bug not present) |
| 12 | 125 → skip (commit untestable) |
| 13 | else → bad (bug present) |
| 14 | |
| 15 | Symbol-scoped bisect |
| 16 | -------------------- |
| 17 | Pass ``--symbol addr`` to ``muse bisect start`` to restrict the candidate |
| 18 | commit list to only commits whose structured_delta touched that symbol. |
| 19 | This can reduce a 9-step bisect over 300 commits to a 3-step bisect over |
| 20 | the 8 commits that actually changed the symbol you care about. |
| 21 | |
| 22 | muse bisect start --bad HEAD --good v1.0.0 \\ |
| 23 | --symbol billing.py::Invoice.compute_total |
| 24 | |
| 25 | At each step, Muse shows which ops the midpoint commit applied to the |
| 26 | symbol so you know exactly what changed before you run your tests. |
| 27 | |
| 28 | JSON output |
| 29 | ----------- |
| 30 | Pass ``--json`` to any subcommand for machine-readable NDJSON output. |
| 31 | The ``run`` subcommand emits one JSON object per bisect step (NDJSON), |
| 32 | plus a final summary line. All other subcommands emit a single JSON |
| 33 | object on stdout. |
| 34 | |
| 35 | Subcommands:: |
| 36 | |
| 37 | muse bisect start [--bad <ref>] [--good <ref>] [--symbol <addr>] [--json] |
| 38 | muse bisect bad [<ref>] [--json] |
| 39 | muse bisect good [<ref>] [--json] |
| 40 | muse bisect skip [<ref>] [--json] |
| 41 | muse bisect run <command> [--json] |
| 42 | muse bisect log [--json] |
| 43 | muse bisect reset [--json] |
| 44 | |
| 45 | Exit codes:: |
| 46 | |
| 47 | 0 — success |
| 48 | 1 — user error (no session, bad ref, bad args) |
| 49 | 2 — internal error (lost state) |
| 50 | """ |
| 51 | |
| 52 | import argparse |
| 53 | import json |
| 54 | import logging |
| 55 | import pathlib |
| 56 | import sys |
| 57 | from typing import TypedDict |
| 58 | |
| 59 | from muse.core.bisect import ( |
| 60 | BisectResult, |
| 61 | BisectStatusDict, |
| 62 | _commits_touching_symbol, |
| 63 | _symbol_ops_in_commit, |
| 64 | get_bisect_log, |
| 65 | get_bisect_next, |
| 66 | get_bisect_status, |
| 67 | is_bisect_active, |
| 68 | mark_bad, |
| 69 | mark_good, |
| 70 | reset_bisect, |
| 71 | run_bisect_command, |
| 72 | skip_commit, |
| 73 | start_bisect, |
| 74 | ) |
| 75 | from muse.core.envelope import EnvelopeJson, make_envelope |
| 76 | from muse.core.timing import start_timer |
| 77 | from muse.core.errors import ExitCode |
| 78 | from muse.core.repo import read_repo_id, require_repo |
| 79 | from muse.core.store import get_head_commit_id, read_current_branch, resolve_commit_ref |
| 80 | from muse.core.validation import sanitize_display |
| 81 | |
| 82 | logger = logging.getLogger(__name__) |
| 83 | |
| 84 | _MAX_SYMBOL_ADDR_LEN = 500 |
| 85 | |
| 86 | # --------------------------------------------------------------------------- |
| 87 | # Typed JSON schemas |
| 88 | # --------------------------------------------------------------------------- |
| 89 | |
| 90 | class _BisectStepPayload(TypedDict): |
| 91 | """Domain fields built by ``_result_to_json``.""" |
| 92 | |
| 93 | done: bool |
| 94 | first_bad: str | None |
| 95 | next_to_test: str | None |
| 96 | remaining_count: int |
| 97 | steps_remaining: int |
| 98 | verdict: str |
| 99 | symbol_changes: list[str] |
| 100 | |
| 101 | class _BisectStepJson(_BisectStepPayload, EnvelopeJson): |
| 102 | """Full wire shape for start / bad / good / skip subcommands.""" |
| 103 | |
| 104 | class _BisectLogEntryJson(TypedDict): |
| 105 | """One structured entry in the bisect log.""" |
| 106 | |
| 107 | commit_id: str |
| 108 | verdict: str |
| 109 | timestamp: str |
| 110 | |
| 111 | class _BisectStatusJson(EnvelopeJson, total=False): |
| 112 | """JSON output for ``muse bisect status --json``. |
| 113 | |
| 114 | ``active`` is always present. All other keys are only present when |
| 115 | ``active`` is ``True``. |
| 116 | """ |
| 117 | |
| 118 | active: bool |
| 119 | bad_id: str |
| 120 | good_ids: list[str] |
| 121 | symbol_filter: str |
| 122 | remaining_count: int |
| 123 | steps_remaining: int |
| 124 | skipped_count: int |
| 125 | branch: str |
| 126 | |
| 127 | class _BisectRunStepJson(EnvelopeJson): |
| 128 | """One NDJSON line emitted by ``muse bisect run --json`` per step.""" |
| 129 | |
| 130 | step: int |
| 131 | testing: str |
| 132 | verdict: str |
| 133 | remaining_count: int |
| 134 | done: bool |
| 135 | symbol_changes: list[str] |
| 136 | |
| 137 | class _BisectRunDoneJson(TypedDict): |
| 138 | """Final NDJSON line emitted by ``muse bisect run --json`` when complete.""" |
| 139 | |
| 140 | done: bool |
| 141 | first_bad: str | None |
| 142 | steps_taken: int |
| 143 | |
| 144 | class _BisectLogJson(EnvelopeJson): |
| 145 | """JSON output for ``muse bisect log --json``.""" |
| 146 | |
| 147 | active: bool |
| 148 | entries: list[_BisectLogEntryJson] |
| 149 | |
| 150 | class _BisectResetJson(EnvelopeJson): |
| 151 | """JSON output for ``muse bisect reset --json``.""" |
| 152 | |
| 153 | reset: bool |
| 154 | |
| 155 | # --------------------------------------------------------------------------- |
| 156 | # Internal helpers |
| 157 | # --------------------------------------------------------------------------- |
| 158 | |
| 159 | def _resolve_ref(root: pathlib.Path, ref: str | None) -> str: |
| 160 | """Resolve *ref* to a full commit ID; fall back to HEAD when *ref* is None.""" |
| 161 | branch = read_current_branch(root) |
| 162 | repo_id = read_repo_id(root) |
| 163 | if ref is None: |
| 164 | commit_id = get_head_commit_id(root, branch) |
| 165 | if not commit_id: |
| 166 | print("❌ No commits on current branch.", file=sys.stderr) |
| 167 | raise SystemExit(ExitCode.USER_ERROR) |
| 168 | return commit_id |
| 169 | commit = resolve_commit_ref(root, repo_id, branch, ref) |
| 170 | if commit is None: |
| 171 | print(f"❌ Ref '{sanitize_display(ref)}' not found.", file=sys.stderr) |
| 172 | raise SystemExit(ExitCode.USER_ERROR) |
| 173 | return commit.commit_id |
| 174 | |
| 175 | def _print_result(result: BisectResult) -> None: |
| 176 | """Render a BisectResult as human-readable text to stdout.""" |
| 177 | if result.done: |
| 178 | print(f"\n✅ First bad commit found: {sanitize_display(result.first_bad or '')}") |
| 179 | print(" Run 'muse bisect reset' to end the session.") |
| 180 | else: |
| 181 | print( |
| 182 | f"Next to test: {sanitize_display(result.next_to_test or '')} " |
| 183 | f"({result.remaining_count} remaining, ~{result.steps_remaining} step(s) left)" |
| 184 | ) |
| 185 | if result.symbol_changes: |
| 186 | print(" Symbol changes in this commit:") |
| 187 | for line in result.symbol_changes: |
| 188 | print(f" {sanitize_display(line)}") |
| 189 | |
| 190 | def _result_to_json(result: BisectResult) -> _BisectStepPayload: |
| 191 | """Convert a BisectResult to a typed JSON-serialisable dict.""" |
| 192 | return _BisectStepPayload( |
| 193 | done=result.done, |
| 194 | first_bad=result.first_bad, |
| 195 | next_to_test=result.next_to_test, |
| 196 | remaining_count=result.remaining_count, |
| 197 | steps_remaining=result.steps_remaining, |
| 198 | verdict=result.verdict, |
| 199 | symbol_changes=[sanitize_display(s) for s in result.symbol_changes], |
| 200 | ) |
| 201 | |
| 202 | def _parse_log_entry(raw: str) -> _BisectLogEntryJson: |
| 203 | """Parse ``"<commit_id> <verdict> <ISO8601_timestamp>"`` into a typed dict. |
| 204 | |
| 205 | The on-disk log format written by :func:`muse.core.bisect._apply_verdict` |
| 206 | and :func:`muse.core.bisect.start_bisect` is a space-separated triple. |
| 207 | Any missing or malformed parts default to empty strings. All fields are |
| 208 | sanitized before output. |
| 209 | """ |
| 210 | parts = raw.split(" ", 2) |
| 211 | return _BisectLogEntryJson( |
| 212 | commit_id=sanitize_display(parts[0]) if len(parts) > 0 else "", |
| 213 | verdict=sanitize_display(parts[1]) if len(parts) > 1 else "", |
| 214 | timestamp=sanitize_display(parts[2]) if len(parts) > 2 else "", |
| 215 | ) |
| 216 | |
| 217 | # --------------------------------------------------------------------------- |
| 218 | # Subcommand handlers |
| 219 | # --------------------------------------------------------------------------- |
| 220 | |
| 221 | def run_bisect_start(args: argparse.Namespace) -> None: |
| 222 | """Start a bisect session between a known-bad and known-good commit. |
| 223 | |
| 224 | Immediately suggests the midpoint commit to test. Use ``--symbol`` to |
| 225 | restrict the search to commits that touched a specific symbol — this can |
| 226 | reduce a 9-step bisect to 3 steps when the regressing symbol is known. |
| 227 | |
| 228 | Agent quickstart |
| 229 | ---------------- |
| 230 | :: |
| 231 | |
| 232 | muse bisect start --bad HEAD --good v1.0.0 --json |
| 233 | muse bisect start --bad HEAD --good v1.0.0 --symbol billing.py::Invoice.compute_total --json |
| 234 | |
| 235 | JSON fields |
| 236 | ----------- |
| 237 | done ``true`` when the first bad commit has been isolated. |
| 238 | first_bad Full commit ID of the first bad commit; ``null`` while in progress. |
| 239 | next_to_test Full commit ID of the next midpoint to test; ``null`` when done. |
| 240 | remaining_count Number of commits still in the candidate set. |
| 241 | steps_remaining Estimated binary-search steps left. |
| 242 | verdict ``"started"`` for this subcommand. |
| 243 | symbol_changes List of symbol-op descriptions for the midpoint commit. |
| 244 | |
| 245 | Exit codes |
| 246 | ---------- |
| 247 | 0 Session started (or immediately resolved when bad/good are adjacent). |
| 248 | 1 Session already active, invalid ``--symbol``, no ``--good``, or ref not found. |
| 249 | 2 Not inside a Muse repository. |
| 250 | """ |
| 251 | elapsed = start_timer() |
| 252 | bad: str | None = args.bad |
| 253 | good: list[str] | None = args.good |
| 254 | symbol: str | None = args.symbol |
| 255 | json_out: bool = args.json_out |
| 256 | |
| 257 | root = require_repo() |
| 258 | if is_bisect_active(root): |
| 259 | print( |
| 260 | "⚠️ A bisect session is already active. Run 'muse bisect reset' first.", |
| 261 | file=sys.stderr, |
| 262 | ) |
| 263 | raise SystemExit(ExitCode.USER_ERROR) |
| 264 | |
| 265 | if symbol is not None: |
| 266 | if "::" not in symbol: |
| 267 | print( |
| 268 | f"❌ --symbol must be a qualified symbol address " |
| 269 | f"(e.g. billing.py::func), got: {sanitize_display(symbol)!r}", |
| 270 | file=sys.stderr, |
| 271 | ) |
| 272 | raise SystemExit(ExitCode.USER_ERROR) |
| 273 | if len(symbol) > _MAX_SYMBOL_ADDR_LEN: |
| 274 | print( |
| 275 | f"❌ --symbol address too long (max {_MAX_SYMBOL_ADDR_LEN} chars).", |
| 276 | file=sys.stderr, |
| 277 | ) |
| 278 | raise SystemExit(ExitCode.USER_ERROR) |
| 279 | |
| 280 | bad_id = _resolve_ref(root, bad) |
| 281 | good_ids = [_resolve_ref(root, g) for g in (good or [])] |
| 282 | if not good_ids: |
| 283 | print( |
| 284 | "❌ Provide at least one --good commit: " |
| 285 | "muse bisect start --bad HEAD --good <ref>", |
| 286 | file=sys.stderr, |
| 287 | ) |
| 288 | raise SystemExit(ExitCode.USER_ERROR) |
| 289 | |
| 290 | branch = read_current_branch(root) |
| 291 | result = start_bisect(root, bad_id, good_ids, branch=branch, symbol_filter=symbol or "") |
| 292 | |
| 293 | if json_out: |
| 294 | print(json.dumps(_BisectStepJson(**make_envelope(elapsed), **_result_to_json(result)))) |
| 295 | return |
| 296 | |
| 297 | symbol_msg = f" symbol={sanitize_display(symbol)}" if symbol else "" |
| 298 | print( |
| 299 | f"Bisect session started. bad={bad_id} " |
| 300 | f"good=[{', '.join(good_ids)}]{symbol_msg}" |
| 301 | ) |
| 302 | if symbol and result.remaining_count == 0 and not result.done: |
| 303 | print( |
| 304 | f"⚠️ No commits between bad and good touched symbol " |
| 305 | f"{sanitize_display(symbol)!r}." |
| 306 | ) |
| 307 | print(" Try bisecting without --symbol, or widen the bad/good range.") |
| 308 | _print_result(result) |
| 309 | |
| 310 | def run_bisect_bad(args: argparse.Namespace) -> None: |
| 311 | """Mark a commit as bad (the bug is present in this commit). |
| 312 | |
| 313 | Narrows the bisect search range by recording that the given commit |
| 314 | (default: HEAD) exhibits the regression. Muse updates the remaining |
| 315 | set and suggests the next midpoint to test. |
| 316 | |
| 317 | Agent quickstart |
| 318 | ---------------- |
| 319 | :: |
| 320 | |
| 321 | muse bisect bad --json |
| 322 | muse bisect bad a1b2c3 --json |
| 323 | |
| 324 | JSON fields |
| 325 | ----------- |
| 326 | done ``true`` when the first bad commit has been isolated. |
| 327 | first_bad Full commit ID of the first bad commit; ``null`` while in progress. |
| 328 | next_to_test Full commit ID of the next midpoint to test; ``null`` when done. |
| 329 | remaining_count Commits still in the candidate set. |
| 330 | steps_remaining Estimated binary-search steps left. |
| 331 | verdict ``"bad"`` for this subcommand. |
| 332 | symbol_changes List of symbol-op descriptions for the midpoint commit. |
| 333 | |
| 334 | Exit codes |
| 335 | ---------- |
| 336 | 0 Verdict recorded; search advanced (or ``done=true`` if isolated). |
| 337 | 1 No active bisect session, or ref not found. |
| 338 | 2 Not inside a Muse repository. |
| 339 | """ |
| 340 | elapsed = start_timer() |
| 341 | ref: str | None = args.ref |
| 342 | json_out: bool = args.json_out |
| 343 | |
| 344 | root = require_repo() |
| 345 | if not is_bisect_active(root): |
| 346 | print("❌ No bisect session in progress. Run 'muse bisect start' first.", file=sys.stderr) |
| 347 | raise SystemExit(ExitCode.USER_ERROR) |
| 348 | commit_id = _resolve_ref(root, ref) |
| 349 | result = mark_bad(root, commit_id) |
| 350 | |
| 351 | if json_out: |
| 352 | print(json.dumps(_BisectStepJson(**make_envelope(elapsed), **_result_to_json(result)))) |
| 353 | return |
| 354 | |
| 355 | print(f"Marked {commit_id} as bad.") |
| 356 | _print_result(result) |
| 357 | |
| 358 | def run_bisect_good(args: argparse.Namespace) -> None: |
| 359 | """Mark a commit as good (the bug is absent in this commit). |
| 360 | |
| 361 | Narrows the bisect search range — Muse updates the candidate set and |
| 362 | suggests the next midpoint to test. |
| 363 | |
| 364 | Agent quickstart |
| 365 | ---------------- |
| 366 | :: |
| 367 | |
| 368 | muse bisect good --json |
| 369 | muse bisect good a1b2c3 --json |
| 370 | |
| 371 | JSON fields |
| 372 | ----------- |
| 373 | done ``true`` when the first bad commit has been isolated. |
| 374 | first_bad Full commit ID of the first bad commit; ``null`` while in progress. |
| 375 | next_to_test Full commit ID of the next midpoint to test; ``null`` when done. |
| 376 | remaining_count Commits still in the candidate set. |
| 377 | steps_remaining Estimated binary-search steps left. |
| 378 | verdict ``"good"`` for this subcommand. |
| 379 | symbol_changes List of symbol-op descriptions for the midpoint commit. |
| 380 | |
| 381 | Exit codes |
| 382 | ---------- |
| 383 | 0 Verdict recorded; search advanced (or ``done=true`` if isolated). |
| 384 | 1 No active bisect session, or ref not found. |
| 385 | 2 Not inside a Muse repository. |
| 386 | """ |
| 387 | elapsed = start_timer() |
| 388 | ref: str | None = args.ref |
| 389 | json_out: bool = args.json_out |
| 390 | |
| 391 | root = require_repo() |
| 392 | if not is_bisect_active(root): |
| 393 | print("❌ No bisect session in progress. Run 'muse bisect start' first.", file=sys.stderr) |
| 394 | raise SystemExit(ExitCode.USER_ERROR) |
| 395 | commit_id = _resolve_ref(root, ref) |
| 396 | result = mark_good(root, commit_id) |
| 397 | |
| 398 | if json_out: |
| 399 | print(json.dumps(_BisectStepJson(**make_envelope(elapsed), **_result_to_json(result)))) |
| 400 | return |
| 401 | |
| 402 | print(f"Marked {commit_id} as good.") |
| 403 | _print_result(result) |
| 404 | |
| 405 | def run_bisect_skip(args: argparse.Namespace) -> None: |
| 406 | """Skip a commit that cannot be tested (e.g. fails to build). |
| 407 | |
| 408 | Muse excludes it from the remaining set and suggests the next midpoint. |
| 409 | In ``muse bisect run`` mode, exit code 125 from the test script triggers |
| 410 | this automatically. |
| 411 | |
| 412 | Agent quickstart |
| 413 | ---------------- |
| 414 | :: |
| 415 | |
| 416 | muse bisect skip --json |
| 417 | muse bisect skip a1b2c3 --json |
| 418 | |
| 419 | JSON fields |
| 420 | ----------- |
| 421 | done ``true`` when the first bad commit has been isolated. |
| 422 | first_bad Full commit ID of the first bad commit; ``null`` while in progress. |
| 423 | next_to_test Full commit ID of the next midpoint to test; ``null`` when done. |
| 424 | remaining_count Commits still in the candidate set. |
| 425 | steps_remaining Estimated binary-search steps left. |
| 426 | verdict ``"skip"`` for this subcommand. |
| 427 | symbol_changes List of symbol-op descriptions for the midpoint commit. |
| 428 | |
| 429 | Exit codes |
| 430 | ---------- |
| 431 | 0 Commit skipped; search advanced (or ``done=true`` if isolated). |
| 432 | 1 No active bisect session, or ref not found. |
| 433 | 2 Not inside a Muse repository. |
| 434 | """ |
| 435 | elapsed = start_timer() |
| 436 | ref: str | None = args.ref |
| 437 | json_out: bool = args.json_out |
| 438 | |
| 439 | root = require_repo() |
| 440 | if not is_bisect_active(root): |
| 441 | print("❌ No bisect session in progress. Run 'muse bisect start' first.", file=sys.stderr) |
| 442 | raise SystemExit(ExitCode.USER_ERROR) |
| 443 | commit_id = _resolve_ref(root, ref) |
| 444 | result = skip_commit(root, commit_id) |
| 445 | |
| 446 | if json_out: |
| 447 | print(json.dumps(_BisectStepJson(**make_envelope(elapsed), **_result_to_json(result)))) |
| 448 | return |
| 449 | |
| 450 | print(f"Skipped {commit_id}.") |
| 451 | _print_result(result) |
| 452 | |
| 453 | def run_bisect_run(args: argparse.Namespace) -> None: |
| 454 | """Automatically bisect by running a command at each step. |
| 455 | |
| 456 | The command exit code determines the verdict:: |
| 457 | |
| 458 | 0 → good |
| 459 | 125 → skip |
| 460 | else → bad |
| 461 | |
| 462 | The command is run in the repository root. Exit code ``0`` → good, |
| 463 | ``125`` → skip, anything else → bad. |
| 464 | |
| 465 | Agent quickstart |
| 466 | ---------------- |
| 467 | :: |
| 468 | |
| 469 | muse bisect run "pytest tests/test_regression.py -x -q" --json |
| 470 | muse bisect run "./check.sh" --json |
| 471 | |
| 472 | JSON fields (NDJSON — one object per step, then a final summary) |
| 473 | ---------------------------------------------------------------- |
| 474 | Per-step fields: |
| 475 | |
| 476 | step Step number (1-based). |
| 477 | testing Full commit ID being tested. |
| 478 | verdict ``"good"``, ``"bad"``, or ``"skip"``. |
| 479 | remaining_count Candidates remaining after this step. |
| 480 | done ``true`` on the final step. |
| 481 | symbol_changes Symbol-op descriptions for the commit tested. |
| 482 | |
| 483 | Final summary fields: |
| 484 | |
| 485 | done Always ``true``. |
| 486 | first_bad Full commit ID of the first bad commit; ``null`` if none isolated. |
| 487 | steps_taken Total number of steps executed. |
| 488 | |
| 489 | Exit codes |
| 490 | ---------- |
| 491 | 0 Bisect run completed; first bad commit identified or session exhausted. |
| 492 | 1 No active bisect session. |
| 493 | 2 Not inside a Muse repository. |
| 494 | """ |
| 495 | elapsed = start_timer() |
| 496 | command: str = args.command |
| 497 | json_out: bool = args.json_out |
| 498 | timeout: int | None = args.timeout |
| 499 | |
| 500 | root = require_repo() |
| 501 | if not is_bisect_active(root): |
| 502 | print("❌ No bisect session in progress. Run 'muse bisect start' first.", file=sys.stderr) |
| 503 | raise SystemExit(ExitCode.USER_ERROR) |
| 504 | |
| 505 | step = 0 |
| 506 | while True: |
| 507 | current, symbol_filter = get_bisect_next(root) |
| 508 | if current is None: |
| 509 | if json_out: |
| 510 | # NDJSON — keep compact (one line per record) |
| 511 | print(json.dumps(_BisectRunDoneJson(done=True, first_bad=None, steps_taken=step))) |
| 512 | else: |
| 513 | print("✅ Bisect complete. Run 'muse bisect reset' to end.") |
| 514 | return |
| 515 | |
| 516 | # Collect symbol changes before running the command (they describe the |
| 517 | # commit we are about to test, not the result of the test). |
| 518 | changes: list[str] = [] |
| 519 | if symbol_filter: |
| 520 | changes = _symbol_ops_in_commit(root, current, symbol_filter) |
| 521 | |
| 522 | if not json_out: |
| 523 | print(f" → Testing {current} …") |
| 524 | if changes: |
| 525 | print(" Symbol changes:") |
| 526 | for line in changes: |
| 527 | print(f" {sanitize_display(line)}") |
| 528 | |
| 529 | result = run_bisect_command(root, command, current, timeout=timeout) |
| 530 | step += 1 |
| 531 | |
| 532 | if json_out: |
| 533 | # NDJSON — keep compact (one line per step record) |
| 534 | print(json.dumps(_BisectRunStepJson( |
| 535 | **make_envelope(elapsed), |
| 536 | step=step, |
| 537 | testing=current, |
| 538 | verdict=result.verdict, |
| 539 | remaining_count=result.remaining_count, |
| 540 | done=result.done, |
| 541 | symbol_changes=[sanitize_display(s) for s in changes], |
| 542 | ))) |
| 543 | else: |
| 544 | print(f" verdict: {result.verdict}") |
| 545 | |
| 546 | if result.done: |
| 547 | if json_out: |
| 548 | # NDJSON — keep compact |
| 549 | print(json.dumps( |
| 550 | _BisectRunDoneJson( |
| 551 | done=True, |
| 552 | first_bad=result.first_bad, |
| 553 | steps_taken=step, |
| 554 | ) |
| 555 | )) |
| 556 | else: |
| 557 | print(f"\n✅ First bad commit: {sanitize_display(result.first_bad or '')}") |
| 558 | return |
| 559 | |
| 560 | def run_bisect_log(args: argparse.Namespace) -> None: |
| 561 | """Show the full bisect session log. |
| 562 | |
| 563 | Displays every verdict applied so far (oldest first). Works whether or |
| 564 | not a session is currently active — returns an empty list when no session |
| 565 | has been started. |
| 566 | |
| 567 | Agent quickstart |
| 568 | ---------------- |
| 569 | :: |
| 570 | |
| 571 | muse bisect log --json |
| 572 | |
| 573 | JSON fields |
| 574 | ----------- |
| 575 | active ``true`` when a bisect session is in progress. |
| 576 | entries List of log entry objects (oldest first). |
| 577 | |
| 578 | Each entry: |
| 579 | |
| 580 | commit_id Commit ID string. |
| 581 | verdict ``"bad"``, ``"good"``, or ``"skip"``. |
| 582 | timestamp ISO-8601 timestamp of the verdict. |
| 583 | |
| 584 | Exit codes |
| 585 | ---------- |
| 586 | 0 Always (empty entries when no session exists). |
| 587 | 2 Not inside a Muse repository. |
| 588 | """ |
| 589 | elapsed = start_timer() |
| 590 | json_out: bool = args.json_out |
| 591 | |
| 592 | root = require_repo() |
| 593 | entries = get_bisect_log(root) |
| 594 | active = is_bisect_active(root) |
| 595 | |
| 596 | if json_out: |
| 597 | print(json.dumps(_BisectLogJson(**make_envelope(elapsed), active=active, entries=[_parse_log_entry(e) for e in entries]))) |
| 598 | return |
| 599 | |
| 600 | if not entries: |
| 601 | print("No bisect log. Start a session with 'muse bisect start'.") |
| 602 | return |
| 603 | print("Bisect log:") |
| 604 | for entry in entries: |
| 605 | print(f" {sanitize_display(entry)}") |
| 606 | |
| 607 | def run_bisect_reset(args: argparse.Namespace) -> None: |
| 608 | """End the bisect session and remove all bisect state. |
| 609 | |
| 610 | Idempotent — safe to call whether or not a session is active. |
| 611 | |
| 612 | Agent quickstart |
| 613 | ---------------- |
| 614 | :: |
| 615 | |
| 616 | muse bisect reset --json |
| 617 | |
| 618 | JSON fields |
| 619 | ----------- |
| 620 | reset Always ``true``. |
| 621 | |
| 622 | Exit codes |
| 623 | ---------- |
| 624 | 0 Always (state removed if it existed, no-op otherwise). |
| 625 | 2 Not inside a Muse repository. |
| 626 | """ |
| 627 | elapsed = start_timer() |
| 628 | json_out: bool = args.json_out |
| 629 | |
| 630 | root = require_repo() |
| 631 | reset_bisect(root) |
| 632 | |
| 633 | if json_out: |
| 634 | print(json.dumps(_BisectResetJson(**make_envelope(elapsed), reset=True))) |
| 635 | return |
| 636 | |
| 637 | print("Bisect session reset.") |
| 638 | |
| 639 | def run_bisect_status(args: argparse.Namespace) -> None: |
| 640 | """Report the current bisect session state without modifying anything. |
| 641 | |
| 642 | Read-only snapshot of the active session. Call this at the start of any |
| 643 | bisect workflow to discover whether a session is already running. |
| 644 | |
| 645 | Agent quickstart |
| 646 | ---------------- |
| 647 | :: |
| 648 | |
| 649 | muse bisect status --json |
| 650 | |
| 651 | JSON fields |
| 652 | ----------- |
| 653 | active ``true`` when a bisect session is in progress; ``false`` otherwise. |
| 654 | |
| 655 | When ``active`` is ``true``: |
| 656 | |
| 657 | bad_id Full commit ID of the known-bad commit. |
| 658 | good_ids List of known-good commit IDs. |
| 659 | symbol_filter Symbol address filter; empty string if none. |
| 660 | remaining_count Commits still in the candidate set. |
| 661 | steps_remaining Estimated binary-search steps left. |
| 662 | skipped_count Number of commits skipped so far. |
| 663 | branch Branch the session was started on. |
| 664 | |
| 665 | Exit codes |
| 666 | ---------- |
| 667 | 0 Always (``active=false`` when no session exists). |
| 668 | 2 Not inside a Muse repository. |
| 669 | """ |
| 670 | elapsed = start_timer() |
| 671 | json_out: bool = args.json_out |
| 672 | |
| 673 | root = require_repo() |
| 674 | status = get_bisect_status(root) |
| 675 | |
| 676 | if json_out: |
| 677 | print(json.dumps(_BisectStatusJson(**make_envelope(elapsed), **dict(status)))) |
| 678 | return |
| 679 | |
| 680 | if not status.get("active"): |
| 681 | print("No bisect session active. Start one with 'muse bisect start'.") |
| 682 | return |
| 683 | |
| 684 | print("Bisect session active.") |
| 685 | bad = status.get("bad_id", "") |
| 686 | print(f" bad: {sanitize_display(bad)}") |
| 687 | good_ids = status.get("good_ids", []) |
| 688 | good_display = ", ".join(sanitize_display(g) for g in good_ids) |
| 689 | print(f" good: {good_display}") |
| 690 | sym = status.get("symbol_filter", "") |
| 691 | if sym: |
| 692 | print(f" symbol: {sanitize_display(sym)}") |
| 693 | remaining = status.get("remaining_count", 0) |
| 694 | steps = status.get("steps_remaining", 0) |
| 695 | print(f" remaining: {remaining} commit(s) (~{steps} step(s))") |
| 696 | print(f" skipped: {status.get('skipped_count', 0)}") |
| 697 | branch = status.get("branch", "") |
| 698 | if branch: |
| 699 | print(f" branch: {sanitize_display(branch)}") |
| 700 | |
| 701 | # --------------------------------------------------------------------------- |
| 702 | # Registration |
| 703 | # --------------------------------------------------------------------------- |
| 704 | |
| 705 | def register(subparsers: "argparse._SubParsersAction[argparse.ArgumentParser]") -> None: |
| 706 | """Register the ``bisect`` subcommand.""" |
| 707 | parser = subparsers.add_parser( |
| 708 | "bisect", |
| 709 | help="Binary search through commit history to find regressions.", |
| 710 | description=__doc__, |
| 711 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 712 | ) |
| 713 | subs = parser.add_subparsers(dest="subcommand", metavar="SUBCOMMAND") |
| 714 | subs.required = True |
| 715 | |
| 716 | # ── bad ─────────────────────────────────────────────────────────────────── |
| 717 | bad_p = subs.add_parser( |
| 718 | "bad", |
| 719 | help="Mark a commit as bad (bug present).", |
| 720 | description=( |
| 721 | "Record that the given commit (default: HEAD) exhibits the\n" |
| 722 | "regression. Muse narrows the search range and suggests the\n" |
| 723 | "next midpoint to test.\n\n" |
| 724 | "Agent quickstart\n" |
| 725 | "----------------\n" |
| 726 | " muse bisect bad --json\n" |
| 727 | " muse bisect bad <commit-id> --json\n\n" |
| 728 | "JSON output schema\n" |
| 729 | "------------------\n" |
| 730 | ' {"done": false, "first_bad": null, "next_to_test": "<commit-id>",\n' |
| 731 | ' "remaining_count": <int>, "steps_remaining": <int>,\n' |
| 732 | ' "verdict": "bad", "symbol_changes": [...]}\n\n' |
| 733 | "Exit codes\n" |
| 734 | "----------\n" |
| 735 | " 0 — verdict recorded (done=true when first bad commit isolated)\n" |
| 736 | " 1 — no active bisect session, or ref not found\n" |
| 737 | " 2 — not inside a Muse repository\n" |
| 738 | ), |
| 739 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 740 | ) |
| 741 | bad_p.add_argument( |
| 742 | "ref", nargs="?", default=None, metavar="REF", |
| 743 | help="Commit to mark bad (default: HEAD).", |
| 744 | ) |
| 745 | bad_p.add_argument( |
| 746 | "--json", "-j", action="store_true", dest="json_out", default=False, |
| 747 | help="Emit machine-readable JSON to stdout.", |
| 748 | ) |
| 749 | bad_p.set_defaults(func=run_bisect_bad) |
| 750 | |
| 751 | # ── good ────────────────────────────────────────────────────────────────── |
| 752 | good_p = subs.add_parser( |
| 753 | "good", |
| 754 | help="Mark a commit as good (bug absent).", |
| 755 | description=( |
| 756 | "Record that the given commit (default: HEAD) does not exhibit\n" |
| 757 | "the regression. Muse narrows the search range and suggests the\n" |
| 758 | "next midpoint to test.\n\n" |
| 759 | "Agent quickstart\n" |
| 760 | "----------------\n" |
| 761 | " muse bisect good --json\n" |
| 762 | " muse bisect good <commit-id> --json\n\n" |
| 763 | "JSON output schema\n" |
| 764 | "------------------\n" |
| 765 | ' {"done": false, "first_bad": null, "next_to_test": "<commit-id>",\n' |
| 766 | ' "remaining_count": <int>, "steps_remaining": <int>,\n' |
| 767 | ' "verdict": "good", "symbol_changes": [...]}\n\n' |
| 768 | "Exit codes\n" |
| 769 | "----------\n" |
| 770 | " 0 — verdict recorded (done=true when first bad commit isolated)\n" |
| 771 | " 1 — no active bisect session, or ref not found\n" |
| 772 | " 2 — not inside a Muse repository\n" |
| 773 | ), |
| 774 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 775 | ) |
| 776 | good_p.add_argument( |
| 777 | "ref", nargs="?", default=None, metavar="REF", |
| 778 | help="Commit to mark good (default: HEAD).", |
| 779 | ) |
| 780 | good_p.add_argument( |
| 781 | "--json", "-j", action="store_true", dest="json_out", default=False, |
| 782 | help="Emit machine-readable JSON to stdout.", |
| 783 | ) |
| 784 | good_p.set_defaults(func=run_bisect_good) |
| 785 | |
| 786 | # ── log ─────────────────────────────────────────────────────────────────── |
| 787 | log_p = subs.add_parser( |
| 788 | "log", |
| 789 | help="Show the bisect session log.", |
| 790 | description=( |
| 791 | "Display every verdict applied in the current (or most recent)\n" |
| 792 | "bisect session, oldest first. Each entry is a space-separated\n" |
| 793 | "triple: <commit-id> <verdict> <timestamp>.\n\n" |
| 794 | "Agent quickstart\n" |
| 795 | "----------------\n" |
| 796 | " muse bisect log --json\n\n" |
| 797 | "JSON output schema\n" |
| 798 | "------------------\n" |
| 799 | ' {"active": true|false,\n' |
| 800 | ' "entries": [{"commit_id": "<id>", "verdict": "bad|good|skip",\n' |
| 801 | ' "timestamp": "<ISO8601>"}, ...]}\n\n' |
| 802 | "Exit codes\n" |
| 803 | "----------\n" |
| 804 | " 0 — always (empty entries list when no session exists)\n" |
| 805 | " 2 — not inside a Muse repository\n" |
| 806 | ), |
| 807 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 808 | ) |
| 809 | log_p.add_argument( |
| 810 | "--json", "-j", action="store_true", dest="json_out", default=False, |
| 811 | help="Emit machine-readable JSON to stdout.", |
| 812 | ) |
| 813 | log_p.set_defaults(func=run_bisect_log) |
| 814 | |
| 815 | # ── reset ───────────────────────────────────────────────────────────────── |
| 816 | reset_p = subs.add_parser( |
| 817 | "reset", |
| 818 | help="End the bisect session and clean up state.", |
| 819 | description=( |
| 820 | "Remove all bisect state. Idempotent — safe to call whether or\n" |
| 821 | "not a session is active. After reset, bad/good/skip commands\n" |
| 822 | "will refuse until a new session is started with 'muse bisect\n" |
| 823 | "start'.\n\n" |
| 824 | "Agent quickstart\n" |
| 825 | "----------------\n" |
| 826 | " muse bisect reset --json\n\n" |
| 827 | "JSON output schema\n" |
| 828 | "------------------\n" |
| 829 | ' {"reset": true}\n\n' |
| 830 | "Exit codes\n" |
| 831 | "----------\n" |
| 832 | " 0 — always (no-op when no session is active)\n" |
| 833 | " 2 — not inside a Muse repository\n" |
| 834 | ), |
| 835 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 836 | ) |
| 837 | reset_p.add_argument( |
| 838 | "--json", "-j", action="store_true", dest="json_out", default=False, |
| 839 | help="Emit machine-readable JSON to stdout.", |
| 840 | ) |
| 841 | reset_p.set_defaults(func=run_bisect_reset) |
| 842 | |
| 843 | # ── run ─────────────────────────────────────────────────────────────────── |
| 844 | run_p = subs.add_parser( |
| 845 | "run", |
| 846 | help="Automatically bisect by running a command.", |
| 847 | description=( |
| 848 | "Run COMMAND at each bisect step; Muse interprets the exit code\n" |
| 849 | "and applies the verdict automatically until the first bad commit\n" |
| 850 | "is isolated. Exit codes: 0=good, 125=skip, anything else=bad.\n\n" |
| 851 | "Agent quickstart\n" |
| 852 | "----------------\n" |
| 853 | " muse bisect run 'pytest tests/test_regression.py -x -q' --json\n" |
| 854 | " muse bisect run './check.sh' --json\n\n" |
| 855 | "NDJSON output — one step line per commit, then a summary line\n" |
| 856 | "-----------------------------------------------------------\n" |
| 857 | ' step line: {"step":<int>, "testing":"<id>", "verdict":"good|bad|skip",\n' |
| 858 | ' "remaining_count":<int>, "done":false,\n' |
| 859 | ' "symbol_changes":[...]}\n' |
| 860 | ' done line: {"done":true, "first_bad":"<id>|null", "steps_taken":<int>}\n\n' |
| 861 | "Exit codes\n" |
| 862 | "----------\n" |
| 863 | " 0 — run complete (first bad isolated or session exhausted)\n" |
| 864 | " 1 — no active bisect session\n" |
| 865 | " 2 — not inside a Muse repository\n" |
| 866 | ), |
| 867 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 868 | ) |
| 869 | run_p.add_argument( |
| 870 | "command", metavar="COMMAND", |
| 871 | help="Shell command to run at each step (exit 0=good, 125=skip, else=bad).", |
| 872 | ) |
| 873 | run_p.add_argument( |
| 874 | "--json", "-j", action="store_true", dest="json_out", default=False, |
| 875 | help="Emit machine-readable NDJSON (one line per step, then a summary line).", |
| 876 | ) |
| 877 | run_p.add_argument( |
| 878 | "--timeout", "-t", type=int, default=None, metavar="SECONDS", |
| 879 | help=( |
| 880 | "Kill the test command after SECONDS and treat the commit as " |
| 881 | "untestable (skip, same as exit 125). Prevents a hanging command " |
| 882 | "from stalling an automated bisect." |
| 883 | ), |
| 884 | ) |
| 885 | run_p.set_defaults(func=run_bisect_run) |
| 886 | |
| 887 | # ── skip ────────────────────────────────────────────────────────────────── |
| 888 | skip_p = subs.add_parser( |
| 889 | "skip", |
| 890 | help="Skip an untestable commit.", |
| 891 | description=( |
| 892 | "Record that the given commit (default: HEAD) cannot be tested —\n" |
| 893 | "e.g. it fails to build. Muse excludes it from the remaining set\n" |
| 894 | "and suggests the next midpoint. In 'muse bisect run' mode, exit\n" |
| 895 | "code 125 from the test script triggers this automatically.\n\n" |
| 896 | "Agent quickstart\n" |
| 897 | "----------------\n" |
| 898 | " muse bisect skip --json\n" |
| 899 | " muse bisect skip <commit-id> --json\n\n" |
| 900 | "JSON output schema\n" |
| 901 | "------------------\n" |
| 902 | ' {"done": false, "first_bad": null, "next_to_test": "<commit-id>",\n' |
| 903 | ' "remaining_count": <int>, "steps_remaining": <int>,\n' |
| 904 | ' "verdict": "skip", "symbol_changes": [...]}\n\n' |
| 905 | "Exit codes\n" |
| 906 | "----------\n" |
| 907 | " 0 — commit skipped (done=true when first bad commit isolated)\n" |
| 908 | " 1 — no active bisect session, or ref not found\n" |
| 909 | " 2 — not inside a Muse repository\n" |
| 910 | ), |
| 911 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 912 | ) |
| 913 | skip_p.add_argument( |
| 914 | "ref", nargs="?", default=None, metavar="REF", |
| 915 | help="Commit to skip (default: HEAD).", |
| 916 | ) |
| 917 | skip_p.add_argument( |
| 918 | "--json", "-j", action="store_true", dest="json_out", default=False, |
| 919 | help="Emit machine-readable JSON to stdout.", |
| 920 | ) |
| 921 | skip_p.set_defaults(func=run_bisect_skip) |
| 922 | |
| 923 | # ── start ───────────────────────────────────────────────────────────────── |
| 924 | start_p = subs.add_parser( |
| 925 | "start", |
| 926 | help="Begin a bisect session.", |
| 927 | description=( |
| 928 | "Mark the known-bad and known-good commits, then let Muse binary-\n" |
| 929 | "search the commits between them. Muse immediately suggests the\n" |
| 930 | "midpoint commit to test. Use --symbol to restrict the search to\n" |
| 931 | "commits that touched a specific symbol.\n\n" |
| 932 | "Agent quickstart\n" |
| 933 | "----------------\n" |
| 934 | " muse bisect start --bad HEAD --good v1.0.0 --json\n" |
| 935 | " muse bisect start --bad HEAD --good v1.0.0 \\\n" |
| 936 | " --symbol billing.py::Invoice.compute_total --json\n\n" |
| 937 | "JSON output schema\n" |
| 938 | "------------------\n" |
| 939 | ' {"done": false, "first_bad": null, "next_to_test": "<commit-id>",\n' |
| 940 | ' "remaining_count": <int>, "steps_remaining": <int>,\n' |
| 941 | ' "verdict": "started", "symbol_changes": [...]}\n\n' |
| 942 | "Exit codes\n" |
| 943 | "----------\n" |
| 944 | " 0 — session started (or immediately resolved when bad/good adjacent)\n" |
| 945 | " 1 — session already active, bad --symbol, missing --good, or bad ref\n" |
| 946 | " 2 — not inside a Muse repository\n" |
| 947 | ), |
| 948 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 949 | ) |
| 950 | start_p.add_argument( |
| 951 | "--bad", default=None, metavar="REF", |
| 952 | help="Known-bad commit (default: HEAD).", |
| 953 | ) |
| 954 | start_p.add_argument( |
| 955 | "--good", nargs="*", default=None, metavar="REF", |
| 956 | help="Known-good commit(s). Repeat for multiple: --good v1.0 --good v0.9.", |
| 957 | ) |
| 958 | start_p.add_argument( |
| 959 | "--symbol", "-s", default=None, metavar="ADDR", |
| 960 | help=( |
| 961 | "Restrict search to commits that touched this symbol " |
| 962 | "(e.g. billing.py::Invoice.compute_total). Dramatically reduces " |
| 963 | "the number of steps when you already know which symbol regressed." |
| 964 | ), |
| 965 | ) |
| 966 | start_p.add_argument( |
| 967 | "--json", "-j", action="store_true", dest="json_out", default=False, |
| 968 | help="Emit machine-readable JSON to stdout.", |
| 969 | ) |
| 970 | start_p.set_defaults(func=run_bisect_start) |
| 971 | |
| 972 | # ── status ──────────────────────────────────────────────────────────────── |
| 973 | status_p = subs.add_parser( |
| 974 | "status", |
| 975 | help="Report the current bisect session state (read-only).", |
| 976 | description=( |
| 977 | "Return a structured read-only snapshot of the active bisect\n" |
| 978 | "session. Safe to call at any time — never modifies state.\n" |
| 979 | "Agents should use this instead of parsing 'muse bisect log'\n" |
| 980 | "to discover whether a session is running.\n\n" |
| 981 | "Agent quickstart\n" |
| 982 | "----------------\n" |
| 983 | " muse bisect status --json\n\n" |
| 984 | "JSON output schema (active)\n" |
| 985 | "---------------------------\n" |
| 986 | ' {"active": true, "bad_id": "<id>", "good_ids": ["<id>", ...],\n' |
| 987 | ' "symbol_filter": "", "remaining_count": <int>,\n' |
| 988 | ' "steps_remaining": <int>, "skipped_count": <int>,\n' |
| 989 | ' "branch": "<branch>"}\n\n' |
| 990 | "JSON output schema (no session)\n" |
| 991 | "-------------------------------\n" |
| 992 | ' {"active": false}\n\n' |
| 993 | "Exit codes\n" |
| 994 | "----------\n" |
| 995 | " 0 — always (active=false when no session exists)\n" |
| 996 | " 2 — not inside a Muse repository\n" |
| 997 | ), |
| 998 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 999 | ) |
| 1000 | status_p.add_argument( |
| 1001 | "--json", "-j", action="store_true", dest="json_out", default=False, |
| 1002 | help="Emit machine-readable JSON to stdout.", |
| 1003 | ) |
| 1004 | status_p.set_defaults(func=run_bisect_status) |
File History
1 commit
sha256:51ce277f663e01a43eaffbe77509b1de7ac2d4251b55d23306304bcdeb92c90d
feat(pack): delta-encode snapshots in MPackBundle wire format
Sonnet 4.6
minor
⚠
128 days ago