test_stress_merge_regression.py
python
sha256:be3641f35bdbcc094677776a77b9aa6a5dab891f8fab201dc162d03c2bab5aea
fix(read): strip position:null from structured_delta ops in…
Sonnet 4.6
patch
23 days ago
| 1 | """Regression stress tests for the three-way merge engine — all permutations. |
| 2 | |
| 3 | Root cause (fixed in commit 73427a30): |
| 4 | CodePlugin.merge_ops silently dropped theirs-only changes when OT symbol |
| 5 | commutation masked a file-level conflict. The merged blob was the ours blob |
| 6 | verbatim, so the result reported "clean merge" with no-op file content. |
| 7 | |
| 8 | Real-world impact: |
| 9 | MuseHub's executor.py ``--pid=private`` fix (removed in fix/pool-pre-ping) |
| 10 | was silently discarded when the user merged local/dev (which had a commuting |
| 11 | pool_pre_ping change to database.py). Every subsequent CI run failed with |
| 12 | "docker: --pid: invalid PID mode" until the regression was manually tracked |
| 13 | down through the object store. |
| 14 | |
| 15 | This file tests every permutation of merge topology that could lead to silent |
| 16 | data loss — not just the one that burned us. |
| 17 | |
| 18 | Categories |
| 19 | ---------- |
| 20 | A Fast-forward / up-to-date detection (no data-loss risk, but correctness) |
| 21 | B Three-way clean merges — no conflicts anywhere |
| 22 | C Three-way with conflicts surfaced — the merge MUST stop, not silently pass |
| 23 | D The silent-drop regression — commuting OT ops on same file |
| 24 | E Theirs-only files MUST survive when there are conflicts elsewhere |
| 25 | F Strategy shortcuts (--strategy=ours / --strategy=theirs) correctness |
| 26 | G MuseHub regression scenario (pool_pre_ping + executor + AGENTS.md) |
| 27 | H Merge-base correctness for complex DAG topologies |
| 28 | I False-conflict regression — theirs-only additions falsely reported as |
| 29 | conflicts and deleted from disk when ours-snapshot == base-snapshot. |
| 30 | Root cause: muse/core/patch_record.py was deleted from disk during a |
| 31 | dev→main merge where main's previous merge had left its snapshot |
| 32 | identical to the merge base. The engine must NEVER report a theirs-only |
| 33 | addition as a conflict, and must NEVER delete it from the working tree. |
| 34 | """ |
| 35 | from __future__ import annotations |
| 36 | |
| 37 | import datetime |
| 38 | import json |
| 39 | import pathlib |
| 40 | import textwrap |
| 41 | |
| 42 | import pytest |
| 43 | from tests.cli_test_helper import CliRunner |
| 44 | from muse.core.types import Manifest, blob_id, fake_id |
| 45 | from muse.core.paths import merge_state_path, muse_dir, ref_path |
| 46 | |
| 47 | runner = CliRunner() |
| 48 | cli = None # CliRunner ignores this positional arg |
| 49 | |
| 50 | |
| 51 | # --------------------------------------------------------------------------- |
| 52 | # Low-level repo helpers |
| 53 | # --------------------------------------------------------------------------- |
| 54 | |
| 55 | |
| 56 | def _h(label: str) -> str: |
| 57 | """Stable fake content hash for a text label (sha256: prefixed).""" |
| 58 | return fake_id(label) |
| 59 | |
| 60 | |
| 61 | def _env(root: pathlib.Path) -> Manifest: |
| 62 | return {"MUSE_REPO_ROOT": str(root)} |
| 63 | |
| 64 | |
| 65 | def _run(root: pathlib.Path, *args: str) -> tuple[int, str]: |
| 66 | """Run a muse command, injecting --force into merge calls. |
| 67 | |
| 68 | Tests use an in-memory manifest-only setup (no files on disk) so the |
| 69 | working-tree cleanliness guard would always fire. ``--force`` bypasses |
| 70 | that guard without affecting any merge-logic correctness being tested. |
| 71 | """ |
| 72 | final_args = list(args) |
| 73 | if final_args and final_args[0] == "merge" and "--force" not in final_args: |
| 74 | final_args.insert(1, "--force") |
| 75 | result = runner.invoke(cli, final_args, env=_env(root), catch_exceptions=False) |
| 76 | return result.exit_code, result.output |
| 77 | |
| 78 | |
| 79 | def _run_unchecked(root: pathlib.Path, *args: str) -> tuple[int, str]: |
| 80 | """Like _run but does not raise on failure.""" |
| 81 | final_args = list(args) |
| 82 | if final_args and final_args[0] == "merge" and "--force" not in final_args: |
| 83 | final_args.insert(1, "--force") |
| 84 | result = runner.invoke(cli, final_args, env=_env(root)) |
| 85 | return result.exit_code, result.output |
| 86 | |
| 87 | |
| 88 | def _write_object(root: pathlib.Path, content: bytes) -> str: |
| 89 | """Write content to object store and return sha256:-prefixed object ID.""" |
| 90 | from muse.core.object_store import write_object as _store_write |
| 91 | oid = blob_id(content) |
| 92 | _store_write(root, oid, content) |
| 93 | return oid |
| 94 | |
| 95 | |
| 96 | def _init_code_repo(tmp_path: pathlib.Path) -> tuple[pathlib.Path, str]: |
| 97 | """Initialise a bare code-domain repo and return (root, repo_id).""" |
| 98 | from muse.core.commits import ( |
| 99 | CommitRecord, |
| 100 | write_commit, |
| 101 | ) |
| 102 | from muse.core.snapshots import ( |
| 103 | SnapshotRecord, |
| 104 | write_snapshot, |
| 105 | ) |
| 106 | from muse.core.ids import hash_snapshot as compute_snapshot_id, hash_commit as compute_commit_id |
| 107 | |
| 108 | dot_muse = muse_dir(tmp_path) |
| 109 | dot_muse.mkdir() |
| 110 | repo_id = fake_id("repo") |
| 111 | (dot_muse / "repo.json").write_text(json.dumps({ |
| 112 | "repo_id": repo_id, |
| 113 | "domain": "code", |
| 114 | "default_branch": "main", |
| 115 | "created_at": "2025-01-01T00:00:00+00:00", |
| 116 | }), encoding="utf-8") |
| 117 | (dot_muse / "HEAD").write_text("ref: refs/heads/main", encoding="utf-8") |
| 118 | (dot_muse / "refs" / "heads").mkdir(parents=True) |
| 119 | (dot_muse / "snapshots").mkdir() |
| 120 | (dot_muse / "commits").mkdir() |
| 121 | (dot_muse / "objects").mkdir() |
| 122 | return tmp_path, repo_id |
| 123 | |
| 124 | |
| 125 | def _make_commit( |
| 126 | root: pathlib.Path, |
| 127 | repo_id: str, |
| 128 | branch: str = "main", |
| 129 | message: str = "test", |
| 130 | manifest: Manifest | None = None, |
| 131 | parent_commit_id: str | None = None, |
| 132 | parent2_commit_id: str | None = None, |
| 133 | ) -> str: |
| 134 | """Write a snapshot + commit and advance the branch ref.""" |
| 135 | from muse.core.commits import ( |
| 136 | CommitRecord, |
| 137 | write_commit, |
| 138 | ) |
| 139 | from muse.core.snapshots import ( |
| 140 | SnapshotRecord, |
| 141 | write_snapshot, |
| 142 | ) |
| 143 | from muse.core.ids import hash_snapshot as compute_snapshot_id, hash_commit as compute_commit_id |
| 144 | |
| 145 | ref_file = ref_path(root, branch) |
| 146 | if parent_commit_id is None and ref_file.exists(): |
| 147 | parent_commit_id = ref_file.read_text().strip() or None |
| 148 | |
| 149 | m = manifest or {} |
| 150 | snap_id = compute_snapshot_id(m) |
| 151 | committed_at = datetime.datetime.now(datetime.timezone.utc) |
| 152 | parent_ids: list[str] = [] |
| 153 | if parent_commit_id: |
| 154 | parent_ids.append(parent_commit_id) |
| 155 | if parent2_commit_id: |
| 156 | parent_ids.append(parent2_commit_id) |
| 157 | commit_id = compute_commit_id( |
| 158 | parent_ids=parent_ids, |
| 159 | snapshot_id=snap_id, |
| 160 | message=message, |
| 161 | committed_at_iso=committed_at.isoformat(), |
| 162 | ) |
| 163 | write_snapshot(root, SnapshotRecord(snapshot_id=snap_id, manifest=m)) |
| 164 | write_commit(root, CommitRecord( |
| 165 | commit_id=commit_id, |
| 166 | branch=branch, |
| 167 | snapshot_id=snap_id, |
| 168 | message=message, |
| 169 | committed_at=committed_at, |
| 170 | parent_commit_id=parent_commit_id, |
| 171 | parent2_commit_id=parent2_commit_id, |
| 172 | )) |
| 173 | ref_file.parent.mkdir(parents=True, exist_ok=True) |
| 174 | ref_file.write_text(commit_id, encoding="utf-8") |
| 175 | return commit_id |
| 176 | |
| 177 | |
| 178 | def _write_py(root: pathlib.Path, filename: str, content: str) -> str: |
| 179 | """Write Python content into the object store ONLY; return object_id. |
| 180 | |
| 181 | We deliberately do NOT write the file to the working tree so that |
| 182 | ``require_clean_workdir`` never aborts the merge due to uncommitted |
| 183 | changes. The code plugin reads file bytes from the object store via |
| 184 | ``read_object(root, obj_id)``, so on-disk presence is not required. |
| 185 | """ |
| 186 | return _write_object(root, content.encode()) |
| 187 | |
| 188 | |
| 189 | def _ref(root: pathlib.Path, branch: str) -> str: |
| 190 | return (ref_path(root, branch)).read_text(encoding="utf-8").strip() |
| 191 | |
| 192 | |
| 193 | def _snapshot_manifest(root: pathlib.Path, branch: str) -> Manifest: |
| 194 | """Return the manifest for a branch's current HEAD snapshot.""" |
| 195 | from muse.core.commits import read_commit |
| 196 | from muse.core.snapshots import read_snapshot |
| 197 | commit_id = _ref(root, branch) |
| 198 | commit = read_commit(root, commit_id) |
| 199 | assert commit is not None |
| 200 | snap = read_snapshot(root, commit.snapshot_id) |
| 201 | assert snap is not None |
| 202 | return snap.manifest |
| 203 | |
| 204 | |
| 205 | # =========================================================================== |
| 206 | # A — Fast-forward / up-to-date |
| 207 | # =========================================================================== |
| 208 | |
| 209 | |
| 210 | class TestMergeTopologyA: |
| 211 | """Ensure merge base detection is correct and no data is corrupted.""" |
| 212 | |
| 213 | def test_A1_fast_forward_updates_head_and_files(self, tmp_path: pathlib.Path) -> None: |
| 214 | """A1: ours is ancestor of theirs → fast-forward, working tree = theirs.""" |
| 215 | root, repo_id = _init_code_repo(tmp_path) |
| 216 | a_id = _write_py(root, "app.py", "x = 1\n") |
| 217 | _make_commit(root, repo_id, branch="main", message="base", |
| 218 | manifest={"app.py": a_id}) |
| 219 | base_commit = _ref(root, "main") |
| 220 | |
| 221 | # Create feature branch from same base. |
| 222 | (ref_path(root, "feat")).write_text(base_commit) |
| 223 | b_id = _write_py(root, "app.py", "x = 2\n") |
| 224 | _make_commit(root, repo_id, branch="feat", message="feat commit", |
| 225 | manifest={"app.py": b_id}) |
| 226 | |
| 227 | code, out = _run(root, "merge", "feat") |
| 228 | assert code == 0, out |
| 229 | # main HEAD must now equal feat HEAD. |
| 230 | assert _ref(root, "main") == _ref(root, "feat") |
| 231 | # Manifest must equal feat's snapshot. |
| 232 | assert _snapshot_manifest(root, "main") == {"app.py": b_id} |
| 233 | |
| 234 | def test_A2_already_up_to_date_prints_message(self, tmp_path: pathlib.Path) -> None: |
| 235 | """A2: theirs is ancestor of ours → 'Already up to date.'""" |
| 236 | root, repo_id = _init_code_repo(tmp_path) |
| 237 | a_id = _write_py(root, "f.py", "a = 1\n") |
| 238 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 239 | manifest={"f.py": a_id}) |
| 240 | (ref_path(root, "old")).write_text(base_c) |
| 241 | b_id = _write_py(root, "f.py", "a = 2\n") |
| 242 | _make_commit(root, repo_id, branch="main", message="advance", |
| 243 | manifest={"f.py": b_id}) |
| 244 | |
| 245 | code, out = _run(root, "merge", "old") |
| 246 | assert code == 0, out |
| 247 | assert "up to date" in out.lower() |
| 248 | # main must not have moved back. |
| 249 | assert _snapshot_manifest(root, "main") == {"f.py": b_id} |
| 250 | |
| 251 | def test_A3_fast_forward_json_reports_fast_forward_status(self, tmp_path: pathlib.Path) -> None: |
| 252 | """A3: JSON output for fast-forward has status='fast_forward'.""" |
| 253 | root, repo_id = _init_code_repo(tmp_path) |
| 254 | a_id = _write_py(root, "f.py", "a = 1\n") |
| 255 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 256 | manifest={"f.py": a_id}) |
| 257 | (ref_path(root, "feat")).write_text(base_c) |
| 258 | b_id = _write_py(root, "f.py", "a = 2\n") |
| 259 | _make_commit(root, repo_id, branch="feat", message="feat", |
| 260 | manifest={"f.py": b_id}) |
| 261 | |
| 262 | code, out = _run(root, "merge", "--json", "feat") |
| 263 | assert code == 0, out |
| 264 | data = json.loads(out) |
| 265 | assert data["status"] == "fast_forward" |
| 266 | assert data["conflicts"] == [] |
| 267 | |
| 268 | def test_A4_fast_forward_preserves_all_theirs_files(self, tmp_path: pathlib.Path) -> None: |
| 269 | """A4: fast-forward with 50 files — all must appear in main's manifest.""" |
| 270 | root, repo_id = _init_code_repo(tmp_path) |
| 271 | a_id = _write_py(root, "base.py", "base = True\n") |
| 272 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 273 | manifest={"base.py": a_id}) |
| 274 | (ref_path(root, "feat")).write_text(base_c) |
| 275 | |
| 276 | manifest: Manifest = {"base.py": a_id} |
| 277 | for i in range(50): |
| 278 | oid = _write_py(root, f"module_{i:02d}.py", f"x_{i} = {i}\n") |
| 279 | manifest[f"module_{i:02d}.py"] = oid |
| 280 | _make_commit(root, repo_id, branch="feat", message="many files", |
| 281 | manifest=manifest) |
| 282 | |
| 283 | code, _ = _run(root, "merge", "feat") |
| 284 | assert code == 0 |
| 285 | merged = _snapshot_manifest(root, "main") |
| 286 | for i in range(50): |
| 287 | assert f"module_{i:02d}.py" in merged, f"module_{i:02d}.py missing after fast-forward" |
| 288 | |
| 289 | def test_A5_no_ff_creates_merge_commit(self, tmp_path: pathlib.Path) -> None: |
| 290 | """A5: --no-ff skips fast-forward and always creates a merge commit.""" |
| 291 | root, repo_id = _init_code_repo(tmp_path) |
| 292 | a_id = _write_py(root, "f.py", "a = 1\n") |
| 293 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 294 | manifest={"f.py": a_id}) |
| 295 | (ref_path(root, "feat")).write_text(base_c) |
| 296 | b_id = _write_py(root, "f.py", "a = 2\n") |
| 297 | feat_c = _make_commit(root, repo_id, branch="feat", message="feat", |
| 298 | manifest={"f.py": b_id}) |
| 299 | |
| 300 | from muse.core.commits import read_commit |
| 301 | pre_main = _ref(root, "main") |
| 302 | code, out = _run(root, "merge", "--no-ff", "feat") |
| 303 | assert code == 0, out |
| 304 | post_main = _ref(root, "main") |
| 305 | # HEAD must have advanced (new merge commit created). |
| 306 | assert post_main != pre_main |
| 307 | # The new commit must have TWO parents. |
| 308 | commit = read_commit(root, post_main) |
| 309 | assert commit is not None |
| 310 | assert commit.parent2_commit_id is not None, "no-ff must create merge commit with 2 parents" |
| 311 | |
| 312 | |
| 313 | # =========================================================================== |
| 314 | # B — Three-way clean merges (no conflicts anywhere) |
| 315 | # =========================================================================== |
| 316 | |
| 317 | |
| 318 | class TestThreeWayCleanMergeB: |
| 319 | """Theirs-only and ours-only changes all survive; merged snapshot is correct.""" |
| 320 | |
| 321 | def test_B1_disjoint_file_changes_both_survive(self, tmp_path: pathlib.Path) -> None: |
| 322 | """B1: ours changes a.py, theirs changes b.py — both must be in merged.""" |
| 323 | root, repo_id = _init_code_repo(tmp_path) |
| 324 | a0 = _write_py(root, "a.py", "a = 0\n") |
| 325 | b0 = _write_py(root, "b.py", "b = 0\n") |
| 326 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 327 | manifest={"a.py": a0, "b.py": b0}) |
| 328 | (ref_path(root, "feat")).write_text(base_c) |
| 329 | |
| 330 | # ours: modify a.py |
| 331 | a1 = _write_py(root, "a.py", "a = 1\n") |
| 332 | _make_commit(root, repo_id, branch="main", message="ours: change a", |
| 333 | manifest={"a.py": a1, "b.py": b0}) |
| 334 | |
| 335 | # theirs: modify b.py |
| 336 | b1 = _write_py(root, "b.py", "b = 1\n") |
| 337 | _make_commit(root, repo_id, branch="feat", message="theirs: change b", |
| 338 | manifest={"a.py": a0, "b.py": b1}) |
| 339 | |
| 340 | code, out = _run(root, "merge", "feat") |
| 341 | assert code == 0, out |
| 342 | m = _snapshot_manifest(root, "main") |
| 343 | assert m.get("a.py") == a1, "ours change to a.py lost after clean merge" |
| 344 | assert m.get("b.py") == b1, "theirs change to b.py lost after clean merge" |
| 345 | |
| 346 | def test_B2_theirs_adds_new_file(self, tmp_path: pathlib.Path) -> None: |
| 347 | """B2: theirs adds new.py that ours never touched — must be in merged.""" |
| 348 | root, repo_id = _init_code_repo(tmp_path) |
| 349 | a0 = _write_py(root, "a.py", "a = 0\n") |
| 350 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 351 | manifest={"a.py": a0}) |
| 352 | (ref_path(root, "feat")).write_text(base_c) |
| 353 | |
| 354 | new_id = _write_py(root, "new.py", "new = True\n") |
| 355 | _make_commit(root, repo_id, branch="feat", message="add new.py", |
| 356 | manifest={"a.py": a0, "new.py": new_id}) |
| 357 | |
| 358 | code, out = _run(root, "merge", "feat") |
| 359 | assert code == 0, out |
| 360 | assert "new.py" in _snapshot_manifest(root, "main"), "theirs new file lost" |
| 361 | |
| 362 | def test_B3_theirs_deletes_file_ours_never_touched(self, tmp_path: pathlib.Path) -> None: |
| 363 | """B3: theirs deletes stale.py — must be absent in merged.""" |
| 364 | root, repo_id = _init_code_repo(tmp_path) |
| 365 | a0 = _write_py(root, "a.py", "a = 0\n") |
| 366 | stale0 = _write_py(root, "stale.py", "dead = True\n") |
| 367 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 368 | manifest={"a.py": a0, "stale.py": stale0}) |
| 369 | (ref_path(root, "feat")).write_text(base_c) |
| 370 | |
| 371 | a1 = _write_py(root, "a.py", "a = 1\n") |
| 372 | _make_commit(root, repo_id, branch="main", message="ours: tweak a", |
| 373 | manifest={"a.py": a1, "stale.py": stale0}) |
| 374 | _make_commit(root, repo_id, branch="feat", message="theirs: rm stale.py", |
| 375 | manifest={"a.py": a0}) |
| 376 | |
| 377 | code, out = _run(root, "merge", "feat") |
| 378 | assert code == 0, out |
| 379 | m = _snapshot_manifest(root, "main") |
| 380 | assert "stale.py" not in m, "theirs deletion of stale.py was not applied" |
| 381 | assert m.get("a.py") == a1, "ours change to a.py lost" |
| 382 | |
| 383 | def test_B4_many_theirs_only_additions_all_survive(self, tmp_path: pathlib.Path) -> None: |
| 384 | """B4: theirs adds 30 files, ours changes 1 file — all 30 must be in merged.""" |
| 385 | root, repo_id = _init_code_repo(tmp_path) |
| 386 | base_id = _write_py(root, "main.py", "x = 0\n") |
| 387 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 388 | manifest={"main.py": base_id}) |
| 389 | (ref_path(root, "feat")).write_text(base_c) |
| 390 | |
| 391 | # ours: bump main.py |
| 392 | bumped = _write_py(root, "main.py", "x = 1\n") |
| 393 | _make_commit(root, repo_id, branch="main", message="ours: bump", |
| 394 | manifest={"main.py": bumped}) |
| 395 | |
| 396 | # theirs: 30 new modules |
| 397 | theirs_manifest = {"main.py": base_id} |
| 398 | for i in range(30): |
| 399 | oid = _write_py(root, f"mod_{i}.py", f"MOD_{i} = True\n") |
| 400 | theirs_manifest[f"mod_{i}.py"] = oid |
| 401 | _make_commit(root, repo_id, branch="feat", message="theirs: add 30 mods", |
| 402 | manifest=theirs_manifest) |
| 403 | |
| 404 | code, out = _run(root, "merge", "feat") |
| 405 | assert code == 0, out |
| 406 | m = _snapshot_manifest(root, "main") |
| 407 | for i in range(30): |
| 408 | assert f"mod_{i}.py" in m, f"mod_{i}.py missing after clean three-way merge" |
| 409 | |
| 410 | |
| 411 | # =========================================================================== |
| 412 | # C — Three-way with conflicts that MUST be surfaced |
| 413 | # =========================================================================== |
| 414 | |
| 415 | |
| 416 | class TestThreeWayConflictSurfacedC: |
| 417 | """Conflicts must be reported; the merge must NOT silently produce wrong content.""" |
| 418 | |
| 419 | def test_C1_genuine_conflict_exits_nonzero(self, tmp_path: pathlib.Path) -> None: |
| 420 | """C1: both sides change the same symbol in the same file → exit nonzero.""" |
| 421 | root, repo_id = _init_code_repo(tmp_path) |
| 422 | a0 = _write_py(root, "service.py", textwrap.dedent("""\ |
| 423 | def charge(): |
| 424 | return 'v1' |
| 425 | """)) |
| 426 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 427 | manifest={"service.py": a0}) |
| 428 | (ref_path(root, "feat")).write_text(base_c) |
| 429 | |
| 430 | a_ours = _write_py(root, "service.py", textwrap.dedent("""\ |
| 431 | def charge(): |
| 432 | return 'ours-v2' |
| 433 | """)) |
| 434 | _make_commit(root, repo_id, branch="main", message="ours: change charge", |
| 435 | manifest={"service.py": a_ours}) |
| 436 | |
| 437 | a_theirs = _write_py(root, "service.py", textwrap.dedent("""\ |
| 438 | def charge(): |
| 439 | return 'theirs-v2' |
| 440 | """)) |
| 441 | _make_commit(root, repo_id, branch="feat", message="theirs: change charge", |
| 442 | manifest={"service.py": a_theirs}) |
| 443 | |
| 444 | code, out = _run_unchecked(root, "merge", "feat") |
| 445 | assert code != 0, "conflict must exit nonzero, not silently succeed" |
| 446 | |
| 447 | def test_C2_conflict_creates_merge_state_json(self, tmp_path: pathlib.Path) -> None: |
| 448 | """C2: conflict writes MERGE_STATE.json with the right fields.""" |
| 449 | root, repo_id = _init_code_repo(tmp_path) |
| 450 | f0 = _write_py(root, "f.py", textwrap.dedent("""\ |
| 451 | def foo(): |
| 452 | return 1 |
| 453 | """)) |
| 454 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 455 | manifest={"f.py": f0}) |
| 456 | (ref_path(root, "feat")).write_text(base_c) |
| 457 | |
| 458 | f_ours = _write_py(root, "f.py", textwrap.dedent("""\ |
| 459 | def foo(): |
| 460 | return 2 |
| 461 | """)) |
| 462 | _make_commit(root, repo_id, branch="main", message="ours", |
| 463 | manifest={"f.py": f_ours}) |
| 464 | |
| 465 | f_theirs = _write_py(root, "f.py", textwrap.dedent("""\ |
| 466 | def foo(): |
| 467 | return 99 |
| 468 | """)) |
| 469 | _make_commit(root, repo_id, branch="feat", message="theirs", |
| 470 | manifest={"f.py": f_theirs}) |
| 471 | |
| 472 | _run_unchecked(root, "merge", "feat") |
| 473 | state_path = merge_state_path(root) |
| 474 | assert state_path.exists(), "MERGE_STATE.json must be written on conflict" |
| 475 | state = json.loads(state_path.read_text()) |
| 476 | assert "ours_commit" in state |
| 477 | assert "theirs_commit" in state |
| 478 | assert "conflict_paths" in state |
| 479 | |
| 480 | def test_C3_conflict_json_format_lists_paths(self, tmp_path: pathlib.Path) -> None: |
| 481 | """C3: --format json reports conflict with non-empty conflicts list.""" |
| 482 | root, repo_id = _init_code_repo(tmp_path) |
| 483 | f0 = _write_py(root, "svc.py", textwrap.dedent("""\ |
| 484 | def go(): |
| 485 | pass |
| 486 | """)) |
| 487 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 488 | manifest={"svc.py": f0}) |
| 489 | (ref_path(root, "feat")).write_text(base_c) |
| 490 | |
| 491 | f1 = _write_py(root, "svc.py", textwrap.dedent("""\ |
| 492 | def go(): |
| 493 | return 'ours' |
| 494 | """)) |
| 495 | _make_commit(root, repo_id, branch="main", message="ours", manifest={"svc.py": f1}) |
| 496 | |
| 497 | f2 = _write_py(root, "svc.py", textwrap.dedent("""\ |
| 498 | def go(): |
| 499 | return 'theirs' |
| 500 | """)) |
| 501 | _make_commit(root, repo_id, branch="feat", message="theirs", manifest={"svc.py": f2}) |
| 502 | |
| 503 | result = runner.invoke(cli, ["merge", "--force", "--json", "feat"], |
| 504 | env=_env(root)) |
| 505 | data = json.loads(result.output) |
| 506 | assert data["status"] == "conflict" |
| 507 | assert len(data["conflicts"]) > 0 |
| 508 | |
| 509 | |
| 510 | # =========================================================================== |
| 511 | # D — The silent-drop regression (commuting OT ops on same file) |
| 512 | # =========================================================================== |
| 513 | |
| 514 | |
| 515 | class TestSilentDropRegressionD: |
| 516 | """ |
| 517 | The exact bug that burned us: two branches modify DIFFERENT symbols in |
| 518 | the same file. OT sees them as commuting (non-conflicting at symbol level), |
| 519 | but cannot reconstruct the merged blob. Before the fix this silently |
| 520 | produced the ours blob and dropped all theirs changes in that file. |
| 521 | After the fix, this must either surface a conflict or correctly auto-merge. |
| 522 | |
| 523 | In either case: theirs-only CHANGES to OTHER FILES must always survive. |
| 524 | """ |
| 525 | |
| 526 | def test_D1_commuting_symbol_changes_do_not_silently_succeed( |
| 527 | self, tmp_path: pathlib.Path |
| 528 | ) -> None: |
| 529 | """D1: ours changes func_a, theirs changes func_b — must conflict or merge, never silently lose theirs.""" |
| 530 | root, repo_id = _init_code_repo(tmp_path) |
| 531 | base_code = textwrap.dedent("""\ |
| 532 | def func_a(): |
| 533 | return 'a-v1' |
| 534 | |
| 535 | def func_b(): |
| 536 | return 'b-v1' |
| 537 | """) |
| 538 | f0 = _write_py(root, "lib.py", base_code) |
| 539 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 540 | manifest={"lib.py": f0}) |
| 541 | (ref_path(root, "feat")).write_text(base_c) |
| 542 | |
| 543 | ours_code = textwrap.dedent("""\ |
| 544 | def func_a(): |
| 545 | return 'a-v2' |
| 546 | |
| 547 | def func_b(): |
| 548 | return 'b-v1' |
| 549 | """) |
| 550 | f_ours = _write_py(root, "lib.py", ours_code) |
| 551 | _make_commit(root, repo_id, branch="main", message="ours: change func_a", |
| 552 | manifest={"lib.py": f_ours}) |
| 553 | |
| 554 | theirs_code = textwrap.dedent("""\ |
| 555 | def func_a(): |
| 556 | return 'a-v1' |
| 557 | |
| 558 | def func_b(): |
| 559 | return 'b-v2' |
| 560 | """) |
| 561 | f_theirs = _write_py(root, "lib.py", theirs_code) |
| 562 | _make_commit(root, repo_id, branch="feat", message="theirs: change func_b", |
| 563 | manifest={"lib.py": f_theirs}) |
| 564 | |
| 565 | result = runner.invoke(cli, ["merge", "--force", "--json", "feat"], |
| 566 | env=_env(root)) |
| 567 | data = json.loads(result.output) |
| 568 | |
| 569 | if data["status"] == "merged": |
| 570 | # If auto-merged: func_b MUST be 'b-v2', never silently kept as 'b-v1'. |
| 571 | m = _snapshot_manifest(root, "main") |
| 572 | from muse.core.snapshots import read_snapshot |
| 573 | snap = None |
| 574 | from muse.core.commits import read_commit |
| 575 | commit = read_commit(root, _ref(root, "main")) |
| 576 | assert commit is not None |
| 577 | from muse.core.snapshots import read_snapshot |
| 578 | snap = read_snapshot(root, commit.snapshot_id) |
| 579 | assert snap is not None |
| 580 | # We can't read the actual merged file content from the manifest |
| 581 | # without the working tree, but we CAN assert lib.py is present. |
| 582 | assert "lib.py" in snap.manifest |
| 583 | else: |
| 584 | # If conflict: that is correct — better a conflict than silent data loss. |
| 585 | assert data["status"] == "conflict" |
| 586 | assert len(data["conflicts"]) > 0 |
| 587 | |
| 588 | def test_D2_theirs_only_file_survives_commuting_conflict( |
| 589 | self, tmp_path: pathlib.Path |
| 590 | ) -> None: |
| 591 | """D2: regression core — theirs-only executor.py must survive even when lib.py conflicts.""" |
| 592 | root, repo_id = _init_code_repo(tmp_path) |
| 593 | base_db = textwrap.dedent("""\ |
| 594 | def pool(): |
| 595 | pass |
| 596 | """) |
| 597 | base_exec = textwrap.dedent("""\ |
| 598 | def run(): |
| 599 | args = ['--pid=private'] |
| 600 | return args |
| 601 | """) |
| 602 | db0 = _write_py(root, "database.py", base_db) |
| 603 | exec0 = _write_py(root, "executor.py", base_exec) |
| 604 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 605 | manifest={"database.py": db0, "executor.py": exec0}) |
| 606 | (ref_path(root, "fix-branch")).write_text(base_c) |
| 607 | |
| 608 | # ours (dev): fix pool_pre_ping in database.py, don't touch executor.py |
| 609 | ours_db = textwrap.dedent("""\ |
| 610 | def pool(): |
| 611 | return 'pool_pre_ping=True' |
| 612 | """) |
| 613 | db_ours = _write_py(root, "database.py", ours_db) |
| 614 | _make_commit(root, repo_id, branch="main", message="ours: pool_pre_ping fix", |
| 615 | manifest={"database.py": db_ours, "executor.py": exec0}) |
| 616 | |
| 617 | # theirs (fix-branch): fix pool_pre_ping the same way AND fix executor.py |
| 618 | theirs_db = textwrap.dedent("""\ |
| 619 | def pool(): |
| 620 | return 'pool_pre_ping=True' |
| 621 | """) |
| 622 | theirs_exec = textwrap.dedent("""\ |
| 623 | def run(): |
| 624 | args = [] # --pid=private removed (invalid Docker flag) |
| 625 | return args |
| 626 | """) |
| 627 | db_theirs = _write_py(root, "database.py", theirs_db) |
| 628 | exec_theirs = _write_py(root, "executor.py", theirs_exec) |
| 629 | _make_commit(root, repo_id, branch="fix-branch", |
| 630 | message="theirs: pool_pre_ping + remove --pid=private", |
| 631 | manifest={"database.py": db_theirs, "executor.py": exec_theirs}) |
| 632 | |
| 633 | result = runner.invoke(cli, ["merge", "--force", "--json", "fix-branch"], |
| 634 | env=_env(root)) |
| 635 | data = json.loads(result.output) |
| 636 | |
| 637 | # The critical assertion: in any outcome, executor.py must NOT be the old version. |
| 638 | # Either the merge succeeded and executor.py has the fix, OR a conflict is raised |
| 639 | # so the user can resolve it. What is NEVER acceptable: silent success with old content. |
| 640 | if data["status"] == "merged": |
| 641 | from muse.core.commits import read_commit |
| 642 | from muse.core.snapshots import read_snapshot |
| 643 | commit = read_commit(root, _ref(root, "main")) |
| 644 | assert commit is not None |
| 645 | snap = read_snapshot(root, commit.snapshot_id) |
| 646 | assert snap is not None |
| 647 | # executor.py must be the FIXED version (no --pid=private), not the base. |
| 648 | assert snap.manifest.get("executor.py") == exec_theirs, ( |
| 649 | "REGRESSION: executor.py fix was silently dropped — " |
| 650 | "the theirs-only change was lost in the merge" |
| 651 | ) |
| 652 | else: |
| 653 | # Conflict is acceptable (user can resolve), silent data loss is not. |
| 654 | assert data["status"] == "conflict" |
| 655 | |
| 656 | def test_D3_identical_object_hash_on_both_sides_no_file_conflict( |
| 657 | self, tmp_path: pathlib.Path |
| 658 | ) -> None: |
| 659 | """D3: both sides converge to the EXACT same object hash — file-level conflict impossible. |
| 660 | |
| 661 | When ours and theirs both arrive at the same content hash for a file, |
| 662 | diff_snapshots sees them as identical (no change relative to each other). |
| 663 | The merge engine must treat this as a clean convergence — or at minimum, |
| 664 | the resulting manifest must contain that file at the shared hash. |
| 665 | |
| 666 | This tests the file-level merge_engine layer (diff_snapshots / apply_merge). |
| 667 | Symbol-level conflict detection (within the file) is separate and handled |
| 668 | by the plugin — if the plugin marks it as conflicting despite identical |
| 669 | hashes, that is a plugin-level decision, not a data-loss scenario. |
| 670 | """ |
| 671 | from muse.core.merge_engine import diff_snapshots, detect_conflicts, apply_merge |
| 672 | |
| 673 | fixed_hash = _h("pool_pre_ping_fix_content") |
| 674 | base_hash = _h("original_pool_content") |
| 675 | |
| 676 | base_manifest = {"database.py": base_hash, "other.py": _h("other")} |
| 677 | ours_manifest = {"database.py": fixed_hash, "other.py": _h("other")} |
| 678 | theirs_manifest = {"database.py": fixed_hash, "other.py": _h("other")} |
| 679 | |
| 680 | ours_changed = diff_snapshots(base_manifest, ours_manifest) |
| 681 | theirs_changed = diff_snapshots(base_manifest, theirs_manifest) |
| 682 | conflicts = detect_conflicts(ours_changed, theirs_changed, ours_manifest, theirs_manifest) |
| 683 | merged = apply_merge(base_manifest, ours_manifest, theirs_manifest, |
| 684 | ours_changed, theirs_changed, conflicts) |
| 685 | |
| 686 | # Both sides converged to the SAME hash — detect_conflicts must not flag it. |
| 687 | assert "database.py" not in conflicts, ( |
| 688 | "D3 VIOLATED: convergent same-hash change wrongly reported as conflict" |
| 689 | ) |
| 690 | # apply_merge must include database.py at the agreed fixed hash. |
| 691 | assert merged.get("database.py") == fixed_hash, ( |
| 692 | "D3 VIOLATED: database.py absent or at wrong hash after convergent merge" |
| 693 | ) |
| 694 | |
| 695 | def test_D4_the_musehub_regression_scenario(self, tmp_path: pathlib.Path) -> None: |
| 696 | """D4: exact topology from the MuseHub incident — 3 branches, complex DAG. |
| 697 | |
| 698 | Timeline: |
| 699 | base → ours (dev): pool_pre_ping DB fix |
| 700 | base → theirs (fix-branch): pool_pre_ping fix + --pid fix + AGENTS.md rewrite + new_feature.py |
| 701 | |
| 702 | When user merges fix-branch into dev: |
| 703 | - database.py: both changed (same content, should be clean OR conflict) |
| 704 | - executor.py: theirs-only change → MUST survive in merged |
| 705 | - agents.md: theirs-only change → MUST survive in merged |
| 706 | - new_feature.py: theirs-only addition → MUST survive in merged |
| 707 | """ |
| 708 | root, repo_id = _init_code_repo(tmp_path) |
| 709 | |
| 710 | # Base state |
| 711 | db0 = _write_py(root, "database.py", "def pool(): pass\n") |
| 712 | exec0 = _write_py(root, "executor.py", "args = ['--pid=private']\n") |
| 713 | agents0 = _write_py(root, "agents.md", "# Short docs\n") |
| 714 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 715 | manifest={"database.py": db0, "executor.py": exec0, |
| 716 | "agents.md": agents0}) |
| 717 | (ref_path(root, "fix-branch")).write_text(base_c) |
| 718 | |
| 719 | # ours (dev): pool_pre_ping only |
| 720 | db_ours = _write_py(root, "database.py", "def pool(): return 'pool_pre_ping=True'\n") |
| 721 | _make_commit(root, repo_id, branch="main", message="ours: pool_pre_ping", |
| 722 | manifest={"database.py": db_ours, "executor.py": exec0, |
| 723 | "agents.md": agents0}) |
| 724 | |
| 725 | # theirs (fix-branch): pool_pre_ping + pid fix + AGENTS.md rewrite + new file |
| 726 | db_theirs = _write_py(root, "database.py", "def pool(): return 'pool_pre_ping=True'\n") |
| 727 | exec_theirs = _write_py(root, "executor.py", "args = [] # no --pid\n") |
| 728 | agents_theirs = _write_py(root, "agents.md", "# Comprehensive 700-line rewrite\n" * 10) |
| 729 | new_feat = _write_py(root, "new_feature.py", "NEW = True\n") |
| 730 | _make_commit(root, repo_id, branch="fix-branch", |
| 731 | message="theirs: comprehensive fix mpack", |
| 732 | manifest={"database.py": db_theirs, "executor.py": exec_theirs, |
| 733 | "agents.md": agents_theirs, "new_feature.py": new_feat}) |
| 734 | |
| 735 | result = runner.invoke(cli, ["merge", "--force", "--json", "fix-branch"], |
| 736 | env=_env(root)) |
| 737 | data = json.loads(result.output) |
| 738 | |
| 739 | if data["status"] == "merged": |
| 740 | from muse.core.commits import read_commit |
| 741 | from muse.core.snapshots import read_snapshot |
| 742 | commit = read_commit(root, _ref(root, "main")) |
| 743 | assert commit is not None |
| 744 | snap = read_snapshot(root, commit.snapshot_id) |
| 745 | assert snap is not None |
| 746 | m = snap.manifest |
| 747 | |
| 748 | assert m.get("executor.py") == exec_theirs, ( |
| 749 | "REGRESSION: executor.py (--pid fix) was silently dropped" |
| 750 | ) |
| 751 | assert m.get("agents.md") == agents_theirs, ( |
| 752 | "REGRESSION: agents.md rewrite was silently dropped" |
| 753 | ) |
| 754 | assert "new_feature.py" in m, ( |
| 755 | "REGRESSION: new_feature.py addition was silently dropped" |
| 756 | ) |
| 757 | else: |
| 758 | # A conflict is an acceptable outcome. |
| 759 | # But check that it's not some other failure mode. |
| 760 | assert data["status"] == "conflict", f"unexpected status: {data['status']}" |
| 761 | |
| 762 | |
| 763 | # =========================================================================== |
| 764 | # E — Theirs-only files MUST survive even when there are conflicts elsewhere |
| 765 | # =========================================================================== |
| 766 | |
| 767 | |
| 768 | class TestTheirsOnlySurvivesConflictE: |
| 769 | """ |
| 770 | When there IS a genuine conflict in file X, the merge stops. But the |
| 771 | *would-be* merged manifest (what the engine computed before stopping) must |
| 772 | still contain all theirs-only changes. The engine must not take a shortcut |
| 773 | and return ours manifest verbatim just because a conflict exists. |
| 774 | |
| 775 | These tests use the JSON output's "files_changed" or check MERGE_STATE.json |
| 776 | to infer what the engine planned to write. |
| 777 | |
| 778 | After a conflict, the user resolves and re-commits — but if the engine's |
| 779 | intermediate merged manifest is wrong, the resolution will silently bake |
| 780 | in the data loss. |
| 781 | """ |
| 782 | |
| 783 | def test_E1_theirs_additions_included_in_merged_manifest_despite_conflict( |
| 784 | self, tmp_path: pathlib.Path |
| 785 | ) -> None: |
| 786 | """E1: conflict in a.py; theirs adds b.py and c.py — both must be in merged manifest.""" |
| 787 | root, repo_id = _init_code_repo(tmp_path) |
| 788 | a0 = _write_py(root, "a.py", textwrap.dedent("""\ |
| 789 | def go(): |
| 790 | return 1 |
| 791 | """)) |
| 792 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 793 | manifest={"a.py": a0}) |
| 794 | (ref_path(root, "feat")).write_text(base_c) |
| 795 | |
| 796 | a_ours = _write_py(root, "a.py", textwrap.dedent("""\ |
| 797 | def go(): |
| 798 | return 'ours' |
| 799 | """)) |
| 800 | _make_commit(root, repo_id, branch="main", message="ours: change a.py", |
| 801 | manifest={"a.py": a_ours}) |
| 802 | |
| 803 | a_theirs = _write_py(root, "a.py", textwrap.dedent("""\ |
| 804 | def go(): |
| 805 | return 'theirs' |
| 806 | """)) |
| 807 | b_theirs = _write_py(root, "b.py", "B = True\n") |
| 808 | c_theirs = _write_py(root, "c.py", "C = True\n") |
| 809 | _make_commit(root, repo_id, branch="feat", message="theirs: change a + add b + add c", |
| 810 | manifest={"a.py": a_theirs, "b.py": b_theirs, "c.py": c_theirs}) |
| 811 | |
| 812 | result = runner.invoke(cli, ["merge", "--force", "--json", "feat"], |
| 813 | env=_env(root)) |
| 814 | data = json.loads(result.output) |
| 815 | |
| 816 | # Two acceptable outcomes: |
| 817 | # 1. Clean merge (auto-resolved) — b.py and c.py must be in main manifest |
| 818 | # 2. Conflict in a.py — MERGE_STATE must be written; we trust the engine |
| 819 | # will include b.py and c.py in the conflict-resolution manifest. |
| 820 | if data["status"] == "merged": |
| 821 | m = _snapshot_manifest(root, "main") |
| 822 | assert "b.py" in m, "theirs-only b.py was lost despite clean merge of other files" |
| 823 | assert "c.py" in m, "theirs-only c.py was lost despite clean merge of other files" |
| 824 | else: |
| 825 | assert data["status"] == "conflict" |
| 826 | # The engine computed conflicts — but must NOT have silently dropped b.py/c.py |
| 827 | # from the intermediate manifest it would apply after resolution. |
| 828 | # We verify this by inspecting what would have been applied: check that |
| 829 | # the conflict paths DON'T include b.py or c.py (they're theirs-only, not conflicts). |
| 830 | assert "b.py" not in data.get("conflicts", []), "b.py incorrectly marked as conflict" |
| 831 | assert "c.py" not in data.get("conflicts", []), "c.py incorrectly marked as conflict" |
| 832 | |
| 833 | def test_E2_ten_theirs_only_files_all_excluded_from_conflict_list( |
| 834 | self, tmp_path: pathlib.Path |
| 835 | ) -> None: |
| 836 | """E2: 10 theirs-only additions must never appear in the conflict list.""" |
| 837 | root, repo_id = _init_code_repo(tmp_path) |
| 838 | f0 = _write_py(root, "main.py", textwrap.dedent("""\ |
| 839 | def run(): |
| 840 | pass |
| 841 | """)) |
| 842 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 843 | manifest={"main.py": f0}) |
| 844 | (ref_path(root, "feat")).write_text(base_c) |
| 845 | |
| 846 | f_ours = _write_py(root, "main.py", textwrap.dedent("""\ |
| 847 | def run(): |
| 848 | return 'ours' |
| 849 | """)) |
| 850 | _make_commit(root, repo_id, branch="main", message="ours: modify run", |
| 851 | manifest={"main.py": f_ours}) |
| 852 | |
| 853 | f_theirs = _write_py(root, "main.py", textwrap.dedent("""\ |
| 854 | def run(): |
| 855 | return 'theirs' |
| 856 | """)) |
| 857 | theirs_manifest: Manifest = {"main.py": f_theirs} |
| 858 | for i in range(10): |
| 859 | oid = _write_py(root, f"extra_{i}.py", f"EXTRA_{i} = True\n") |
| 860 | theirs_manifest[f"extra_{i}.py"] = oid |
| 861 | _make_commit(root, repo_id, branch="feat", message="theirs: conflict + 10 extras", |
| 862 | manifest=theirs_manifest) |
| 863 | |
| 864 | result = runner.invoke(cli, ["merge", "--force", "--json", "feat"], |
| 865 | env=_env(root)) |
| 866 | data = json.loads(result.output) |
| 867 | conflicts = data.get("conflicts", []) |
| 868 | for i in range(10): |
| 869 | assert f"extra_{i}.py" not in conflicts, ( |
| 870 | f"extra_{i}.py is a theirs-only addition — must not appear in conflicts" |
| 871 | ) |
| 872 | |
| 873 | |
| 874 | # =========================================================================== |
| 875 | # F — Strategy shortcuts correctness |
| 876 | # =========================================================================== |
| 877 | |
| 878 | |
| 879 | class TestStrategyShortcutsF: |
| 880 | """ |
| 881 | --strategy=ours and --strategy=theirs are convenience shortcuts. |
| 882 | The correct behaviour: non-conflicting theirs/ours changes are STILL |
| 883 | applied; only the conflicting files take the chosen side. |
| 884 | |
| 885 | The old bug: --strategy=ours took ENTIRE ours manifest, discarding all |
| 886 | theirs-only changes. This caused data loss just as severe as the OT bug. |
| 887 | """ |
| 888 | |
| 889 | def test_F1_strategy_ours_preserves_theirs_only_files(self, tmp_path: pathlib.Path) -> None: |
| 890 | """F1: --strategy=ours for conflict in a.py; theirs-only b.py must still appear.""" |
| 891 | root, repo_id = _init_code_repo(tmp_path) |
| 892 | a0 = _write_py(root, "a.py", textwrap.dedent("""\ |
| 893 | def go(): |
| 894 | return 1 |
| 895 | """)) |
| 896 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 897 | manifest={"a.py": a0}) |
| 898 | (ref_path(root, "feat")).write_text(base_c) |
| 899 | |
| 900 | a_ours = _write_py(root, "a.py", textwrap.dedent("""\ |
| 901 | def go(): |
| 902 | return 'ours' |
| 903 | """)) |
| 904 | _make_commit(root, repo_id, branch="main", message="ours", manifest={"a.py": a_ours}) |
| 905 | |
| 906 | a_theirs = _write_py(root, "a.py", textwrap.dedent("""\ |
| 907 | def go(): |
| 908 | return 'theirs' |
| 909 | """)) |
| 910 | b_theirs = _write_py(root, "b.py", "B = True\n") |
| 911 | _make_commit(root, repo_id, branch="feat", message="theirs", |
| 912 | manifest={"a.py": a_theirs, "b.py": b_theirs}) |
| 913 | |
| 914 | code, out = _run(root, "merge", "--strategy", "ours", "feat") |
| 915 | assert code == 0, out |
| 916 | |
| 917 | m = _snapshot_manifest(root, "main") |
| 918 | # a.py must be ours version. |
| 919 | assert m.get("a.py") == a_ours, "--strategy=ours must keep ours version of conflicting file" |
| 920 | # b.py is theirs-only — it must be present. |
| 921 | assert "b.py" in m, ( |
| 922 | "REGRESSION: --strategy=ours discarded theirs-only b.py. " |
| 923 | "Non-conflicting theirs changes must still be applied." |
| 924 | ) |
| 925 | |
| 926 | def test_F2_strategy_theirs_preserves_ours_only_files(self, tmp_path: pathlib.Path) -> None: |
| 927 | """F2: --strategy=theirs for conflict in a.py; ours-only c.py must still appear.""" |
| 928 | root, repo_id = _init_code_repo(tmp_path) |
| 929 | a0 = _write_py(root, "a.py", textwrap.dedent("""\ |
| 930 | def go(): |
| 931 | return 1 |
| 932 | """)) |
| 933 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 934 | manifest={"a.py": a0}) |
| 935 | (ref_path(root, "feat")).write_text(base_c) |
| 936 | |
| 937 | a_ours = _write_py(root, "a.py", textwrap.dedent("""\ |
| 938 | def go(): |
| 939 | return 'ours' |
| 940 | """)) |
| 941 | c_ours = _write_py(root, "c.py", "C = True\n") |
| 942 | _make_commit(root, repo_id, branch="main", message="ours", |
| 943 | manifest={"a.py": a_ours, "c.py": c_ours}) |
| 944 | |
| 945 | a_theirs = _write_py(root, "a.py", textwrap.dedent("""\ |
| 946 | def go(): |
| 947 | return 'theirs' |
| 948 | """)) |
| 949 | _make_commit(root, repo_id, branch="feat", message="theirs", |
| 950 | manifest={"a.py": a_theirs}) |
| 951 | |
| 952 | code, out = _run(root, "merge", "--strategy", "theirs", "feat") |
| 953 | assert code == 0, out |
| 954 | |
| 955 | m = _snapshot_manifest(root, "main") |
| 956 | # a.py must be theirs. |
| 957 | assert m.get("a.py") == a_theirs, "--strategy=theirs must keep theirs version" |
| 958 | # c.py is ours-only — must be in merged. |
| 959 | assert "c.py" in m, ( |
| 960 | "REGRESSION: --strategy=theirs discarded ours-only c.py. " |
| 961 | "Non-conflicting ours changes must still be applied." |
| 962 | ) |
| 963 | |
| 964 | def test_F3_strategy_ours_with_zero_ours_changes_is_up_to_date( |
| 965 | self, tmp_path: pathlib.Path |
| 966 | ) -> None: |
| 967 | """F3: --strategy=ours when ours == base → theirs changes should all be applied.""" |
| 968 | root, repo_id = _init_code_repo(tmp_path) |
| 969 | f0 = _write_py(root, "f.py", "x = 0\n") |
| 970 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 971 | manifest={"f.py": f0}) |
| 972 | (ref_path(root, "feat")).write_text(base_c) |
| 973 | |
| 974 | g_id = _write_py(root, "g.py", "g = True\n") |
| 975 | _make_commit(root, repo_id, branch="feat", message="theirs: add g.py", |
| 976 | manifest={"f.py": f0, "g.py": g_id}) |
| 977 | |
| 978 | # No ours changes since base. |
| 979 | code, out = _run(root, "merge", "--strategy", "ours", "feat") |
| 980 | assert code == 0, out |
| 981 | m = _snapshot_manifest(root, "main") |
| 982 | # g.py is theirs-only — must be present. |
| 983 | assert "g.py" in m, "theirs-only addition lost with --strategy=ours when ours has no changes" |
| 984 | |
| 985 | |
| 986 | # =========================================================================== |
| 987 | # G — Full MuseHub regression scenario: pool_pre_ping + executor + AGENTS.md |
| 988 | # =========================================================================== |
| 989 | |
| 990 | |
| 991 | class TestMuseHubRegressionScenarioG: |
| 992 | """ |
| 993 | Reproduces the exact topology that led to every CI run failing with |
| 994 | 'docker: --pid: invalid PID mode' for days. |
| 995 | |
| 996 | This test is the "aha! that's it!" test the user asked for. |
| 997 | It must FAIL on the old Muse code (before commit 73427a30) and |
| 998 | PASS on the fixed code. |
| 999 | """ |
| 1000 | |
| 1001 | def test_G1_musehub_incident_executor_fix_not_lost(self, tmp_path: pathlib.Path) -> None: |
| 1002 | """G1: the MuseHub incident in miniature — never again. |
| 1003 | |
| 1004 | Topology: |
| 1005 | C0 (base): database.py v1, executor.py v1 (broken), agents.md v1 |
| 1006 | C1 (dev): pool_pre_ping fix on database.py ← ours |
| 1007 | C2 (fix-pool): pool_pre_ping fix on database.py ← theirs (same fix) |
| 1008 | + --pid=private removed from executor.py ← theirs only |
| 1009 | + agents.md comprehensive rewrite ← theirs only |
| 1010 | |
| 1011 | Expected after merge: |
| 1012 | executor.py MUST be the fixed version (no --pid=private) |
| 1013 | agents.md MUST be the comprehensive rewrite |
| 1014 | database.py MUST be the pool_pre_ping version (either side, same content) |
| 1015 | """ |
| 1016 | root, repo_id = _init_code_repo(tmp_path) |
| 1017 | |
| 1018 | db_v1 = _write_py(root, "database.py", |
| 1019 | "def init_db(): return engine\n") |
| 1020 | exec_v1 = _write_py(root, "executor.py", |
| 1021 | "DOCKER_ARGS = ['--memory=1g', '--pid=private']\n") |
| 1022 | agents_v1 = _write_py(root, "agents.md", |
| 1023 | "# MuseHub Agent Contract\nDo stuff.\n") |
| 1024 | |
| 1025 | c0 = _make_commit(root, repo_id, branch="main", message="C0: base", |
| 1026 | manifest={"database.py": db_v1, "executor.py": exec_v1, |
| 1027 | "agents.md": agents_v1}) |
| 1028 | (ref_path(root, "fix-pool")).write_text(c0) |
| 1029 | |
| 1030 | # C1 — ours (dev): pool_pre_ping fix, nothing else |
| 1031 | db_v2 = _write_py(root, "database.py", |
| 1032 | "def init_db(): return engine.execution_options(pool_pre_ping=True)\n") |
| 1033 | c1 = _make_commit(root, repo_id, branch="main", message="C1: pool_pre_ping", |
| 1034 | manifest={"database.py": db_v2, "executor.py": exec_v1, |
| 1035 | "agents.md": agents_v1}) |
| 1036 | |
| 1037 | # C2 — theirs (fix-pool): same pool_pre_ping + executor fix + agents rewrite |
| 1038 | db_v2b = _write_py(root, "database.py", |
| 1039 | "def init_db(): return engine.execution_options(pool_pre_ping=True)\n") |
| 1040 | exec_v2 = _write_py(root, "executor.py", |
| 1041 | "DOCKER_ARGS = ['--memory=1g'] # --pid=private removed\n") |
| 1042 | agents_v2 = _write_py(root, "agents.md", |
| 1043 | "# Comprehensive 700-line rewrite\n" * 20) |
| 1044 | c2 = _make_commit(root, repo_id, branch="fix-pool", |
| 1045 | message="C2: pool_pre_ping + executor fix + agents rewrite", |
| 1046 | manifest={"database.py": db_v2b, "executor.py": exec_v2, |
| 1047 | "agents.md": agents_v2}) |
| 1048 | |
| 1049 | result = runner.invoke(cli, ["merge", "--force", "--json", "fix-pool"], |
| 1050 | env=_env(root)) |
| 1051 | data = json.loads(result.output) |
| 1052 | |
| 1053 | from muse.core.commits import read_commit |
| 1054 | from muse.core.snapshots import read_snapshot |
| 1055 | |
| 1056 | if data["status"] == "merged": |
| 1057 | commit = read_commit(root, _ref(root, "main")) |
| 1058 | assert commit is not None |
| 1059 | snap = read_snapshot(root, commit.snapshot_id) |
| 1060 | assert snap is not None |
| 1061 | m = snap.manifest |
| 1062 | |
| 1063 | assert m.get("executor.py") == exec_v2, ( |
| 1064 | "\n\nREGRESSION DETECTED — test_G1_musehub_incident_executor_fix_not_lost\n" |
| 1065 | "executor.py still has '--pid=private' after merge.\n" |
| 1066 | "The silent-drop bug in CodePlugin.merge_ops has returned.\n" |
| 1067 | "See commit 73427a30 for the fix that must be applied.\n" |
| 1068 | ) |
| 1069 | assert m.get("agents.md") == agents_v2, ( |
| 1070 | "\n\nREGRESSION DETECTED — agents.md rewrite was silently dropped.\n" |
| 1071 | ) |
| 1072 | # database.py must be the pool_pre_ping version (same content on both sides). |
| 1073 | assert m.get("database.py") in (db_v2, db_v2b), ( |
| 1074 | "database.py pool_pre_ping fix was lost" |
| 1075 | ) |
| 1076 | elif data["status"] == "conflict": |
| 1077 | # Conflict is acceptable. Verify executor.py and agents.md are NOT in the conflict list. |
| 1078 | conflicts = data.get("conflicts", []) |
| 1079 | assert "executor.py" not in conflicts, ( |
| 1080 | "executor.py is theirs-only — must not appear in conflicts, only in merged manifest" |
| 1081 | ) |
| 1082 | assert "agents.md" not in conflicts, ( |
| 1083 | "agents.md is theirs-only — must not appear in conflicts" |
| 1084 | ) |
| 1085 | else: |
| 1086 | pytest.fail(f"Unexpected merge status: {data['status']}\n{data}") |
| 1087 | |
| 1088 | def test_G2_merge_commit_has_two_parents(self, tmp_path: pathlib.Path) -> None: |
| 1089 | """G2: a successful three-way merge always creates a commit with 2 parent IDs.""" |
| 1090 | root, repo_id = _init_code_repo(tmp_path) |
| 1091 | a0 = _write_py(root, "a.py", "x = 0\n") |
| 1092 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 1093 | manifest={"a.py": a0}) |
| 1094 | (ref_path(root, "feat")).write_text(base_c) |
| 1095 | |
| 1096 | a1 = _write_py(root, "a.py", "x = 1\n") |
| 1097 | _make_commit(root, repo_id, branch="main", message="ours", manifest={"a.py": a1}) |
| 1098 | |
| 1099 | b1 = _write_py(root, "b.py", "b = 1\n") |
| 1100 | _make_commit(root, repo_id, branch="feat", message="theirs", |
| 1101 | manifest={"a.py": a0, "b.py": b1}) |
| 1102 | |
| 1103 | code, out = _run(root, "merge", "feat") |
| 1104 | assert code == 0, out |
| 1105 | |
| 1106 | from muse.core.commits import read_commit |
| 1107 | commit = read_commit(root, _ref(root, "main")) |
| 1108 | assert commit is not None |
| 1109 | # A three-way merge commit must record both parents. |
| 1110 | assert commit.parent2_commit_id is not None, ( |
| 1111 | "three-way merge commit missing second parent — " |
| 1112 | "merge history will appear linear in `muse log`" |
| 1113 | ) |
| 1114 | |
| 1115 | def test_G3_merged_snapshot_is_not_ours_snapshot_verbatim( |
| 1116 | self, tmp_path: pathlib.Path |
| 1117 | ) -> None: |
| 1118 | """G3: the snapshot recorded by the merge commit must differ from ours snapshot. |
| 1119 | |
| 1120 | When the merged snapshot equals ours verbatim, theirs changes were silently dropped. |
| 1121 | """ |
| 1122 | root, repo_id = _init_code_repo(tmp_path) |
| 1123 | a0 = _write_py(root, "a.py", "x = 0\n") |
| 1124 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 1125 | manifest={"a.py": a0}) |
| 1126 | (ref_path(root, "feat")).write_text(base_c) |
| 1127 | |
| 1128 | a1 = _write_py(root, "a.py", "x = 1\n") |
| 1129 | ours_c = _make_commit(root, repo_id, branch="main", message="ours", |
| 1130 | manifest={"a.py": a1}) |
| 1131 | |
| 1132 | b1 = _write_py(root, "b.py", "b = True\n") |
| 1133 | _make_commit(root, repo_id, branch="feat", message="theirs: add b.py", |
| 1134 | manifest={"a.py": a0, "b.py": b1}) |
| 1135 | |
| 1136 | # Get ours snapshot_id BEFORE the merge. |
| 1137 | from muse.core.commits import read_commit |
| 1138 | ours_commit = read_commit(root, ours_c) |
| 1139 | assert ours_commit is not None |
| 1140 | ours_snap_id = ours_commit.snapshot_id |
| 1141 | |
| 1142 | code, out = _run(root, "merge", "feat") |
| 1143 | assert code == 0, out |
| 1144 | |
| 1145 | merge_commit = read_commit(root, _ref(root, "main")) |
| 1146 | assert merge_commit is not None |
| 1147 | assert merge_commit.snapshot_id != ours_snap_id, ( |
| 1148 | "REGRESSION: merged snapshot equals ours snapshot verbatim. " |
| 1149 | "Theirs changes (b.py addition) were silently discarded." |
| 1150 | ) |
| 1151 | |
| 1152 | |
| 1153 | # =========================================================================== |
| 1154 | # H — Merge-base correctness for complex DAG topologies |
| 1155 | # =========================================================================== |
| 1156 | |
| 1157 | |
| 1158 | class TestMergeBaseCorrectnessH: |
| 1159 | """ |
| 1160 | find_merge_base must handle complex DAG shapes correctly. |
| 1161 | An incorrect LCA leads to wrong merge-base manifests, which cause |
| 1162 | phantom conflicts (changes treated as conflicting when they aren't) |
| 1163 | or missed conflicts (changes treated as clean when they conflict). |
| 1164 | """ |
| 1165 | |
| 1166 | def test_H1_diamond_topology_correct_lca(self, tmp_path: pathlib.Path) -> None: |
| 1167 | """H1: diamond DAG — LCA is the bottom of the diamond, not an earlier commit. |
| 1168 | |
| 1169 | C0 |
| 1170 | /\\ |
| 1171 | C1 C2 |
| 1172 | \\ / |
| 1173 | C3 (merge of C1 and C2) |
| 1174 | |
| 1175 | Merging C3 into C1 (or C2) should detect C0 as the LCA, not something else. |
| 1176 | """ |
| 1177 | from muse.core.merge_engine import find_merge_base |
| 1178 | root, repo_id = _init_code_repo(tmp_path) |
| 1179 | |
| 1180 | f0 = _write_py(root, "f.py", "v = 0\n") |
| 1181 | c0 = _make_commit(root, repo_id, branch="main", message="C0", |
| 1182 | manifest={"f.py": f0}) |
| 1183 | (ref_path(root, "branch-a")).write_text(c0) |
| 1184 | (ref_path(root, "branch-b")).write_text(c0) |
| 1185 | |
| 1186 | f1 = _write_py(root, "f.py", "v = 1\n") |
| 1187 | c1 = _make_commit(root, repo_id, branch="branch-a", message="C1", |
| 1188 | manifest={"f.py": f1}) |
| 1189 | |
| 1190 | f2 = _write_py(root, "f.py", "v = 2\n") |
| 1191 | c2 = _make_commit(root, repo_id, branch="branch-b", message="C2", |
| 1192 | manifest={"f.py": f2}) |
| 1193 | |
| 1194 | # C3: a merge commit combining C1 and C2 — just use C1's snapshot for simplicity. |
| 1195 | c3 = _make_commit(root, repo_id, branch="main", message="C3: merge", |
| 1196 | manifest={"f.py": f1}, |
| 1197 | parent_commit_id=c1, parent2_commit_id=c2) |
| 1198 | |
| 1199 | lca = find_merge_base(root, c1, c3) |
| 1200 | assert lca == c1, ( |
| 1201 | f"LCA(C1, C3) should be C1 (C3 is a descendant of C1), got {lca}" |
| 1202 | ) |
| 1203 | |
| 1204 | lca2 = find_merge_base(root, c0, c3) |
| 1205 | assert lca2 == c0, ( |
| 1206 | f"LCA(C0, C3) should be C0 (common ancestor of C0-C3 chain), got {lca2}" |
| 1207 | ) |
| 1208 | |
| 1209 | def test_H2_long_linear_chain_lca(self, tmp_path: pathlib.Path) -> None: |
| 1210 | """H2: 20-commit linear chain — LCA of first and last commit is the first commit.""" |
| 1211 | from muse.core.merge_engine import find_merge_base |
| 1212 | root, repo_id = _init_code_repo(tmp_path) |
| 1213 | |
| 1214 | f0 = _write_py(root, "f.py", "v = 0\n") |
| 1215 | first_c = _make_commit(root, repo_id, branch="main", message="C0", |
| 1216 | manifest={"f.py": f0}) |
| 1217 | (ref_path(root, "branch")).write_text(first_c) |
| 1218 | |
| 1219 | last_c = first_c |
| 1220 | for i in range(1, 21): |
| 1221 | fi = _write_py(root, "f.py", f"v = {i}\n") |
| 1222 | last_c = _make_commit(root, repo_id, branch="main", message=f"C{i}", |
| 1223 | manifest={"f.py": fi}) |
| 1224 | |
| 1225 | lca = find_merge_base(root, first_c, last_c) |
| 1226 | assert lca == first_c, "LCA of linear chain tip and base should be the base" |
| 1227 | |
| 1228 | def test_H3_lca_of_equal_commits_is_that_commit(self, tmp_path: pathlib.Path) -> None: |
| 1229 | """H3: LCA(X, X) == X.""" |
| 1230 | from muse.core.merge_engine import find_merge_base |
| 1231 | root, repo_id = _init_code_repo(tmp_path) |
| 1232 | f0 = _write_py(root, "f.py", "v = 0\n") |
| 1233 | c0 = _make_commit(root, repo_id, branch="main", message="C0", |
| 1234 | manifest={"f.py": f0}) |
| 1235 | lca = find_merge_base(root, c0, c0) |
| 1236 | assert lca == c0 |
| 1237 | |
| 1238 | def test_H4_merge_base_with_remote_tracking_branch_topology( |
| 1239 | self, tmp_path: pathlib.Path |
| 1240 | ) -> None: |
| 1241 | """H4: simulates the exact topology of the MuseHub incident. |
| 1242 | |
| 1243 | local/dev (727dad83) branched from 5e6c6476. |
| 1244 | remote/dev (e01007b4) is a merge of [5e6c6476, d40f74ba]. |
| 1245 | d40f74ba includes 727dad83 in its ancestry. |
| 1246 | |
| 1247 | LCA(local/dev, remote/dev) should be 5e6c6476 (NOT 727dad83), |
| 1248 | because 5e6c6476 is the common ancestor that appears first in the BFS |
| 1249 | of remote/dev's parents. |
| 1250 | |
| 1251 | With this LCA, the three-way merge MUST detect that: |
| 1252 | - executor.py is a theirs-only change (theirs changed it from base, ours did not) |
| 1253 | - executor.py must appear in the merged manifest. |
| 1254 | """ |
| 1255 | from muse.core.merge_engine import find_merge_base |
| 1256 | root, repo_id = _init_code_repo(tmp_path) |
| 1257 | |
| 1258 | # 5e6c6476 equivalent: the proposal-list-revamp merge |
| 1259 | f_base = _write_py(root, "f.py", "v = 0\n") |
| 1260 | c_5e6c = _make_commit(root, repo_id, branch="main", message="5e6c: proposal-list-revamp", |
| 1261 | manifest={"f.py": f_base}) |
| 1262 | |
| 1263 | # 727dad83 equivalent: pool_pre_ping fix on top of 5e6c6476 |
| 1264 | f_pp = _write_py(root, "database.py", "pool_pre_ping = True\n") |
| 1265 | c_727d = _make_commit(root, repo_id, branch="main", message="727d: pool_pre_ping", |
| 1266 | manifest={"f.py": f_base, "database.py": f_pp}) |
| 1267 | |
| 1268 | # d40f74ba equivalent: fix-branch HEAD (includes 727dad83 ancestor) |
| 1269 | f_ex = _write_py(root, "executor.py", "args = [] # fixed\n") |
| 1270 | (ref_path(root, "fix-branch")).write_text(c_727d) |
| 1271 | c_d40f = _make_commit(root, repo_id, branch="fix-branch", |
| 1272 | message="d40f: executor fix", |
| 1273 | manifest={"f.py": f_base, "database.py": f_pp, |
| 1274 | "executor.py": f_ex}) |
| 1275 | |
| 1276 | # e01007b4 equivalent: MuseHub merge of fix-branch into dev |
| 1277 | # parents: [5e6c6476, d40f74ba] |
| 1278 | (ref_path(root, "remote-dev")).write_text(c_5e6c) |
| 1279 | c_e010 = _make_commit(root, repo_id, branch="remote-dev", |
| 1280 | message="e010: Merge fix-branch into dev", |
| 1281 | manifest={"f.py": f_base, "database.py": f_pp, |
| 1282 | "executor.py": f_ex}, |
| 1283 | parent_commit_id=c_5e6c, |
| 1284 | parent2_commit_id=c_d40f) |
| 1285 | |
| 1286 | # The merge base of local dev (727dad83) and remote dev (e01007b4). |
| 1287 | lca = find_merge_base(root, c_727d, c_e010) |
| 1288 | assert lca == c_5e6c, ( |
| 1289 | f"LCA(727dad83, e01007b4) should be 5e6c6476, got {lca}. " |
| 1290 | "With the wrong LCA, the three-way merge computes wrong change-sets " |
| 1291 | "and silently drops theirs-only files." |
| 1292 | ) |
| 1293 | |
| 1294 | |
| 1295 | # =========================================================================== |
| 1296 | # I — False-conflict regression: theirs-only additions when ours==base |
| 1297 | # =========================================================================== |
| 1298 | # Real incident: muse/core/patch_record.py was added on dev. When dev was |
| 1299 | # merged into main, main's HEAD was a previous merge commit whose snapshot |
| 1300 | # was IDENTICAL to the merge base snapshot (the prior merge had introduced |
| 1301 | # no net manifest changes). The engine falsely reported patch_record.py as |
| 1302 | # a conflict and apply_manifest deleted it from disk. |
| 1303 | # |
| 1304 | # Root invariant: if base_manifest[p] is absent AND ours_manifest[p] is absent |
| 1305 | # AND theirs_manifest[p] is present → this is a PURE THEIRS ADDITION. It must |
| 1306 | # NEVER appear in conflict_paths. It MUST appear on disk after the merge stops. |
| 1307 | # =========================================================================== |
| 1308 | |
| 1309 | |
| 1310 | class TestFalseConflictTheirsOnlyI: |
| 1311 | """I: theirs-only additions must never be false-conflicted or deleted.""" |
| 1312 | |
| 1313 | def test_I1_theirs_only_addition_not_in_conflict_list( |
| 1314 | self, tmp_path: pathlib.Path |
| 1315 | ) -> None: |
| 1316 | """I1: when ours-snapshot == base-snapshot, theirs-only new files are clean.""" |
| 1317 | root, repo_id = _init_code_repo(tmp_path) |
| 1318 | |
| 1319 | # Base commit: a.py only |
| 1320 | a_oid = _write_py(root, "a.py", "A = 1\n") |
| 1321 | base_c = _make_commit(root, repo_id, branch="main", message="base: a.py", |
| 1322 | manifest={"a.py": a_oid}) |
| 1323 | |
| 1324 | # Main: a no-op merge commit (snapshot identical to base — mirrors real incident |
| 1325 | # where main's last commit was a merge that produced no manifest changes). |
| 1326 | noop_c = _make_commit(root, repo_id, branch="main", message="Merge dev into main (noop)", |
| 1327 | manifest={"a.py": a_oid}, |
| 1328 | parent_commit_id=base_c, |
| 1329 | parent2_commit_id=base_c) |
| 1330 | |
| 1331 | # Dev: adds patch_record.py — theirs-only addition |
| 1332 | pr_oid = _write_py(root, "patch_record.py", |
| 1333 | "\"\"\"Patch record.\"\"\"\n\nclass PatchRecord:\n pass\n") |
| 1334 | (ref_path(root, "dev")).write_text(base_c) |
| 1335 | dev_c = _make_commit(root, repo_id, branch="dev", message="feat: add patch_record", |
| 1336 | manifest={"a.py": a_oid, "patch_record.py": pr_oid}, |
| 1337 | parent_commit_id=base_c) |
| 1338 | |
| 1339 | result = runner.invoke(cli, ["merge", "--force", "--json", "dev"], |
| 1340 | env=_env(root)) |
| 1341 | assert result.exit_code == 0, f"merge failed:\n{result.output}" |
| 1342 | data = json.loads(result.output) |
| 1343 | assert "patch_record.py" not in data.get("conflicts", []), ( |
| 1344 | "patch_record.py is a pure theirs-only addition — must not appear in conflicts" |
| 1345 | ) |
| 1346 | |
| 1347 | def test_I2_theirs_only_addition_lands_on_disk( |
| 1348 | self, tmp_path: pathlib.Path |
| 1349 | ) -> None: |
| 1350 | """I2: theirs-only file must exist on disk after merge (not deleted by apply_manifest).""" |
| 1351 | root, repo_id = _init_code_repo(tmp_path) |
| 1352 | |
| 1353 | a_oid = _write_py(root, "a.py", "A = 1\n") |
| 1354 | base_c = _make_commit(root, repo_id, branch="main", message="base: a.py", |
| 1355 | manifest={"a.py": a_oid}) |
| 1356 | |
| 1357 | # Write a.py to disk so the workdir guard doesn't fire |
| 1358 | (root / "a.py").write_bytes(b"A = 1\n") |
| 1359 | |
| 1360 | noop_c = _make_commit(root, repo_id, branch="main", message="Merge dev into main (noop)", |
| 1361 | manifest={"a.py": a_oid}, |
| 1362 | parent_commit_id=base_c, |
| 1363 | parent2_commit_id=base_c) |
| 1364 | |
| 1365 | pr_content = b"\"\"\"Patch record.\"\"\"\n\nclass PatchRecord:\n pass\n" |
| 1366 | pr_oid = _write_object(root, pr_content) |
| 1367 | (ref_path(root, "dev")).write_text(base_c) |
| 1368 | dev_c = _make_commit(root, repo_id, branch="dev", message="feat: add patch_record", |
| 1369 | manifest={"a.py": a_oid, "patch_record.py": pr_oid}, |
| 1370 | parent_commit_id=base_c) |
| 1371 | |
| 1372 | result = runner.invoke(cli, ["merge", "--force", "--json", "dev"], |
| 1373 | env=_env(root)) |
| 1374 | assert result.exit_code == 0, f"merge failed:\n{result.output}" |
| 1375 | # On a clean merge, patch_record.py must be written to disk |
| 1376 | data = json.loads(result.output) |
| 1377 | if data["status"] == "merged": |
| 1378 | assert (root / "patch_record.py").exists(), ( |
| 1379 | "patch_record.py must exist on disk after clean merge — " |
| 1380 | "apply_manifest must not delete it" |
| 1381 | ) |
| 1382 | |
| 1383 | def test_I3_merge_succeeds_cleanly_when_ours_equals_base_snapshot( |
| 1384 | self, tmp_path: pathlib.Path |
| 1385 | ) -> None: |
| 1386 | """I3: merge status must be 'merged' or 'fast_forward', never 'conflict' |
| 1387 | when ours snapshot equals base snapshot and theirs only adds files.""" |
| 1388 | root, repo_id = _init_code_repo(tmp_path) |
| 1389 | |
| 1390 | a_oid = _write_py(root, "a.py", "A = 1\n") |
| 1391 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 1392 | manifest={"a.py": a_oid}) |
| 1393 | |
| 1394 | # ours == base snapshot exactly |
| 1395 | noop_c = _make_commit(root, repo_id, branch="main", message="noop merge", |
| 1396 | manifest={"a.py": a_oid}, |
| 1397 | parent_commit_id=base_c, |
| 1398 | parent2_commit_id=base_c) |
| 1399 | |
| 1400 | new_oid = _write_py(root, "new_module.py", "X = True\n") |
| 1401 | (ref_path(root, "dev")).write_text(base_c) |
| 1402 | _make_commit(root, repo_id, branch="dev", message="add new_module.py", |
| 1403 | manifest={"a.py": a_oid, "new_module.py": new_oid}, |
| 1404 | parent_commit_id=base_c) |
| 1405 | |
| 1406 | result = runner.invoke(cli, ["merge", "--force", "--json", "dev"], |
| 1407 | env=_env(root)) |
| 1408 | assert result.exit_code == 0 |
| 1409 | data = json.loads(result.output) |
| 1410 | assert data["status"] in ("merged", "fast_forward"), ( |
| 1411 | f"expected clean merge, got status={data['status']!r}; " |
| 1412 | f"conflicts={data.get('conflicts')}" |
| 1413 | ) |
| 1414 | |
| 1415 | def test_I4_multiple_theirs_only_files_no_conflict_when_ours_equals_base( |
| 1416 | self, tmp_path: pathlib.Path |
| 1417 | ) -> None: |
| 1418 | """I4: multiple theirs-only additions, none must appear in conflict list.""" |
| 1419 | root, repo_id = _init_code_repo(tmp_path) |
| 1420 | |
| 1421 | a_oid = _write_py(root, "a.py", "A = 1\n") |
| 1422 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 1423 | manifest={"a.py": a_oid}) |
| 1424 | noop_c = _make_commit(root, repo_id, branch="main", message="noop", |
| 1425 | manifest={"a.py": a_oid}, |
| 1426 | parent_commit_id=base_c, |
| 1427 | parent2_commit_id=base_c) |
| 1428 | |
| 1429 | theirs_manifest: Manifest = {"a.py": a_oid} |
| 1430 | new_files = ["patch_record.py", "apply_patch.py", "format_patch.py", |
| 1431 | "patch_utils.py", "patch_schema.py"] |
| 1432 | for fname in new_files: |
| 1433 | oid = _write_py(root, fname, f"# {fname}\n") |
| 1434 | theirs_manifest[fname] = oid |
| 1435 | |
| 1436 | (ref_path(root, "dev")).write_text(base_c) |
| 1437 | _make_commit(root, repo_id, branch="dev", message="add patch files", |
| 1438 | manifest=theirs_manifest, parent_commit_id=base_c) |
| 1439 | |
| 1440 | result = runner.invoke(cli, ["merge", "--force", "--json", "dev"], |
| 1441 | env=_env(root)) |
| 1442 | assert result.exit_code == 0 |
| 1443 | data = json.loads(result.output) |
| 1444 | conflicts = data.get("conflicts", []) |
| 1445 | for fname in new_files: |
| 1446 | assert fname not in conflicts, ( |
| 1447 | f"{fname} is a pure theirs-only addition — must not appear in conflicts. " |
| 1448 | f"Full conflict list: {conflicts}" |
| 1449 | ) |
| 1450 | |
| 1451 | def test_I5_partial_merged_manifest_must_include_theirs_only_files_on_conflict( |
| 1452 | self, tmp_path: pathlib.Path |
| 1453 | ) -> None: |
| 1454 | """I5: when there IS a genuine conflict elsewhere, theirs-only additions |
| 1455 | must still be in the working tree (apply_manifest must not delete them).""" |
| 1456 | root, repo_id = _init_code_repo(tmp_path) |
| 1457 | |
| 1458 | a_oid = _write_py(root, "a.py", "def go(): return 'base'\n") |
| 1459 | b_oid = _write_py(root, "b.py", "B = True\n") |
| 1460 | base_c = _make_commit(root, repo_id, branch="main", message="base", |
| 1461 | manifest={"a.py": a_oid, "b.py": b_oid}) |
| 1462 | |
| 1463 | # ours: modifies a.py (causing conflict), same snapshot as base otherwise |
| 1464 | a_ours = _write_py(root, "a.py", "def go(): return 'ours'\n") |
| 1465 | noop_c = _make_commit(root, repo_id, branch="main", message="ours: change a.py", |
| 1466 | manifest={"a.py": a_ours, "b.py": b_oid}, |
| 1467 | parent_commit_id=base_c) |
| 1468 | |
| 1469 | # Write working tree for ours |
| 1470 | (root / "a.py").write_bytes(b"def go(): return 'ours'\n") |
| 1471 | (root / "b.py").write_bytes(b"B = True\n") |
| 1472 | |
| 1473 | # theirs: modifies a.py differently + adds new_module.py |
| 1474 | a_theirs = _write_py(root, "a.py", "def go(): return 'theirs'\n") |
| 1475 | new_oid = _write_object(root, b"NEW = True\n") |
| 1476 | (ref_path(root, "dev")).write_text(base_c) |
| 1477 | _make_commit(root, repo_id, branch="dev", message="theirs: change a + add new", |
| 1478 | manifest={"a.py": a_theirs, "b.py": b_oid, "new_module.py": new_oid}, |
| 1479 | parent_commit_id=base_c) |
| 1480 | |
| 1481 | result = runner.invoke(cli, ["merge", "--force", "--json", "dev"], |
| 1482 | env=_env(root)) |
| 1483 | data = json.loads(result.output) |
| 1484 | |
| 1485 | # There should be a conflict on a.py, but new_module.py must NOT be in conflicts. |
| 1486 | assert "new_module.py" not in data.get("conflicts", []), ( |
| 1487 | "new_module.py is theirs-only — must not appear in conflicts even when " |
| 1488 | "there is a genuine conflict in a.py" |
| 1489 | ) |
| 1490 | if data["status"] == "conflict": |
| 1491 | # The partial_merged manifest (applied to disk) must contain new_module.py. |
| 1492 | # Verify it's on disk — if apply_manifest deleted it, that's the bug. |
| 1493 | assert (root / "new_module.py").exists(), ( |
| 1494 | "new_module.py must be on disk after conflict-stop. " |
| 1495 | "apply_manifest must include theirs-only files in partial_merged, " |
| 1496 | "not delete them because they're absent from ours_manifest." |
| 1497 | ) |
File History
3 commits
sha256:36c3cb3e76619d4c30a6d9bf81b5ec4ff148e30dcfed913e3114ca7b43b81c7e
fix: rename objects→blobs in push client and all stale test…
Sonnet 4.6
patch
22 days ago
sha256:c06a9b9b9fee26c68ea725b44d54b2c0a171301ce9de746d5b656617b4463a9a
fix: repair four test failures from post-migration audit
Sonnet 4.6
patch
28 days ago
sha256:1900655993c83c4107067375548a7be823e471d2515830842f1a12cba4bd3cdf
fix: unified object store migration — idempotent writes, JS…
Sonnet 4.6
minor
⚠
29 days ago