Merge branch 'master' into feat/issue-658-mcp-restart-coordinator

This commit is contained in:
2026-07-24 05:48:25 -05:00
15 changed files with 3282 additions and 29 deletions
+304 -20
View File
@@ -2446,6 +2446,20 @@ def _evaluate_issue_lock_recovery(
descendant_sha=local_head,
)
# #871: the inverse of the #768 descendant relation — the *remote* head may
# have advanced past the local/recorded head via a sanctioned merge-based
# branch sync (``gitea_update_pr_branch_by_merge``) while the local worktree
# stayed put. Observe that provenance server-side so the assessor can prove
# it and nothing else. Probed only when the heads differ; never from any
# caller-supplied value.
sync_provenance: dict | None = None
if remote_head and local_head and remote_head != local_head:
sync_provenance = issue_lock_worktree.read_merge_sync_provenance(
worktree_path,
prior_head_sha=local_head,
synced_head_sha=remote_head,
)
# #772: with no remote branch there is no head to measure against, so the
# base the branch was cut from is observed instead. Probed only in that
# case, so the published path's evidence is untouched (#772 AC8).
@@ -2488,6 +2502,7 @@ def _evaluate_issue_lock_recovery(
remote_branch_exists=remote_branch_exists,
recorded_base_sha=recorded_base,
base_ancestry=base_ancestry,
sync_provenance=sync_provenance,
)
@@ -11117,6 +11132,9 @@ def _collect_branch_ownership_records(
"""
records: list[dict] = []
inventory_error = False
# #855 AC4: expired/stale reviewer-lease records eligible for an explicit
# reclaim decision, evaluated after the full ownership inventory is built.
reviewer_reclaim_candidates: list[tuple[dict, bool | None]] = []
target_branch = (branch or "").strip()
if not target_branch:
return {"records": records, "inventory_error": False}
@@ -11265,15 +11283,28 @@ def _collect_branch_ownership_records(
else:
status = freshness_status
reclaim_allowed = False
records.append(
_base_rec(
category=category,
status=status,
reclaim_allowed=reclaim_allowed,
role=role,
host=lease_host or host_n or host,
)
rec = _base_rec(
category=category,
status=status,
reclaim_allowed=reclaim_allowed,
role=role,
host=lease_host or host_n or host,
)
records.append(rec)
# #855 AC4: a reviewer lease that is expired/stale (its owner
# gone) becomes a candidate for an explicit, fail-closed
# reclaim decision made once the full inventory is known.
if (
role == "reviewer"
and status
in branch_cleanup_guard._RECLAIMABLE_REVIEWER_STATUSES
):
owner_alive = (
fr.get("owner_pid_alive") if isinstance(fr, dict) else None
)
reviewer_reclaim_candidates.append(
(rec, owner_alive if isinstance(owner_alive, bool) else None)
)
except Exception:
# O1: fail closed on control-plane inventory errors.
inventory_error = True
@@ -11344,6 +11375,44 @@ def _collect_branch_ownership_records(
)
)
# #855 AC4: decide, explicitly and fail-closed, whether any expired/stale
# reviewer lease may stop protecting an already-merged branch. This runs
# only after the full ownership inventory is built, so a competing active
# claimant (an active lease, author session, worktree binding, or active
# reviewer comment lease) is visible. An inventory failure keeps every
# reclaim candidate protective (reclaim_allowed stays False).
if reviewer_reclaim_candidates and not inventory_error:
pr_merged_state: bool | None = None
if pr_number is not None and auth and base_api:
try:
pr_live = api_request(
"GET", f"{base_api}/pulls/{int(pr_number)}", auth
)
if isinstance(pr_live, dict) and pr_live:
pr_merged_state = bool(
pr_live.get("merged") or pr_live.get("merged_at")
)
except Exception:
# Unknown merged state fails closed (candidate stays protective).
pr_merged_state = None
for cand_rec, owner_alive in reviewer_reclaim_candidates:
competing = any(
other is not cand_rec
and branch_cleanup_guard.is_active_ownership_status(
other.get("status")
)
for other in records
)
decision = branch_cleanup_guard.assess_expired_reviewer_lease_reclaim(
role=str(cand_rec.get("role")),
status=str(cand_rec.get("status")),
pr_merged=pr_merged_state,
owner_pid_alive=owner_alive,
competing_active_claimant=competing,
)
cand_rec["reclaim_allowed"] = decision["reclaim_allowed"]
cand_rec["reclaim_decision"] = decision["decision"]
return {"records": records, "inventory_error": inventory_error}
@@ -11406,6 +11475,7 @@ def gitea_reconcile_merged_cleanups(
dry_run: bool = True,
execute_confirmed: bool = False,
limit: int = 50,
pr_number: int | None = None,
remote: str = "dadeschools",
host: str | None = None,
org: str | None = None,
@@ -11416,7 +11486,11 @@ def gitea_reconcile_merged_cleanups(
Args:
dry_run: Defaults to True. When True, only builds the reconciliation report.
execute_confirmed: Must be True when dry_run=False.
limit: Max number of closed PRs to inspect.
limit: Max number of closed PRs to inspect (batch mode only; ignored when
``pr_number`` is set).
pr_number: Optional exact merged PR selector (#855). When set, only that
PR is assessed/acted on (fail closed if missing, unmerged, or
ambiguous). When omitted, existing batch behaviour is preserved.
remote: Known Gitea instance ('dadeschools' or 'prgs').
host: Override the Gitea host.
org: Override the owner/organization.
@@ -11451,11 +11525,120 @@ def gitea_reconcile_merged_cleanups(
"audit_phase": audit_reconciliation_mode.current_phase(),
}
# #855: optional exact PR pin. Fail closed before any inventory mutation.
exact_pr: int | None = None
if pr_number is not None:
try:
exact_pr = int(pr_number)
except (TypeError, ValueError):
return {
"success": False,
"performed": False,
"executed": False,
"dry_run": bool(dry_run),
"selection_mode": "exact_pr",
"selected_pr_number": pr_number,
"reasons": [
f"pr_number={pr_number!r} is not a valid integer "
"(fail closed; no mutation)"
],
"blocker_kind": "invalid_pr_number",
}
if exact_pr <= 0:
return {
"success": False,
"performed": False,
"executed": False,
"dry_run": bool(dry_run),
"selection_mode": "exact_pr",
"selected_pr_number": exact_pr,
"reasons": [
f"pr_number={exact_pr} must be a positive integer "
"(fail closed; no mutation)"
],
"blocker_kind": "invalid_pr_number",
}
h, o, r = _resolve(remote, host, org, repo)
auth = _auth(h)
base = repo_api_url(h, o, r)
closed_prs = api_get_all(f"{base}/pulls?state=closed", auth, limit=limit)
open_prs = api_get_all(f"{base}/pulls?state=open", auth)
selection_mode = "batch"
closed_prs: list[dict] = []
open_prs: list[dict] = []
if exact_pr is not None:
selection_mode = "exact_pr"
try:
pr_live = api_request("GET", f"{base}/pulls/{exact_pr}", auth)
except Exception as exc:
return {
"success": False,
"performed": False,
"executed": False,
"dry_run": bool(dry_run),
"selection_mode": selection_mode,
"selected_pr_number": exact_pr,
"reasons": [
f"PR #{exact_pr} could not be uniquely resolved "
f"(fail closed; no mutation): {_redact(str(exc))}"
],
"blocker_kind": "pr_unresolvable",
}
if not isinstance(pr_live, dict) or not pr_live:
return {
"success": False,
"performed": False,
"executed": False,
"dry_run": bool(dry_run),
"selection_mode": selection_mode,
"selected_pr_number": exact_pr,
"reasons": [
f"PR #{exact_pr} could not be uniquely resolved "
"(empty response; fail closed; no mutation)"
],
"blocker_kind": "pr_unresolvable",
}
live_number = pr_live.get("number")
try:
live_number_int = int(live_number) if live_number is not None else None
except (TypeError, ValueError):
live_number_int = None
if live_number_int != exact_pr:
return {
"success": False,
"performed": False,
"executed": False,
"dry_run": bool(dry_run),
"selection_mode": selection_mode,
"selected_pr_number": exact_pr,
"reasons": [
f"PR #{exact_pr} resolution is ambiguous or mismatched "
f"(live number={live_number!r}; fail closed; no mutation)"
],
"blocker_kind": "pr_ambiguous",
}
if not (pr_live.get("merged") or pr_live.get("merged_at")):
return {
"success": False,
"performed": False,
"executed": False,
"dry_run": bool(dry_run),
"selection_mode": selection_mode,
"selected_pr_number": exact_pr,
"reasons": [
f"PR #{exact_pr} is not merged "
"(exact-target cleanup requires a merged PR; "
"fail closed; no mutation)"
],
"blocker_kind": "pr_not_merged",
}
closed_prs = [pr_live]
# Exact mode still needs open heads for remote-delete safety gates.
open_prs = api_get_all(f"{base}/pulls?state=open", auth)
else:
# Preserve historical call order (closed then open) for batch callers/tests.
closed_prs = api_get_all(f"{base}/pulls?state=closed", auth, limit=limit)
open_prs = api_get_all(f"{base}/pulls?state=open", auth)
merged_closed: list[dict] = []
remote_branch_exists: dict[str, bool] = {}
@@ -11482,6 +11665,13 @@ def gitea_reconcile_merged_cleanups(
scratch_candidates = merged_cleanup_reconcile.discover_reviewer_scratch_worktrees(
_canonical_local_git_root()
)
# #855: exact-target never inventories or mutates foreign PR scratch trees.
if exact_pr is not None:
scratch_candidates = [
s
for s in scratch_candidates
if int(s.get("pr_number") or 0) == int(exact_pr)
]
active_reviewer_leases: dict[int, bool] = {}
pr_states: dict[int, dict] = {}
for scratch in scratch_candidates:
@@ -11516,6 +11706,33 @@ def gitea_reconcile_merged_cleanups(
active_reviewer_leases=active_reviewer_leases,
pr_states=pr_states,
)
report["selection_mode"] = selection_mode
if exact_pr is not None:
report["selected_pr_number"] = exact_pr
# Fail closed if exact pin somehow produced other or zero entries.
entries = list(report.get("entries") or [])
entry_numbers = []
for entry in entries:
try:
entry_numbers.append(int(entry.get("pr_number")))
except (TypeError, ValueError):
entry_numbers.append(entry.get("pr_number"))
if entry_numbers != [exact_pr]:
return {
"success": False,
"performed": False,
"executed": False,
"dry_run": bool(dry_run),
"selection_mode": selection_mode,
"selected_pr_number": exact_pr,
"reasons": [
f"exact PR #{exact_pr} selection produced unexpected "
f"candidate set {entry_numbers!r} "
"(fail closed; no mutation)"
],
"blocker_kind": "exact_selection_mismatch",
"entries": entries,
}
if dry_run:
report["dry_run"] = True
@@ -18890,10 +19107,59 @@ def gitea_update_pr_branch_by_merge(
prepared_verdict_head_sha=live_pr_head,
)
return {
"success": True,
# #871: the remote head is now advanced; the durable linked-issue lock must
# be advanced with it, or a later dead-session recovery can never prove
# ownership at the new head. This runs AFTER the successful remote update, so
# a failure here is a *partial* lifecycle failure — the remote moved but the
# durable state did not — and must never be reported as a full success.
claimant = _work_lease_claimant(h)
matched_issue = ownership.get("matched_issue")
synced_at = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
lock_refresh: dict = {
"refreshed": False,
"reasons": ["durable lock head refresh was not attempted"],
}
if new_head and matched_issue and source_branch and (wt or None):
try:
lock_refresh = issue_lock_store.apply_durable_lock_head_refresh(
remote=remote,
org=o,
repo=r,
issue_number=int(matched_issue),
branch_name=source_branch,
worktree_path=wt,
pr_number=pr_number,
identity=claimant.get("username"),
profile=claimant.get("profile"),
current_pid=os.getpid(),
expected_old_head=live_pr_head,
new_head=new_head,
synced_at=synced_at,
base_head=live_base_head,
)
except Exception as exc:
lock_refresh = {
"refreshed": False,
"reasons": [
f"durable lock head refresh raised (fail closed): {_redact(str(exc))}"
],
}
else:
lock_refresh = {
"refreshed": False,
"reasons": [
"durable lock head refresh could not run: missing new head, "
"linked issue, source branch, or worktree binding"
],
}
durable_refreshed = bool(lock_refresh.get("refreshed"))
base_result = {
"performed": True,
"mutation_allowed": True,
"durable_lock_refreshed": durable_refreshed,
"durable_lock_refresh": lock_refresh,
"fully_synchronized": durable_refreshed,
"style": "merge",
"force_push": False,
"rebase": False,
@@ -18915,19 +19181,37 @@ def gitea_update_pr_branch_by_merge(
"prepared_verdict_invalidated": transition.get(
"prepared_verdict_invalidated"
),
"recommended_next_action": transition.get(
"recommended_next_action",
pr_sync_status.ACTION_FRESH_REVIEW_REQUIRED,
),
"transition": transition,
"role_kind": role,
"profile_name": profile.get("profile_name"),
"worktree_path": wt or None,
"reasons": list(transition.get("reasons") or []) + [
"update-by-merge completed via native Gitea API (style=merge only)"
],
}
if durable_refreshed:
base_result["success"] = True
base_result["recommended_next_action"] = transition.get(
"recommended_next_action",
pr_sync_status.ACTION_FRESH_REVIEW_REQUIRED,
)
base_result["reasons"] = list(transition.get("reasons") or []) + [
"update-by-merge completed via native Gitea API (style=merge only)",
f"durable linked-issue lock #{matched_issue} head refreshed to "
f"{new_head} (verified by read-after-write)",
]
return base_result
# Partial lifecycle failure: the remote advanced but the durable lock did
# not. Do NOT report a fully successful synchronization (#871).
base_result["success"] = False
base_result["partial_lifecycle_failure"] = True
base_result["recommended_next_action"] = pr_sync_status.ACTION_BLOCKED
base_result["reasons"] = list(lock_refresh.get("reasons") or []) + [
f"PARTIAL LIFECYCLE FAILURE: PR #{pr_number} remote head advanced to "
f"{new_head} but the durable linked-issue lock head was not refreshed; "
"the synchronization is NOT complete",
]
return base_result
@mcp.tool()
def gitea_assess_conflict_fix_push(