Merge commit 'b5ee6567aade4fdce4c57d826184bfd01ebb522e' into HEAD
# Conflicts: # tests/test_resolve_task_capability.py
This commit is contained in:
@@ -16,6 +16,7 @@ def test_all_workflow_files_exist():
|
||||
"reconcile-landed-pr.md",
|
||||
"create-issue.md",
|
||||
"work-issue.md",
|
||||
"pr-queue-cleanup.md",
|
||||
):
|
||||
assert (SKILL_DIR / "workflows" / name).is_file(), name
|
||||
|
||||
@@ -26,6 +27,7 @@ def test_skill_references_all_workflow_files():
|
||||
"workflows/reconcile-landed-pr.md",
|
||||
"workflows/create-issue.md",
|
||||
"workflows/work-issue.md",
|
||||
"workflows/pr-queue-cleanup.md",
|
||||
):
|
||||
assert name in SKILL
|
||||
|
||||
@@ -36,6 +38,7 @@ def test_skill_contains_mode_isolation_language():
|
||||
assert "reconcile-landed-pr" in SKILL
|
||||
assert "create-issue" in SKILL
|
||||
assert "work-issue" in SKILL
|
||||
assert "pr-queue-cleanup" in SKILL
|
||||
assert "Do not mix modes" in SKILL or "do not mix modes" in SKILL.lower()
|
||||
|
||||
|
||||
@@ -99,12 +102,24 @@ def test_work_issue_workflow_contract():
|
||||
assert "## 21. PR creation rules" in text
|
||||
|
||||
|
||||
def test_pr_queue_cleanup_workflow_contract():
|
||||
text = (SKILL_DIR / "workflows" / "pr-queue-cleanup.md").read_text(
|
||||
encoding="utf-8"
|
||||
)
|
||||
assert "canonical: true" in text
|
||||
assert "exactly one" in text
|
||||
assert "check_cleanup_task_allowed" in text
|
||||
assert "resolve_cleanup_run_state" in text
|
||||
assert "Next suggested PR" in text
|
||||
|
||||
|
||||
def test_final_report_schemas_exist():
|
||||
for name in (
|
||||
"review-merge-final-report.md",
|
||||
"reconcile-landed-final-report.md",
|
||||
"create-issue-final-report.md",
|
||||
"work-issue-final-report.md",
|
||||
"pr-queue-cleanup-final-report.md",
|
||||
):
|
||||
assert (SKILL_DIR / "schemas" / name).is_file(), name
|
||||
|
||||
@@ -124,6 +139,20 @@ def test_start_issue_template_references_work_issue_workflow():
|
||||
assert "workflows/work-issue.md" in text
|
||||
|
||||
|
||||
def test_pr_queue_cleanup_template_references_workflow():
|
||||
text = (SKILL_DIR / "templates" / "pr-queue-cleanup.md").read_text(
|
||||
encoding="utf-8"
|
||||
)
|
||||
assert "workflows/pr-queue-cleanup.md" in text
|
||||
assert "pr_queue_cleanup" in text
|
||||
|
||||
|
||||
def test_pr_queue_cleanup_verifier_exported():
|
||||
from review_proofs import assess_pr_queue_cleanup_report
|
||||
|
||||
assert callable(assess_pr_queue_cleanup_report)
|
||||
|
||||
|
||||
def test_reviewer_fallback_verifier_exported():
|
||||
from review_proofs import assess_reviewer_fallback_report
|
||||
|
||||
@@ -172,6 +201,12 @@ def test_validation_worktree_edit_verifier_exported():
|
||||
assert callable(assess_validation_worktree_edit_report)
|
||||
|
||||
|
||||
def test_already_landed_handoff_verifier_exported():
|
||||
from review_proofs import assess_already_landed_handoff_report
|
||||
|
||||
assert callable(assess_already_landed_handoff_report)
|
||||
|
||||
|
||||
def test_inventory_worktree_verifier_exported():
|
||||
from review_proofs import assess_inventory_worktree_report
|
||||
|
||||
@@ -193,4 +228,23 @@ def test_mutation_categories_verifier_exported():
|
||||
def test_worktree_ownership_verifier_exported():
|
||||
from review_proofs import assess_worktree_ownership_report
|
||||
|
||||
assert callable(assess_worktree_ownership_report)
|
||||
assert callable(assess_worktree_ownership_report)
|
||||
|
||||
|
||||
def test_pr_queue_cleanup_verifier_exported():
|
||||
from review_proofs import assess_pr_queue_cleanup_report
|
||||
|
||||
assert callable(assess_pr_queue_cleanup_report)
|
||||
|
||||
|
||||
def test_pr_queue_cleanup_workflow_contract():
|
||||
text = (SKILL_DIR / "workflows" / "pr-queue-cleanup.md").read_text(encoding="utf-8")
|
||||
assert "canonical: true" in text
|
||||
assert "task_mode: pr-queue-cleanup" in text
|
||||
assert "## 1. Mode isolation" in text
|
||||
assert "## 4. Terminal mutation chain" in text
|
||||
assert "## 5. Fresh run per PR" in text
|
||||
assert "exactly one" in text.lower()
|
||||
assert "review-merge-pr.md" in text
|
||||
assert "Next suggested PR" in text
|
||||
assert "schemas/pr-queue-cleanup-final-report.md" in text
|
||||
|
||||
@@ -0,0 +1,269 @@
|
||||
"""Tests for PR-only queue cleanup mode (#390)."""
|
||||
import sys
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
|
||||
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
|
||||
|
||||
from pr_queue_cleanup import (
|
||||
CLEANUP_FORBIDDEN_TASKS,
|
||||
CLEANUP_WORKFLOW_PATH,
|
||||
CONTINUE_TO_SAME_PR_MERGE,
|
||||
STOP_AFTER_DECISION,
|
||||
STOP_AFTER_MERGE,
|
||||
STOP_AFTER_MERGE_BLOCKER,
|
||||
STOP_AFTER_REQUEST_CHANGES,
|
||||
STOP_APPROVED_MERGE_GATES_FAILED,
|
||||
STOP_APPROVED_NO_MERGE_AUTH,
|
||||
STOP_GATE_NOT_PROVEN,
|
||||
assess_pr_queue_cleanup_report,
|
||||
check_cleanup_task_allowed,
|
||||
resolve_cleanup_run_state,
|
||||
)
|
||||
from review_proofs import assess_pr_queue_cleanup_report as proofs_assess
|
||||
from role_session_router import REVIEWER_TASKS, route_task_session
|
||||
from task_capability_map import required_permission, required_role
|
||||
|
||||
|
||||
class TestCleanupCapabilityMapping(unittest.TestCase):
|
||||
def test_cleanup_task_is_reviewer_role(self):
|
||||
for key in ("pr_queue_cleanup", "pr-queue-cleanup"):
|
||||
self.assertEqual(required_permission(key), "gitea.pr.review")
|
||||
self.assertEqual(required_role(key), "reviewer")
|
||||
|
||||
def test_cleanup_task_registered_as_reviewer_route(self):
|
||||
self.assertIn("pr_queue_cleanup", REVIEWER_TASKS)
|
||||
|
||||
def test_author_session_gets_wrong_role_stop(self):
|
||||
result = route_task_session(
|
||||
"pr_queue_cleanup",
|
||||
active_profile="prgs-author",
|
||||
active_role_kind="author",
|
||||
allowed_in_current_session=False,
|
||||
)
|
||||
self.assertEqual(result["route_result"], "wrong_role_stop")
|
||||
self.assertFalse(result["downstream_allowed"])
|
||||
|
||||
def test_reviewer_session_allowed(self):
|
||||
result = route_task_session(
|
||||
"pr_queue_cleanup",
|
||||
active_profile="prgs-reviewer",
|
||||
active_role_kind="reviewer",
|
||||
allowed_in_current_session=True,
|
||||
)
|
||||
self.assertEqual(result["route_result"], "allowed_current_session")
|
||||
self.assertTrue(result["downstream_allowed"])
|
||||
|
||||
|
||||
class TestCleanupTaskGate(unittest.TestCase):
|
||||
def test_author_tasks_blocked(self):
|
||||
for task in (
|
||||
"claim_issue",
|
||||
"mark_issue",
|
||||
"lock_issue",
|
||||
"create_issue",
|
||||
"create_branch",
|
||||
"push_branch",
|
||||
"create_pr",
|
||||
"commit_files",
|
||||
):
|
||||
allowed, reasons = check_cleanup_task_allowed(task)
|
||||
self.assertFalse(allowed, task)
|
||||
self.assertTrue(reasons, task)
|
||||
|
||||
def test_review_task_allowed(self):
|
||||
allowed, reasons = check_cleanup_task_allowed("review_pr")
|
||||
self.assertTrue(allowed)
|
||||
self.assertEqual(reasons, [])
|
||||
|
||||
def test_forbidden_set_covers_issue_filing_and_impl(self):
|
||||
self.assertIn("create_issue", CLEANUP_FORBIDDEN_TASKS)
|
||||
self.assertIn("create_branch", CLEANUP_FORBIDDEN_TASKS)
|
||||
self.assertIn("commit_files", CLEANUP_FORBIDDEN_TASKS)
|
||||
|
||||
|
||||
class TestCleanupRunState(unittest.TestCase):
|
||||
def test_request_changes_stops_run(self):
|
||||
state = resolve_cleanup_run_state("request_changes")
|
||||
self.assertEqual(state["outcome"], STOP_AFTER_REQUEST_CHANGES)
|
||||
self.assertFalse(state["further_mutation_allowed"])
|
||||
|
||||
def test_comment_and_skip_stop_run(self):
|
||||
for decision in ("comment", "skip"):
|
||||
state = resolve_cleanup_run_state(decision)
|
||||
self.assertEqual(state["outcome"], STOP_AFTER_DECISION)
|
||||
self.assertFalse(state["further_mutation_allowed"])
|
||||
|
||||
def test_approved_without_merge_auth_stops(self):
|
||||
state = resolve_cleanup_run_state("approved", merge_gates_passed=True)
|
||||
self.assertEqual(state["outcome"], STOP_APPROVED_NO_MERGE_AUTH)
|
||||
self.assertFalse(state["further_mutation_allowed"])
|
||||
|
||||
def test_approved_with_merge_auth_continues_to_merge(self):
|
||||
state = resolve_cleanup_run_state(
|
||||
"approved",
|
||||
merge_authorized_for_pr=True,
|
||||
merge_gates_passed=True,
|
||||
)
|
||||
self.assertEqual(state["outcome"], CONTINUE_TO_SAME_PR_MERGE)
|
||||
self.assertTrue(state["further_mutation_allowed"])
|
||||
self.assertEqual(state["allowed_mutation"], "merge same PR only")
|
||||
|
||||
def test_approved_with_auth_but_failed_gates_stops(self):
|
||||
state = resolve_cleanup_run_state(
|
||||
"approved",
|
||||
merge_authorized_for_pr=True,
|
||||
merge_gates_passed=False,
|
||||
)
|
||||
self.assertEqual(state["outcome"], STOP_APPROVED_MERGE_GATES_FAILED)
|
||||
self.assertFalse(state["further_mutation_allowed"])
|
||||
|
||||
def test_merge_completed_stops(self):
|
||||
state = resolve_cleanup_run_state(
|
||||
"approved",
|
||||
merge_authorized_for_pr=True,
|
||||
merge_gates_passed=True,
|
||||
merge_completed=True,
|
||||
)
|
||||
self.assertEqual(state["outcome"], STOP_AFTER_MERGE)
|
||||
self.assertFalse(state["further_mutation_allowed"])
|
||||
|
||||
def test_merge_blocker_stops(self):
|
||||
state = resolve_cleanup_run_state(
|
||||
"approved",
|
||||
merge_authorized_for_pr=True,
|
||||
merge_gates_passed=True,
|
||||
merge_blocker=True,
|
||||
)
|
||||
self.assertEqual(state["outcome"], STOP_AFTER_MERGE_BLOCKER)
|
||||
self.assertFalse(state["further_mutation_allowed"])
|
||||
|
||||
def test_unknown_decision_fails_closed(self):
|
||||
state = resolve_cleanup_run_state("ship_it")
|
||||
self.assertEqual(state["outcome"], STOP_GATE_NOT_PROVEN)
|
||||
self.assertFalse(state["further_mutation_allowed"])
|
||||
|
||||
|
||||
def _clean_report(**overrides):
|
||||
lines = {
|
||||
"task": "Task: pr-queue-cleanup",
|
||||
"workflow": f"Workflow source: {CLEANUP_WORKFLOW_PATH} (hash abc123def456)",
|
||||
"pagination": (
|
||||
"PR inventory pagination proof: inventory_complete=true, final "
|
||||
"page, total_count 15"
|
||||
),
|
||||
"selected": "Selected PR: #371",
|
||||
"decision": "Review decision: approved",
|
||||
"merge_auth": "Merge authorized for PR: false",
|
||||
"merge_result": "Merge result: not attempted",
|
||||
"next": "Next suggested PR: #372",
|
||||
"issue_mut": "Issue mutations: none",
|
||||
"branch_mut": "Branch mutations: none",
|
||||
}
|
||||
lines.update(overrides)
|
||||
return "\n".join(v for v in lines.values() if v)
|
||||
|
||||
|
||||
class TestCleanupReportVerifier(unittest.TestCase):
|
||||
def test_clean_single_pr_report_passes(self):
|
||||
result = assess_pr_queue_cleanup_report(_clean_report())
|
||||
self.assertTrue(result["proven"], result["reasons"])
|
||||
self.assertFalse(result["block"])
|
||||
|
||||
def test_approved_and_merged_with_authorization_passes(self):
|
||||
report = _clean_report(
|
||||
merge_auth="Merge authorized for PR: true",
|
||||
merge_result="Merge result: merged 0123456789ab",
|
||||
)
|
||||
result = assess_pr_queue_cleanup_report(report)
|
||||
self.assertTrue(result["proven"], result["reasons"])
|
||||
|
||||
def test_request_changes_then_stop_passes(self):
|
||||
report = _clean_report(decision="Review decision: request_changes")
|
||||
result = assess_pr_queue_cleanup_report(report)
|
||||
self.assertTrue(result["proven"], result["reasons"])
|
||||
|
||||
def test_multi_pr_selection_blocked(self):
|
||||
report = _clean_report() + "\nSelected PR: #403\n"
|
||||
result = assess_pr_queue_cleanup_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(
|
||||
any("multiple prs" in r.lower() for r in result["reasons"]),
|
||||
result["reasons"],
|
||||
)
|
||||
|
||||
def test_two_terminal_mutations_blocked(self):
|
||||
report = _clean_report() + "\nReview decision: request_changes"
|
||||
result = assess_pr_queue_cleanup_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(
|
||||
any("terminal review mutations" in r for r in result["reasons"]),
|
||||
result["reasons"],
|
||||
)
|
||||
|
||||
def test_missing_pagination_proof_blocked(self):
|
||||
report = _clean_report(
|
||||
pagination="PR inventory pagination proof: inventory listed"
|
||||
)
|
||||
result = assess_pr_queue_cleanup_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(
|
||||
any("pagination" in r.lower() for r in result["reasons"]),
|
||||
result["reasons"],
|
||||
)
|
||||
|
||||
def test_missing_next_suggested_pr_blocked(self):
|
||||
report = _clean_report(next="")
|
||||
result = assess_pr_queue_cleanup_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(
|
||||
any("Next suggested PR" in r for r in result["reasons"]),
|
||||
result["reasons"],
|
||||
)
|
||||
|
||||
def test_merge_without_authorization_blocked(self):
|
||||
report = _clean_report(
|
||||
merge_result="Merge result: merged abc123abc123",
|
||||
)
|
||||
result = assess_pr_queue_cleanup_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(
|
||||
any("authorization" in r.lower() for r in result["reasons"]),
|
||||
result["reasons"],
|
||||
)
|
||||
|
||||
def test_issue_mutations_forbidden(self):
|
||||
report = _clean_report(issue_mut="Issue mutations: gitea_mark_issue")
|
||||
result = assess_pr_queue_cleanup_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(
|
||||
any("issue mutations" in r.lower() for r in result["reasons"]),
|
||||
result["reasons"],
|
||||
)
|
||||
|
||||
def test_branch_mutations_forbidden(self):
|
||||
report = _clean_report(branch_mut="Branch mutations: created feat/x")
|
||||
result = assess_pr_queue_cleanup_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(
|
||||
any("branch mutations" in r.lower() for r in result["reasons"]),
|
||||
result["reasons"],
|
||||
)
|
||||
|
||||
def test_missing_workflow_citation_blocked(self):
|
||||
report = _clean_report(workflow="Workflow source: none")
|
||||
result = assess_pr_queue_cleanup_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(
|
||||
any("pr-queue-cleanup.md" in r for r in result["reasons"]),
|
||||
result["reasons"],
|
||||
)
|
||||
|
||||
def test_review_proofs_wrapper_matches_module(self):
|
||||
direct = assess_pr_queue_cleanup_report(_clean_report())
|
||||
wrapped = proofs_assess(_clean_report())
|
||||
self.assertEqual(direct["proven"], wrapped["proven"])
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -287,6 +287,28 @@ class TestResolveTaskCapability(unittest.TestCase):
|
||||
self.assertEqual(res["required_role_kind"], "reconciler")
|
||||
self.assertFalse(res["allowed_in_current_session"])
|
||||
|
||||
@patch("mcp_server.api_request", return_value={"login": "author-user"})
|
||||
@patch("mcp_server.get_auth_header", return_value="token author-pass")
|
||||
def test_resolve_pr_queue_cleanup_author_profile_blocked(self, _auth, _api):
|
||||
with patch.dict(os.environ, self._env("author-profile")):
|
||||
res = mcp_server.gitea_resolve_task_capability(
|
||||
task="pr_queue_cleanup", remote="prgs"
|
||||
)
|
||||
self.assertEqual(res["required_role_kind"], "reviewer")
|
||||
self.assertFalse(res["allowed_in_current_session"])
|
||||
self.assertTrue(res["stop_required"])
|
||||
|
||||
@patch("mcp_server.api_request", return_value={"login": "reviewer-user"})
|
||||
@patch("mcp_server.get_auth_header", return_value="token reviewer-pass")
|
||||
def test_resolve_pr_queue_cleanup_reviewer_profile_allowed(self, _auth, _api):
|
||||
with patch.dict(os.environ, self._env("reviewer-profile")):
|
||||
res = mcp_server.gitea_resolve_task_capability(
|
||||
task="pr_queue_cleanup", remote="prgs"
|
||||
)
|
||||
self.assertEqual(res["required_operation_permission"], "gitea.pr.review")
|
||||
self.assertTrue(res["allowed_in_current_session"])
|
||||
self.assertFalse(res["stop_required"])
|
||||
|
||||
@patch("mcp_server.api_request", return_value={"login": "author-user"})
|
||||
@patch("mcp_server.get_auth_header", return_value="token author-pass")
|
||||
def test_close_pr_known_and_lookalike_tasks_still_fail_closed(self, _auth, _api):
|
||||
|
||||
@@ -0,0 +1,121 @@
|
||||
"""Tests for enforced PR review/merge state machine (#290)."""
|
||||
import sys
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
|
||||
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
|
||||
|
||||
import review_merge_state_machine as rmsm # noqa: E402
|
||||
|
||||
FULL_REVIEW_COMPLETION = {state: True for state in rmsm.REVIEW_MERGE_STATES}
|
||||
|
||||
|
||||
def _completion_through(state_name: str) -> dict[str, bool]:
|
||||
idx = rmsm.state_index(state_name)
|
||||
return {state: i <= idx for i, state in enumerate(rmsm.REVIEW_MERGE_STATES)}
|
||||
|
||||
|
||||
class TestWorkflowBlockers(unittest.TestCase):
|
||||
def test_infra_stop_blocks_all_states(self):
|
||||
result = rmsm.assess_workflow_blockers(infra_stop=True)
|
||||
self.assertTrue(result["block"])
|
||||
self.assertEqual(result["forbidden_states"], list(rmsm.REVIEW_MERGE_STATES))
|
||||
|
||||
def test_infra_stop_forbids_advancement(self):
|
||||
result = rmsm.assess_state_advancement(
|
||||
{},
|
||||
target_state="INVENTORY",
|
||||
infra_stop=True,
|
||||
)
|
||||
self.assertTrue(result["block"])
|
||||
self.assertFalse(result["allowed"])
|
||||
|
||||
|
||||
class TestStateAdvancement(unittest.TestCase):
|
||||
def test_cannot_skip_inventory(self):
|
||||
result = rmsm.assess_state_advancement(
|
||||
{"PRECHECK": True},
|
||||
target_state="SELECT_PR",
|
||||
)
|
||||
self.assertTrue(result["block"])
|
||||
self.assertEqual(result["next_allowed_state"], "INVENTORY")
|
||||
|
||||
def test_sequential_advance_allowed(self):
|
||||
completion = _completion_through("INVENTORY")
|
||||
result = rmsm.assess_state_advancement(
|
||||
completion,
|
||||
target_state="SELECT_PR",
|
||||
)
|
||||
self.assertFalse(result["block"])
|
||||
self.assertTrue(result["allowed"])
|
||||
|
||||
|
||||
class TestApproveAndMergeGates(unittest.TestCase):
|
||||
def test_approve_blocked_without_validate(self):
|
||||
completion = _completion_through("PIN_HEAD_SHA")
|
||||
result = rmsm.can_approve(completion)
|
||||
self.assertTrue(result["block"])
|
||||
self.assertIn("VALIDATE", ",".join(result["missing_states"]))
|
||||
|
||||
def test_approve_allowed_when_review_path_complete(self):
|
||||
completion = _completion_through("REVIEW_DECISION")
|
||||
result = rmsm.can_approve(completion)
|
||||
self.assertFalse(result["block"])
|
||||
self.assertTrue(result["allowed"])
|
||||
|
||||
def test_merge_blocked_without_pre_merge_gates(self):
|
||||
completion = _completion_through("APPROVE_OR_REQUEST_CHANGES")
|
||||
result = rmsm.can_merge(completion)
|
||||
self.assertTrue(result["block"])
|
||||
self.assertTrue(result["missing_pre_merge_gates"])
|
||||
|
||||
def test_merge_allowed_with_pre_merge_gates(self):
|
||||
completion = _completion_through("PRE_MERGE_RECHECK")
|
||||
gates = {gate: True for gate in rmsm._PRE_MERGE_REQUIRED_GATES}
|
||||
result = rmsm.can_merge(completion, pre_merge_gates=gates)
|
||||
self.assertFalse(result["block"])
|
||||
self.assertTrue(result["allowed"])
|
||||
|
||||
def test_merge_blocked_when_head_sha_gate_fails(self):
|
||||
completion = _completion_through("PRE_MERGE_RECHECK")
|
||||
gates = {gate: True for gate in rmsm._PRE_MERGE_REQUIRED_GATES}
|
||||
gates["reviewed_head_sha_unchanged"] = False
|
||||
result = rmsm.can_merge(completion, pre_merge_gates=gates)
|
||||
self.assertTrue(result["block"])
|
||||
self.assertIn("reviewed_head_sha_unchanged", result["missing_pre_merge_gates"])
|
||||
|
||||
|
||||
class TestRecoveryHandoff(unittest.TestCase):
|
||||
def test_blocked_handoff_without_replay_passes(self):
|
||||
report = (
|
||||
"Blocked recovery handoff.\n"
|
||||
"Restart the full workflow after infra_stop clears.\n"
|
||||
"No approve/merge performed."
|
||||
)
|
||||
result = rmsm.assess_blocked_recovery_handoff(report)
|
||||
self.assertFalse(result["block"])
|
||||
|
||||
def test_blocked_handoff_with_merge_replay_fails(self):
|
||||
report = "Please merge PR #12 now using gitea_merge_pr."
|
||||
result = rmsm.assess_blocked_recovery_handoff(report)
|
||||
self.assertTrue(result["block"])
|
||||
|
||||
|
||||
class TestFinalReportClaims(unittest.TestCase):
|
||||
def test_ready_to_merge_claim_without_gates_fails(self):
|
||||
result = rmsm.assess_final_report_state_claims(
|
||||
"PR is ready to merge after validation.",
|
||||
state_completion=_completion_through("VALIDATE"),
|
||||
)
|
||||
self.assertTrue(result["block"])
|
||||
|
||||
def test_reviewed_claim_without_validate_fails(self):
|
||||
result = rmsm.assess_final_report_state_claims(
|
||||
"PR reviewed and looks good.",
|
||||
state_completion={"PRECHECK": True},
|
||||
)
|
||||
self.assertTrue(result["block"])
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,108 @@
|
||||
"""Tests for canonical review-workflow citation and version proof (#296).
|
||||
|
||||
PR review/merge reports must prove they loaded the canonical workflow
|
||||
(workflows/review-merge-pr.md) and cite the workflow version/hash used;
|
||||
a stale or missing hash fails so prompt drift is caught.
|
||||
"""
|
||||
import sys
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
|
||||
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
|
||||
|
||||
from review_proofs import ( # noqa: E402
|
||||
assess_review_workflow_source,
|
||||
compute_workflow_hash,
|
||||
)
|
||||
|
||||
|
||||
CANONICAL_TEXT = "# review-merge-pr workflow v1\nstep one\n"
|
||||
CANONICAL_HASH = compute_workflow_hash(CANONICAL_TEXT)
|
||||
|
||||
|
||||
def _report(citation=True, version=CANONICAL_HASH):
|
||||
lines = ["Task: review next eligible PR"]
|
||||
if citation:
|
||||
lines.append(
|
||||
"Workflow source: skills/llm-project-workflow/"
|
||||
"workflows/review-merge-pr.md (loaded via mcp_get_skill_guide)"
|
||||
)
|
||||
if version is not None:
|
||||
lines.append(f"- Workflow version: {version}")
|
||||
return "\n".join(lines) + "\n"
|
||||
|
||||
|
||||
class TestComputeWorkflowHash(unittest.TestCase):
|
||||
def test_deterministic(self):
|
||||
self.assertEqual(
|
||||
compute_workflow_hash(CANONICAL_TEXT),
|
||||
compute_workflow_hash(CANONICAL_TEXT),
|
||||
)
|
||||
|
||||
def test_changes_with_content(self):
|
||||
self.assertNotEqual(
|
||||
compute_workflow_hash(CANONICAL_TEXT),
|
||||
compute_workflow_hash(CANONICAL_TEXT + "extra rule\n"),
|
||||
)
|
||||
|
||||
def test_short_hex(self):
|
||||
h = compute_workflow_hash(CANONICAL_TEXT)
|
||||
self.assertEqual(len(h), 12)
|
||||
int(h, 16) # raises if not hex
|
||||
|
||||
|
||||
class TestReviewWorkflowSource(unittest.TestCase):
|
||||
def test_citation_and_matching_hash_passes(self):
|
||||
result = assess_review_workflow_source(
|
||||
_report(), canonical_hash=CANONICAL_HASH
|
||||
)
|
||||
self.assertTrue(result["complete"])
|
||||
self.assertEqual(result["reasons"], [])
|
||||
|
||||
def test_missing_citation_fails(self):
|
||||
result = assess_review_workflow_source(
|
||||
_report(citation=False), canonical_hash=CANONICAL_HASH
|
||||
)
|
||||
self.assertFalse(result["complete"])
|
||||
self.assertTrue(
|
||||
any("review-merge-pr" in r.lower() for r in result["reasons"])
|
||||
)
|
||||
|
||||
def test_missing_version_field_fails(self):
|
||||
result = assess_review_workflow_source(
|
||||
_report(version=None), canonical_hash=CANONICAL_HASH
|
||||
)
|
||||
self.assertFalse(result["complete"])
|
||||
self.assertTrue(
|
||||
any("workflow version" in r.lower() for r in result["reasons"])
|
||||
)
|
||||
|
||||
def test_stale_hash_fails(self):
|
||||
result = assess_review_workflow_source(
|
||||
_report(version="deadbeef0000"), canonical_hash=CANONICAL_HASH
|
||||
)
|
||||
self.assertFalse(result["complete"])
|
||||
self.assertTrue(
|
||||
any("stale" in r.lower() or "does not match" in r.lower()
|
||||
for r in result["reasons"])
|
||||
)
|
||||
|
||||
def test_version_none_fails(self):
|
||||
result = assess_review_workflow_source(
|
||||
_report(version="none"), canonical_hash=CANONICAL_HASH
|
||||
)
|
||||
self.assertFalse(result["complete"])
|
||||
|
||||
def test_without_canonical_hash_field_still_required(self):
|
||||
# No canonical hash supplied (e.g. offline validation): citation
|
||||
# and a populated version field are still required.
|
||||
self.assertTrue(
|
||||
assess_review_workflow_source(_report())["complete"]
|
||||
)
|
||||
self.assertFalse(
|
||||
assess_review_workflow_source(_report(version=None))["complete"]
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -0,0 +1,131 @@
|
||||
"""Tests for already-landed handoff verifier (#299)."""
|
||||
import sys
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
|
||||
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
|
||||
|
||||
from reviewer_already_landed_handoff import ( # noqa: E402
|
||||
ALREADY_LANDED_STATE,
|
||||
assess_already_landed_handoff_report,
|
||||
)
|
||||
|
||||
|
||||
def _good_report() -> str:
|
||||
return "\n".join([
|
||||
"## PR reconciliation summary",
|
||||
f"Eligibility class: {ALREADY_LANDED_STATE}",
|
||||
"Candidate head SHA: 2a544f582026b72a229d59a172c0a63ac4aaeaf9",
|
||||
"Reviewed head SHA: none",
|
||||
"Review worktree used: false",
|
||||
"Ancestor proof: PR head is ancestor of prgs/master",
|
||||
"",
|
||||
"## Controller Handoff",
|
||||
"- Selected PR: #278",
|
||||
f"- Eligibility class: {ALREADY_LANDED_STATE}",
|
||||
"- Candidate head SHA: 2a544f582026b72a229d59a172c0a63ac4aaeaf9",
|
||||
"- Reviewed head SHA: none",
|
||||
"- Target branch: master",
|
||||
"- Target branch SHA: abc123def4567890abcdef1234567890abcdef12",
|
||||
"- Review worktree used: false",
|
||||
"- Git ref mutations: git fetch prgs master",
|
||||
"- Review decision: none",
|
||||
"- Merge result: none",
|
||||
"- Linked issue status: open",
|
||||
"- Safe next action: reconcile open PR and linked issue",
|
||||
])
|
||||
|
||||
|
||||
class TestAlreadyLandedHandoff(unittest.TestCase):
|
||||
def test_clean_handoff_passes(self):
|
||||
result = assess_already_landed_handoff_report(_good_report())
|
||||
self.assertTrue(result["proven"], result["reasons"])
|
||||
self.assertTrue(result["gate_active"])
|
||||
|
||||
def test_inactive_gate_skips_checks(self):
|
||||
report = "## Controller Handoff\n- Selected PR: #12\n- Review decision: approve"
|
||||
result = assess_already_landed_handoff_report(report)
|
||||
self.assertTrue(result["proven"])
|
||||
self.assertFalse(result["gate_active"])
|
||||
|
||||
def test_rejects_stale_pinned_reviewed_head(self):
|
||||
report = _good_report().replace(
|
||||
"- Candidate head SHA:",
|
||||
"- Pinned reviewed head: 2a544f582026b72a229d59a172c0a63ac4aaeaf9\n"
|
||||
"- Candidate head SHA:",
|
||||
)
|
||||
result = assess_already_landed_handoff_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(any("pinned reviewed head" in r.lower() for r in result["reasons"]))
|
||||
|
||||
def test_rejects_scratch_worktree_used(self):
|
||||
report = _good_report() + "\n- Scratch worktree used: false"
|
||||
result = assess_already_landed_handoff_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(any("scratch worktree" in r.lower() for r in result["reasons"]))
|
||||
|
||||
def test_rejects_mutations_none_with_git_fetch(self):
|
||||
report = _good_report().replace(
|
||||
"- Git ref mutations: git fetch prgs master",
|
||||
"- Git ref mutations: none\n- Mutations: None",
|
||||
)
|
||||
result = assess_already_landed_handoff_report(
|
||||
report,
|
||||
command_log=[{"command": "git fetch prgs master", "performed": True}],
|
||||
)
|
||||
self.assertFalse(result["proven"])
|
||||
joined = " ".join(result["reasons"]).lower()
|
||||
self.assertIn("mutations", joined)
|
||||
self.assertIn("git ref", joined)
|
||||
|
||||
def test_rejects_workspace_mutations_none(self):
|
||||
report = _good_report() + "\n- Workspace mutations: None"
|
||||
result = assess_already_landed_handoff_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(any("workspace mutations" in r.lower() for r in result["reasons"]))
|
||||
|
||||
def test_rejects_reviewed_head_sha_when_gate_fired(self):
|
||||
report = _good_report().replace("- Reviewed head SHA: none", "- Reviewed head SHA: abc123")
|
||||
result = assess_already_landed_handoff_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(any("reviewed head sha" in r.lower() for r in result["reasons"]))
|
||||
|
||||
def test_rejects_narrative_handoff_eligibility_mismatch(self):
|
||||
report = _good_report().replace(
|
||||
f"Eligibility class: {ALREADY_LANDED_STATE}",
|
||||
"Eligibility class: NORMAL_REVIEW_CANDIDATE",
|
||||
1,
|
||||
)
|
||||
result = assess_already_landed_handoff_report(report)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(any("eligibility class" in r.lower() for r in result["reasons"]))
|
||||
|
||||
def test_session_gate_fired_without_text_marker(self):
|
||||
report = "## Controller Handoff\n- Selected PR: #1\n- Pinned reviewed head: abc"
|
||||
result = assess_already_landed_handoff_report(
|
||||
report,
|
||||
handoff_session={"gate_fired": True},
|
||||
)
|
||||
self.assertFalse(result["proven"])
|
||||
self.assertTrue(result["gate_active"])
|
||||
|
||||
def test_infra_stop_style_normal_handoff_not_blocked(self):
|
||||
report = "\n".join([
|
||||
"## Controller Handoff",
|
||||
"- Selected PR: #12",
|
||||
"- Review decision: request_changes",
|
||||
"- Pinned reviewed head: abc123def4567890abcdef1234567890abcdef12",
|
||||
])
|
||||
result = assess_already_landed_handoff_report(report)
|
||||
self.assertTrue(result["proven"])
|
||||
|
||||
|
||||
class TestExport(unittest.TestCase):
|
||||
def test_review_proofs_reexport(self):
|
||||
from review_proofs import assess_already_landed_handoff_report as exported
|
||||
|
||||
self.assertTrue(callable(exported))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
unittest.main()
|
||||
@@ -116,6 +116,24 @@ class TestRoleSessionRouter(unittest.TestCase):
|
||||
self.assertFalse(route["downstream_allowed"])
|
||||
self.assertIn("Wrong role/session for reviewer task", route["message"])
|
||||
|
||||
def test_pr_queue_cleanup_under_author_profile_wrong_role_stop(self):
|
||||
with patch.dict(os.environ, self._env("prgs-author")):
|
||||
route = mcp_server.gitea_route_task_session(
|
||||
task_type="pr_queue_cleanup", remote="prgs"
|
||||
)
|
||||
self.assertEqual(route["route_result"], role_session_router.ROUTE_WRONG_ROLE)
|
||||
self.assertFalse(route["downstream_allowed"])
|
||||
|
||||
@patch("mcp_server.api_request", return_value={"login": "sysadmin"})
|
||||
@patch("mcp_server.get_auth_header", return_value="token reviewer-pass")
|
||||
def test_pr_queue_cleanup_under_reviewer_profile_allowed(self, _auth, _api):
|
||||
with patch.dict(os.environ, self._env("prgs-reviewer")):
|
||||
route = mcp_server.gitea_route_task_session(
|
||||
task_type="pr_queue_cleanup", remote="prgs"
|
||||
)
|
||||
self.assertEqual(route["route_result"], role_session_router.ROUTE_ALLOWED)
|
||||
self.assertTrue(route["downstream_allowed"])
|
||||
|
||||
@patch("mcp_server.api_request")
|
||||
@patch("mcp_server.get_auth_header", return_value="token author-pass")
|
||||
def test_issue_creation_blocked_after_reviewer_wrong_role_stop(
|
||||
|
||||
Reference in New Issue
Block a user