Compare commits

..
Author SHA1 Message Date
jcwalker3 b4afc8cefd Merge branch 'master' into feat/issue-657-mcp-restart-path-inventory-guard 2026-07-24 02:07:25 -05:00
jcwalker3 d542b08ced Merge branch 'master' into feat/issue-657-mcp-restart-path-inventory-guard 2026-07-24 00:26:15 -05:00
sysadminandClaude Opus 4.8 3428fb4190 feat(mcp-health): inventory and guard MCP restart/reload/kill paths (#657)
Enumerate every code/script/host path that can restart, reload, reconnect,
kill, or force-recreate an MCP process, classify each, and link it to the
guard that constrains it.

- mcp_restart_paths.py: machine-readable registry (single source of truth)
  with classifications (sanctioned_narrow / guarded_fail_closed / forbidden /
  removed / host_residual) plus fail-closed guards:
  * assert_restart_attempt_registered() -- unknown restart attempts fail closed
  * assert_no_daemon_self_replacement() -- daemon never os.execv/os.kill/os._exit
    itself (source-tree scan; comment/docstring mentions ignored)
  * assert_auto_restart_helper_absent() -- keeps the #685-removed
    _trigger_mcp_auto_restart from returning
  * assert_registry_wellformed() -- every path classified, guarded, referenced
- docs/mcp-restart-path-inventory.md: complete inventory table linked from
  #655; documents residual host behaviors (/mcp reconnect) and rollout.
- tests/test_mcp_restart_paths.py: 17 tests -- registry well-formedness,
  unknown-attempt fail-closed, daemon-self-replacement scan (with injected
  violation + comment/docstring negative case), legacy-helper-removed
  regression, pkill-stays-contamination (#630), and doc/module lock-step.

No behavior change to existing modules; regression assertions codify invariants
that already hold (per #657 flag-free-before-hard-block rollout). Links
#652 #653 #655 #656.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
2026-07-24 00:57:33 -04:00
13 changed files with 891 additions and 1821 deletions
-976
View File
@@ -1,976 +0,0 @@
"""Sanctioned author issue worktree bootstrap for allocated issues (#850).
Bootstraps an allocated author branch, canonical worktree under ``branches/``,
worktree registration, issue lock, and lease/assignment binding without
caller-side Git, Bash, or helper scripts.
Features:
1. Durable phase journal with read-after-write evidence for every phase.
2. Idempotent replay handling via idempotency keys.
3. Authoritative expected-base / concurrency-pin validation.
4. Typed stale-pin refusal without silent rebasing or repointing.
5. Canonical branches-root enforcement (path MUST be inside branches/).
6. Preexisting work preservation (dirty tracked/untracked check, foreign ownership refusal).
7. Compensating recovery limited strictly to artifacts created by this transition.
8. Satisfiable exact_next_action for MCP scheduled workers.
"""
from __future__ import annotations
import json
import os
import shutil
import subprocess
from typing import Any, Mapping
import author_mutation_worktree
import issue_lock_store
import issue_lock_worktree
import lease_lifecycle
from reviewer_worktree import parse_dirty_tracked_files
import task_capability_map
BOOTSTRAP_TASKS = frozenset(
{
"bootstrap_author_issue_worktree",
"gitea_bootstrap_author_issue_worktree",
}
)
PHASE_1_REQUEST_ACCEPTED = "1_request_accepted"
PHASE_2_BRANCH_CONFIRMED = "2_branch_confirmed"
PHASE_3_PATH_RESERVED = "3_path_reserved"
PHASE_4_WORKTREE_CONFIRMED = "4_worktree_confirmed"
PHASE_5_REGISTRATION_VERIFIED = "5_registration_verified"
PHASE_6_STATE_ESTABLISHED = "6_state_established"
PHASE_7_TRANSITION_COMPLETED = "7_transition_completed"
PHASE_COMPENSATING_RECOVERY = "compensating_recovery"
JOURNAL_DIR_NAME = "bootstrap-journals"
def is_author_issue_bootstrap_task(task: str | None) -> bool:
"""True when *task* is the author issue worktree bootstrap task."""
return (task or "").strip() in BOOTSTRAP_TASKS
def get_journal_dir(override: str | None = None) -> str:
"""Return the root directory for durable bootstrap phase journals."""
if override:
path = override
elif os.environ.get("GITEA_BOOTSTRAP_JOURNAL_DIR"):
path = os.environ["GITEA_BOOTSTRAP_JOURNAL_DIR"]
else:
cache_dir = os.path.expanduser("~/.cache/gitea-tools")
path = os.path.join(cache_dir, JOURNAL_DIR_NAME)
os.makedirs(path, exist_ok=True)
return path
def _journal_file_path(idempotency_key: str, journal_dir: str | None = None) -> str:
safe_key = "".join(
c if c.isalnum() or c in ("-", "_", ".") else "_"
for c in idempotency_key
)
return os.path.join(get_journal_dir(journal_dir), f"{safe_key}.json")
def load_phase_journal(
idempotency_key: str, journal_dir: str | None = None
) -> dict[str, Any] | None:
"""Load a durable phase journal if it exists."""
path = _journal_file_path(idempotency_key, journal_dir=journal_dir)
if not os.path.isfile(path):
return None
try:
with open(path, "r", encoding="utf-8") as f:
return json.load(f)
except Exception:
return None
def save_phase_journal(
journal: dict[str, Any], journal_dir: str | None = None
) -> None:
"""Persist a durable phase journal with write-through file sync."""
key = journal["idempotency_key"]
path = _journal_file_path(key, journal_dir=journal_dir)
tmp_path = f"{path}.tmp.{os.getpid()}"
with open(tmp_path, "w", encoding="utf-8") as f:
json.dump(journal, f, indent=2, sort_keys=True)
os.replace(tmp_path, path)
def derive_default_idempotency_key(
remote: str,
org: str | None,
repo: str | None,
issue_number: int,
assignment_id: str | None = None,
lease_id: str | None = None,
) -> str:
parts = [
"bootstrap",
(remote or "prgs").strip(),
(org or "Scaled-Tech-Consulting").strip(),
(repo or "Gitea-Tools").strip(),
f"issue-{issue_number}",
]
if assignment_id:
parts.append(assignment_id.strip())
if lease_id:
parts.append(lease_id.strip())
return ":".join(parts)
def run_compensating_recovery(
journal: dict[str, Any],
canonical_repo_root: str,
journal_dir: str | None = None,
) -> dict[str, Any]:
"""Execute compensating recovery for artifacts created by this transition only."""
artifacts = journal.get("artifacts_created") or {}
pending = journal.get("pending_creations") or {}
rolled_back: list[str] = []
worktree_path = journal.get("worktree_path")
branch_name = journal.get("branch_name")
# Roll back issue lock if created
if artifacts.get("lock_created") or pending.get("lock"):
issue_num = journal.get("issue_number")
session_id = journal.get("owner_session")
if issue_num and session_id:
try:
issue_lock_store.release_session_lock(
issue_number=issue_num,
session=session_id,
lock_dir=journal_dir,
)
rolled_back.append(f"lock:issue-{issue_num}")
except Exception:
pass
artifacts["lock_created"] = False
worktree_created = (
artifacts.get("worktree_registered")
or artifacts.get("worktree_dir_created")
or (pending.get("worktree_path") == worktree_path and worktree_path)
)
if worktree_created and worktree_path:
if os.path.exists(worktree_path):
# Re-verify cleanliness before destructive removal (F-4)
porc_res = subprocess.run(
["git", "-C", worktree_path, "status", "--porcelain"],
capture_output=True,
text=True,
check=False,
)
is_dirty = porc_res.returncode == 0 and bool(porc_res.stdout.strip())
if is_dirty:
rolled_back.append(f"worktree_path_preserved_dirty:{worktree_path}")
else:
try:
subprocess.run(
[
"git",
"-C",
canonical_repo_root,
"worktree",
"remove",
"--force",
worktree_path,
],
capture_output=True,
text=True,
check=False,
)
except Exception:
pass
if os.path.exists(worktree_path):
shutil.rmtree(worktree_path, ignore_errors=True)
try:
subprocess.run(
["git", "-C", canonical_repo_root, "worktree", "prune"],
capture_output=True,
text=True,
check=False,
)
except Exception:
pass
rolled_back.append(f"worktree_path:{worktree_path}")
else:
rolled_back.append(f"worktree_path:{worktree_path}")
branch_created = (
artifacts.get("branch_created")
or (pending.get("branch_name") == branch_name and branch_name)
)
if branch_created and branch_name:
try:
res = subprocess.run(
[
"git",
"-C",
canonical_repo_root,
"rev-parse",
"--verify",
branch_name,
],
capture_output=True,
text=True,
check=False,
)
if res.returncode == 0:
# Check for author commits on branch before branch deletion (F-4)
resolved_base = journal.get("resolved_base_sha") or "master"
rev_list_res = subprocess.run(
[
"git",
"-C",
canonical_repo_root,
"rev-list",
f"{resolved_base}..{branch_name}",
],
capture_output=True,
text=True,
check=False,
)
has_commits = rev_list_res.returncode == 0 and bool(rev_list_res.stdout.strip())
if has_commits:
rolled_back.append(f"branch_preserved_commits:{branch_name}")
else:
subprocess.run(
[
"git",
"-C",
canonical_repo_root,
"branch",
"-D",
branch_name,
],
capture_output=True,
text=True,
check=False,
)
rolled_back.append(f"branch:{branch_name}")
except Exception:
pass
recovery_info = {
"executed": True,
"rolled_back": rolled_back,
"reason": journal.get("failure_reason"),
}
journal["compensating_recovery"] = recovery_info
journal["current_phase"] = PHASE_COMPENSATING_RECOVERY
save_phase_journal(journal, journal_dir=journal_dir)
return recovery_info
def assess_author_issue_bootstrap(
*,
workspace_path: str,
canonical_repo_root: str,
current_branch: str | None = None,
head_sha: str | None = None,
porcelain_status: str = "",
remote_master_sha: str | None = None,
remote_master_sha_error: str | None = None,
task: str | None = None,
) -> dict[str, Any]:
"""Assess whether author issue worktree bootstrap may proceed from control or worktree root."""
root = os.path.realpath(canonical_repo_root or "")
workspace = os.path.realpath(workspace_path or root or ".")
branch = (current_branch or "").strip()
dirty = parse_dirty_tracked_files(porcelain_status or "")
under_branches = (
author_mutation_worktree.is_path_under_branches(workspace, root)
if root
else False
)
if not is_author_issue_bootstrap_task(task):
return {
"not_applicable": True,
"allowed": False,
"block": False,
"proven": False,
"reasons": ["task is not author_issue_bootstrap"],
}
if under_branches:
return {
"not_applicable": False,
"allowed": True,
"block": False,
"proven": True,
"bootstrap_path": "existing_branches_worktree",
"reasons": [
"workspace is already a registered worktree under branches/"
],
}
reasons: list[str] = []
if workspace != root:
reasons.append(
"bootstrap requires workspace to be canonical control checkout or branches/ worktree"
)
if branch not in author_mutation_worktree.BASE_BRANCHES:
reasons.append(
f"control checkout branch '{branch}' is not an accepted base branch "
f"({', '.join(sorted(author_mutation_worktree.BASE_BRANCHES))})"
)
if dirty:
reasons.append(
f"control checkout has tracked local edits: {', '.join(dirty[:5])}"
)
if remote_master_sha_error:
reasons.append(
f"could not verify live master tip: {remote_master_sha_error}"
)
elif remote_master_sha and head_sha:
h = head_sha.strip().lower()
rm = remote_master_sha.strip().lower()
if h != rm:
reasons.append(
f"control checkout HEAD ({h[:12]}) != live master tip ({rm[:12]})"
)
if reasons:
return {
"not_applicable": False,
"allowed": False,
"block": True,
"proven": False,
"reasons": reasons,
}
return {
"not_applicable": False,
"allowed": True,
"block": False,
"proven": True,
"bootstrap_path": "clean_canonical_control_checkout",
"reasons": [
"control checkout is clean on accepted base branch matching live master"
],
}
import fcntl
import stat
class BootstrapTransitionLock:
"""Inter-process file lock scoped to the transition identity / idempotency key."""
def __init__(self, idempotency_key: str, journal_dir: str | None = None):
safe_key = "".join(
c if c.isalnum() or c in ("-", "_", ".") else "_"
for c in idempotency_key
)
lock_dir = os.path.realpath(get_journal_dir(journal_dir))
if not os.path.isdir(lock_dir):
raise RuntimeError(f"Lock directory '{lock_dir}' does not exist or is not a directory")
raw_lock_path = os.path.abspath(os.path.join(lock_dir, f"{safe_key}.lock"))
try:
common = os.path.commonpath([lock_dir, os.path.dirname(raw_lock_path)])
except Exception:
common = None
if common != lock_dir:
raise RuntimeError(f"Lock path '{raw_lock_path}' escapes canonical lock directory '{lock_dir}'")
self.lock_path = raw_lock_path
self.fd = None
def __enter__(self):
if os.path.islink(self.lock_path):
raise RuntimeError(f"Refusing lock acquisition: lock path '{self.lock_path}' is a symlink")
flags = os.O_RDWR | os.O_CREAT
if hasattr(os, "O_NOFOLLOW"):
flags |= os.O_NOFOLLOW
if hasattr(os, "O_CLOEXEC"):
flags |= os.O_CLOEXEC
try:
fd = os.open(self.lock_path, flags, 0o600)
except OSError as exc:
raise RuntimeError(f"Failed to open lock file safely '{self.lock_path}': {exc}") from exc
st = os.fstat(fd)
if not stat.S_ISREG(st.st_mode):
os.close(fd)
raise RuntimeError(f"Lock target '{self.lock_path}' is not a regular file")
self.fd = fd
fcntl.flock(self.fd, fcntl.LOCK_EX)
return self
def __exit__(self, exc_type, exc_val, exc_tb):
if self.fd is not None:
try:
fcntl.flock(self.fd, fcntl.LOCK_UN)
except Exception:
pass
try:
os.close(self.fd)
except Exception:
pass
self.fd = None
def bootstrap_author_issue_worktree(
*,
issue_number: int,
canonical_repo_root: str,
assignment_id: str | None = None,
lease_id: str | None = None,
expected_base_sha: str | None = None,
branch_name: str | None = None,
worktree_path: str | None = None,
idempotency_key: str | None = None,
remote: str = "prgs",
host: str | None = None,
org: str | None = "Scaled-Tech-Consulting",
repo: str | None = "Gitea-Tools",
active_identity: str | None = "jcwalker3",
active_profile: str | None = "prgs-author",
owner_session: str | None = None,
lock_dir: str | None = None,
dry_run: bool = False,
) -> dict[str, Any]:
"""Execute the sanctioned author issue worktree bootstrap transition."""
root = os.path.realpath(canonical_repo_root)
session = (owner_session or "").strip()
if not session:
return {
"success": False,
"reason_code": "missing_owner_session",
"message": "Missing required owner_session parameter (fail closed). Session identifier cannot be fabricated or defaulted.",
"exact_next_action": (
"Pass explicit owner_session resolved from gitea_whoami or session context."
),
}
if active_identity is None or not str(active_identity).strip():
return {
"success": False,
"reason_code": "missing_active_identity",
"message": "Missing required active_identity parameter (fail closed). Identity cannot be fabricated or defaulted.",
"exact_next_action": "Pass explicit active_identity resolved from gitea_whoami.",
}
identity = str(active_identity).strip()
if active_profile is None or not str(active_profile).strip():
return {
"success": False,
"reason_code": "missing_active_profile",
"message": "Missing required active_profile parameter (fail closed). Profile cannot be fabricated or defaulted.",
"exact_next_action": "Pass explicit active_profile resolved from gitea_whoami.",
}
profile = str(active_profile).strip()
# Derive standard inputs
expected_pattern = f"issue-{issue_number}"
target_branch = (branch_name or "").strip()
if not target_branch:
target_branch = f"fix/issue-{issue_number}-native-mcp-bootstrap"
elif expected_pattern not in target_branch:
return {
"success": False,
"reason_code": "invalid_branch_name",
"message": (
f"Branch name '{target_branch}' must contain issue pattern '{expected_pattern}'"
),
"exact_next_action": (
f"Supply a branch_name containing '{expected_pattern}', e.g., 'fix/issue-{issue_number}-...'"
),
}
worktree_name = target_branch.replace("/", "-")
target_worktree = (worktree_path or "").strip()
if not target_worktree:
target_worktree = os.path.join(root, "branches", worktree_name)
target_worktree = os.path.realpath(os.path.abspath(target_worktree))
key = (idempotency_key or "").strip()
if not key:
key = derive_default_idempotency_key(
remote=remote,
org=org,
repo=repo,
issue_number=issue_number,
assignment_id=assignment_id,
lease_id=lease_id,
)
# Acquire cross-process file lock scoped to the idempotency key / transition identity
with BootstrapTransitionLock(key, journal_dir=lock_dir):
# Idempotency check
existing = load_phase_journal(key, journal_dir=lock_dir)
if existing and existing.get("completed"):
if (
existing.get("issue_number") == issue_number
and existing.get("branch_name") == target_branch
and os.path.realpath(existing.get("worktree_path", ""))
== target_worktree
):
return {
"success": True,
"replayed": True,
"message": (
f"Idempotent replay: worktree for issue #{issue_number} already bootstrapped at {target_worktree}"
),
"issue_number": issue_number,
"branch_name": target_branch,
"worktree_path": target_worktree,
"base_sha": existing.get("resolved_base_sha"),
"lease_id": existing.get("lease_id"),
"assignment_id": existing.get("assignment_id"),
"idempotency_key": key,
"phase_journal": existing,
"exact_next_action": (
"Call gitea_whoami, then gitea_resolve_task_capability(task='work_issue') "
"and proceed with author implementation in the bootstrapped worktree."
),
}
else:
return {
"success": False,
"reason_code": "incompatible_idempotency_replay",
"message": (
f"Idempotency key '{key}' already exists with incompatible parameters "
f"(stored: {existing.get('branch_name')}, {existing.get('worktree_path')}; "
f"requested: {target_branch}, {target_worktree})"
),
"exact_next_action": (
"Supply a unique idempotency_key or pass compatible parameters."
),
}
# Initialize or resume Phase Journal
if existing:
journal = existing
artifacts = journal.setdefault("artifacts_created", {})
artifacts.setdefault("branch_created", False)
artifacts.setdefault("worktree_dir_created", False)
artifacts.setdefault("worktree_registered", False)
artifacts.setdefault("lock_created", False)
else:
journal = {
"idempotency_key": key,
"issue_number": issue_number,
"assignment_id": assignment_id,
"lease_id": lease_id,
"expected_base_sha": expected_base_sha,
"resolved_base_sha": None,
"branch_name": target_branch,
"worktree_path": target_worktree,
"active_identity": identity,
"active_profile": profile,
"owner_session": session,
"remote": remote,
"org": org,
"repo": repo,
"phases": {},
"artifacts_created": {
"branch_created": False,
"worktree_dir_created": False,
"worktree_registered": False,
"lock_created": False,
},
"current_phase": PHASE_1_REQUEST_ACCEPTED,
"completed": False,
}
# Fetch current live master SHA
try:
rev_res = subprocess.run(
["git", "-C", root, "rev-parse", "HEAD"],
capture_output=True,
text=True,
check=True,
)
live_master_sha = rev_res.stdout.strip()
except Exception as exc:
return {
"success": False,
"reason_code": "git_rev_parse_failed",
"message": f"Could not determine repository HEAD: {exc}",
"exact_next_action": "Verify repository git state and retry.",
}
# Phase 1: REQUEST_ACCEPTED & Concurrency Pin Check
if expected_base_sha:
exp_norm = expected_base_sha.strip().lower()
live_norm = live_master_sha.lower()
if exp_norm != live_norm:
journal["failure_reason"] = (
f"stale concurrency pin: expected {exp_norm[:12]} != live {live_norm[:12]}"
)
save_phase_journal(journal, journal_dir=lock_dir)
return {
"success": False,
"reason_code": "stale_concurrency_pin",
"message": (
f"Expected base SHA {exp_norm[:12]} does not match live master SHA {live_norm[:12]} (fail closed)."
),
"expected_base_sha": expected_base_sha,
"live_master_sha": live_master_sha,
"exact_next_action": (
"Re-evaluate assignment against current live master SHA and retry with updated expected_base_sha."
),
}
journal["resolved_base_sha"] = live_master_sha
journal["phases"][PHASE_1_REQUEST_ACCEPTED] = {
"status": "completed",
"live_master_sha": live_master_sha,
"expected_base_sha": expected_base_sha,
}
journal["current_phase"] = PHASE_2_BRANCH_CONFIRMED
save_phase_journal(journal, journal_dir=lock_dir)
if dry_run:
return {
"success": True,
"dry_run": True,
"message": f"Dry-run: validated bootstrap intent for issue #{issue_number}",
"issue_number": issue_number,
"branch_name": target_branch,
"worktree_path": target_worktree,
"base_sha": live_master_sha,
"phase_journal": journal,
"exact_next_action": "Run without dry_run=True to execute bootstrap.",
}
# Phase 2: BRANCH_CONFIRMED
was_branch_created_previously = journal["artifacts_created"].get("branch_created", False)
pending_branch = (journal.get("pending_creations") or {}).get("branch_name")
branch_check = subprocess.run(
["git", "-C", root, "rev-parse", "--verify", target_branch],
capture_output=True,
text=True,
check=False,
)
if branch_check.returncode == 0:
branch_head = branch_check.stdout.strip()
# Verify branch head descends from base
anc_check = subprocess.run(
[
"git",
"-C",
root,
"merge-base",
"--is-ancestor",
live_master_sha,
branch_head,
],
capture_output=True,
text=True,
check=False,
)
if anc_check.returncode != 0 and branch_head.lower() != live_master_sha.lower():
# F-8: Check if branch shares a common ancestor with live master
mb_check = subprocess.run(
["git", "-C", root, "merge-base", live_master_sha, branch_head],
capture_output=True,
text=True,
check=False,
)
if mb_check.returncode != 0 or not mb_check.stdout.strip():
journal["failure_reason"] = (
f"existing branch '{target_branch}' HEAD ({branch_head[:12]}) is incompatible with live master ({live_master_sha[:12]})"
)
save_phase_journal(journal, journal_dir=lock_dir)
return {
"success": False,
"reason_code": "incompatible_existing_branch",
"message": (
f"Existing branch '{target_branch}' HEAD ({branch_head[:12]}) is incompatible with live master ({live_master_sha[:12]})."
),
"exact_next_action": (
"Inspect or sync the existing branch with master before bootstrapping."
),
}
# Preserve creation provenance monotonically across interruption and replay
if was_branch_created_previously or pending_branch == target_branch:
journal["artifacts_created"]["branch_created"] = True
else:
journal["artifacts_created"]["branch_created"] = False
else:
# Persist creation intent/provenance to disk BEFORE executing external mutation
journal.setdefault("pending_creations", {})["branch_name"] = target_branch
journal["artifacts_created"]["branch_created"] = True
save_phase_journal(journal, journal_dir=lock_dir)
# Create branch
create_res = subprocess.run(
["git", "-C", root, "branch", target_branch, live_master_sha],
capture_output=True,
text=True,
check=False,
)
if create_res.returncode != 0:
journal["artifacts_created"]["branch_created"] = False
journal.get("pending_creations", {}).pop("branch_name", None)
journal["failure_reason"] = (
f"failed to create git branch '{target_branch}': {create_res.stderr.strip()}"
)
save_phase_journal(journal, journal_dir=lock_dir)
return {
"success": False,
"reason_code": "branch_creation_failed",
"message": f"Failed to create git branch '{target_branch}': {create_res.stderr.strip()}",
"exact_next_action": "Verify branch availability and retry.",
}
journal["phases"][PHASE_2_BRANCH_CONFIRMED] = {
"status": "completed",
"branch_name": target_branch,
"created": journal["artifacts_created"]["branch_created"],
}
journal["current_phase"] = PHASE_3_PATH_RESERVED
save_phase_journal(journal, journal_dir=lock_dir)
# Phase 3: PATH_RESERVED & Phase 4: WORKTREE_CONFIRMED
if not author_mutation_worktree.is_path_under_branches(
target_worktree, root
):
journal["failure_reason"] = (
f"target_worktree '{target_worktree}' is outside canonical branches/ root"
)
run_compensating_recovery(journal, root, journal_dir=lock_dir)
return {
"success": False,
"reason_code": "path_outside_canonical_branches_root",
"message": (
f"Worktree path '{target_worktree}' is outside canonical branches/ root (fail closed)."
),
"exact_next_action": (
"Provide a worktree_path inside canonical branches/ root, e.g., 'branches/issue-...'"
),
}
was_dir_created_previously = journal["artifacts_created"].get("worktree_dir_created", False)
was_registered_previously = journal["artifacts_created"].get("worktree_registered", False)
pending_wt = (journal.get("pending_creations") or {}).get("worktree_path")
dir_exists = os.path.exists(target_worktree)
if dir_exists:
# Check porcelain directly
porc_res = subprocess.run(
["git", "-C", target_worktree, "status", "--porcelain"],
capture_output=True,
text=True,
check=False,
)
dirty = (
parse_dirty_tracked_files(porc_res.stdout)
if porc_res.returncode == 0
else []
)
if dirty or (porc_res.returncode == 0 and porc_res.stdout.strip()):
journal["failure_reason"] = (
f"target worktree '{target_worktree}' contains dirty tracked/untracked files"
)
run_compensating_recovery(journal, root, journal_dir=lock_dir)
return {
"success": False,
"reason_code": "preexisting_dirty_worktree",
"message": (
f"Preexisting worktree '{target_worktree}' has dirty tracked/untracked files (fail closed)."
),
"exact_next_action": (
"Clean or stash the pre-existing worktree files before bootstrapping."
),
}
# Check registered branch
wt_state = issue_lock_worktree.read_worktree_git_state(target_worktree)
wt_branch = (wt_state.get("current_branch") or "").strip()
if wt_branch and wt_branch != target_branch:
journal["failure_reason"] = (
f"existing worktree '{target_worktree}' is on branch '{wt_branch}' != expected '{target_branch}'"
)
run_compensating_recovery(journal, root, journal_dir=lock_dir)
return {
"success": False,
"reason_code": "incompatible_existing_directory",
"message": (
f"Existing worktree '{target_worktree}' is registered to branch '{wt_branch}' instead of '{target_branch}'."
),
"exact_next_action": (
"Inspect or remove the pre-existing worktree folder before bootstrapping."
),
}
if was_dir_created_previously or pending_wt == target_worktree:
journal["artifacts_created"]["worktree_dir_created"] = True
journal["artifacts_created"]["worktree_registered"] = True
else:
journal["artifacts_created"]["worktree_dir_created"] = False
journal["artifacts_created"]["worktree_registered"] = False
else:
# Persist creation intent/provenance to disk BEFORE executing external worktree add mutation
journal.setdefault("pending_creations", {})["worktree_path"] = target_worktree
journal["artifacts_created"]["worktree_dir_created"] = True
journal["artifacts_created"]["worktree_registered"] = True
save_phase_journal(journal, journal_dir=lock_dir)
wt_add_res = subprocess.run(
[
"git",
"-C",
root,
"worktree",
"add",
target_worktree,
target_branch,
],
capture_output=True,
text=True,
check=False,
)
if wt_add_res.returncode != 0:
journal["artifacts_created"]["worktree_dir_created"] = False
journal["artifacts_created"]["worktree_registered"] = False
journal.get("pending_creations", {}).pop("worktree_path", None)
journal["failure_reason"] = (
f"git worktree add failed: {wt_add_res.stderr.strip()}"
)
run_compensating_recovery(journal, root, journal_dir=lock_dir)
return {
"success": False,
"reason_code": "worktree_add_failed",
"message": f"Failed to execute git worktree add: {wt_add_res.stderr.strip()}",
"exact_next_action": "Verify git worktree capabilities and retry.",
}
journal["phases"][PHASE_3_PATH_RESERVED] = {
"status": "completed",
"worktree_path": target_worktree,
"preexisting_dir": dir_exists,
}
journal["current_phase"] = PHASE_4_WORKTREE_CONFIRMED
save_phase_journal(journal, journal_dir=lock_dir)
# Phase 5: REGISTRATION_VERIFIED
wt_list_res = subprocess.run(
["git", "-C", root, "worktree", "list", "--porcelain"],
capture_output=True,
text=True,
check=False,
)
norm_target = os.path.realpath(target_worktree)
found_registration = False
if wt_list_res.returncode == 0:
for block in wt_list_res.stdout.split("\n\n"):
lines = block.strip().splitlines()
worktree_line = next(
(l[9:].strip() for l in lines if l.startswith("worktree ")),
None,
)
if worktree_line and os.path.realpath(worktree_line) == norm_target:
found_registration = True
break
if not found_registration:
journal["failure_reason"] = (
f"worktree registration for '{target_worktree}' not found in git worktree list"
)
run_compensating_recovery(journal, root, journal_dir=lock_dir)
return {
"success": False,
"reason_code": "worktree_registration_verification_failed",
"message": f"Worktree '{target_worktree}' registration verification failed.",
"exact_next_action": "Check git worktree list integrity and retry.",
}
journal["phases"][PHASE_4_WORKTREE_CONFIRMED] = {
"status": "completed",
"worktree_path": target_worktree,
}
journal["phases"][PHASE_5_REGISTRATION_VERIFIED] = {
"status": "completed",
"registered": True,
}
journal["current_phase"] = PHASE_6_STATE_ESTABLISHED
save_phase_journal(journal, journal_dir=lock_dir)
# Phase 6: STATE_ESTABLISHED — Issue Lock Acquisition
from datetime import datetime, timezone
try:
lock_data = {
"remote": remote,
"org": org or "Scaled-Tech-Consulting",
"repo": repo or "Gitea-Tools",
"issue_number": issue_number,
"branch": target_branch,
"branch_name": target_branch,
"worktree_path": target_worktree,
"owner_session": session,
"claimant": {
"username": identity,
"profile": profile,
},
"assignment_id": assignment_id,
"lease_id": lease_id,
"expected_base_sha": live_master_sha,
"created_at": datetime.now(timezone.utc).isoformat(),
}
journal.setdefault("pending_creations", {})["lock"] = True
journal["artifacts_created"]["lock_created"] = True
save_phase_journal(journal, journal_dir=lock_dir)
lock_res = issue_lock_store.bind_session_lock(lock_data, lock_dir=lock_dir)
except Exception as exc:
journal["artifacts_created"]["lock_created"] = False
journal.get("pending_creations", {}).pop("lock", None)
journal["failure_reason"] = f"issue lock binding failed: {exc}"
run_compensating_recovery(journal, root, journal_dir=lock_dir)
return {
"success": False,
"reason_code": "issue_lock_acquisition_failed",
"message": f"Could not bind canonical issue lock for issue #{issue_number}: {exc}",
"exact_next_action": "Verify lease/assignment state and retry.",
}
journal["phases"][PHASE_6_STATE_ESTABLISHED] = {
"status": "completed",
"lock": lock_res,
}
journal["phases"][PHASE_7_TRANSITION_COMPLETED] = {
"status": "completed",
}
journal["current_phase"] = PHASE_7_TRANSITION_COMPLETED
journal["completed"] = True
save_phase_journal(journal, journal_dir=lock_dir)
return {
"success": True,
"replayed": False,
"message": (
f"Successfully bootstrapped author issue worktree for issue #{issue_number} "
f"at branch '{target_branch}' and worktree '{target_worktree}'."
),
"issue_number": issue_number,
"branch_name": target_branch,
"worktree_path": target_worktree,
"base_sha": live_master_sha,
"lease_id": lease_id,
"assignment_id": assignment_id,
"idempotency_key": key,
"lock_state": lock_res,
"phase_journal": journal,
"exact_next_action": (
"Call gitea_whoami, then gitea_resolve_task_capability(task='work_issue') "
"and proceed with author implementation in the bootstrapped worktree."
),
}
+37 -58
View File
@@ -40,66 +40,22 @@ def _normalize_path(path: str) -> str:
return (path or "").replace("\\", "/").rstrip("/")
def get_canonical_branches_root(project_root: str | None = None) -> str:
"""Return the absolute path of the canonical branches directory for *project_root*."""
root = os.path.realpath(project_root) if project_root else os.path.realpath(os.getcwd())
canonical_repo_root = resolve_canonical_repo_root(root, root)
return os.path.realpath(os.path.join(canonical_repo_root, "branches"))
def is_path_under_branches(path: str, project_root: str | None = None) -> bool:
"""True when *path* resolves inside a canonical ``branches/`` directory."""
if not path or not str(path).strip():
"""True when *path* resolves inside ``<project_root>/branches/``."""
normalized = _normalize_path(path)
if not normalized:
return False
try:
real_path = os.path.realpath(os.path.abspath(str(path).strip()))
except Exception:
return False
branches_root = get_canonical_branches_root(project_root or real_path)
try:
common = os.path.commonpath([branches_root, real_path])
except Exception:
return False
if common != branches_root:
return False
rel = os.path.relpath(real_path, branches_root)
return rel != "." and not rel.startswith("..")
def resolve_canonical_repo_root(workspace_path: str, fallback_project_root: str) -> str:
"""Return the stable repository root for *workspace_path* via git metadata (#460)."""
p = (workspace_path or "").strip()
if p:
try:
res = subprocess.run(
["git", "-C", p, "rev-parse", "--git-common-dir"],
capture_output=True,
text=True,
check=True,
)
common = _realpath_git_common_dir(p, res.stdout)
if common.endswith(f"{os.sep}.git") or os.path.basename(common) == ".git":
candidate_root = os.path.dirname(common)
real_p = os.path.realpath(p)
try:
if os.path.commonpath([candidate_root, real_p]) == candidate_root:
return candidate_root
except Exception:
pass
except Exception:
pass
fallback = os.path.realpath(fallback_project_root or workspace_path or ".")
norm = fallback.replace("\\", "/")
if "/branches/" in norm:
return os.path.realpath(norm.split("/branches/")[0])
elif norm.endswith("/branches"):
return os.path.realpath(os.path.dirname(fallback))
return fallback
if "/branches/" in f"{normalized}/":
return True
if normalized.endswith("/branches"):
return True
if project_root:
root = _normalize_path(os.path.realpath(project_root))
real = _normalize_path(os.path.realpath(path))
if real.startswith(f"{root}/"):
rel = real[len(root) + 1 :]
return rel == "branches" or rel.startswith("branches/")
return False
def resolve_mutation_workspace(
@@ -131,6 +87,29 @@ def _realpath_git_common_dir(workspace_path: str, common_dir: str) -> str:
return os.path.realpath(os.path.join(workspace_path, raw))
def resolve_canonical_repo_root(workspace_path: str, fallback_project_root: str) -> str:
"""Return the stable repository root for *workspace_path* via git metadata (#460)."""
path = (workspace_path or "").strip()
fallback = os.path.realpath(fallback_project_root)
if not path:
return fallback
try:
res = subprocess.run(
["git", "-C", path, "rev-parse", "--git-common-dir"],
capture_output=True,
text=True,
check=True,
)
common = _realpath_git_common_dir(path, res.stdout)
except Exception:
return fallback
if common.endswith(f"{os.sep}.git"):
return os.path.dirname(common)
if os.path.basename(common) == ".git":
return os.path.dirname(common)
return fallback
def resolve_author_mutation_context(
worktree_path: str | None,
process_project_root: str,
+1 -2
View File
@@ -247,8 +247,7 @@ def bootstrap_permits_control_checkout(
"""
if not isinstance(assessment, dict):
return False
import author_issue_bootstrap
if not is_create_issue_task(task) and not author_issue_bootstrap.is_author_issue_bootstrap_task(task):
if not is_create_issue_task(task):
return False
# Positive proof: the assessment must affirmatively allow, with no
+90
View File
@@ -0,0 +1,90 @@
# MCP restart / reload / kill path inventory (#657)
Complete inventory of every code, script, and host path that can **restart,
reload, reconnect, kill, or force-recreate** an MCP process in this project,
with each path classified and linked to the guard that constrains it.
This document is the human-readable companion to the machine-readable registry
in [`mcp_restart_paths.py`](../mcp_restart_paths.py). The two are kept in
lock-step by [`tests/test_mcp_restart_paths.py`](../tests/test_mcp_restart_paths.py):
every `path_id` below must appear in this file, and the source guards are run
against the live tree.
Roadmap linkage: this inventory is the enumeration step of the restart
governance work — parent **#655**, restart-governance ADR **#656**, vision
**#652**, roadmap **#653**. Related detection/guard work: master-advance
staleness **#591**/**#420**, side-effect-free resolver **#685**, transport flap
**#584**, manual-kill contamination **#630**.
## Classifications
| Classification | Meaning |
|---|---|
| `sanctioned_narrow_recovery` | One-shot, safe-by-construction recovery that never targets the running daemon. |
| `guarded_fail_closed` | Detects a restart-requiring condition, then fails mutations closed and emits reconnect guidance. Never self-restarts. |
| `forbidden` | A workflow-safety violation; where an LLM tool could invoke it, it is marked contamination. |
| `removed` | A previously-existing unguarded restart primitive that has been deleted; a regression guard keeps it absent. |
| `host_residual` | Behavior owned by the host/IDE, outside this process's control. Documented, not code-guarded here. |
## The rule
**No component may perform an unguarded full restart of the MCP daemon.** The
in-process daemon (`gitea_mcp_server.py`, `mcp_server.py`,
`role_session_router.py`) must never replace or terminate its own process:
replacing the process after the host has wired up the stdio pipes desyncs the
JSON-RPC transport (observed with Antigravity/Cascade hosts). Recovery is owned
by the host/operator via a client reconnect — the daemon only ever *detects*
and *fails closed*.
## Inventory
| path_id | Classification | Mechanism | Guard | Refs |
|---|---|---|---|---|
| `cli_venv_bootstrap_execv` | sanctioned_narrow_recovery | CLI wrapper scripts re-exec into `venv/bin/python3` via `os.execv`, guarded by `sys.executable != venv_python`. | One-shot pre-import bootstrap; runs before any MCP transport exists and only when not already on the venv interpreter; idempotent guard prevents a re-exec loop. | #657 |
| `daemon_self_replacement` | forbidden | The daemon replacing/terminating its own process (`os.execv`/`os.kill`/`os._exit`) to reload code. | Forbidden by design; enforced against the source tree by `assert_no_daemon_self_replacement()`. | #657, #584 |
| `legacy_auto_restart_helper` | removed | A helper (`_trigger_mcp_auto_restart`) that actively restarted the server from the read-only resolver path. | Removed in #685; kept absent by `assert_auto_restart_helper_absent()`. | #685, #657 |
| `config_touch_reload` | removed | Touching (utime) the MCP client config to make the host reload the server. | Removed from the resolver in #685: stale detection is report-only, never mutating config, spawning threads, or calling `os._exit`. | #685, #657 |
| `master_advance_auto_restart` | guarded_fail_closed | On-disk master advancing past the running code. | `master_parity_gate` captures startup parity and blocks mutations while stale, emitting restart guidance; the process never self-restarts. | #420, #591, #657 |
| `stale_runtime_resolver_reconnect` | guarded_fail_closed | The capability resolver detecting a stale serving process. | Report-only (#685): returns `restart_required`/`stop_required` and an exact reconnect action; no restart, thread, config touch, or `os._exit`. | #685, #657 |
| `manual_daemon_kill` | forbidden | Shell kills of the daemon: `pkill -f mcp_server.py`, `killall`, broad `pkill -f python` sweeps, or `kill <pid>` of a daemon pid. | Forbidden (#630): `runtime_recovery_guard` classifies these as contamination and `gitea_record_daemon_process_kill_attempt` writes a durable marker that fails later mutations closed. Operator maintenance authorization is read only from the environment. | #630, #657 |
| `conflict_marker_infra_stop` | guarded_fail_closed | The daemon entrypoint scans for unresolved merge-conflict markers at startup and stops (`sys.exit(1)`). | Fail-closed startup stop, not a restart: the process exits and waits for the operator to resolve conflicts and relaunch; never loops. | #657 |
| `ide_client_reconnect` | host_residual | A manual `/mcp reconnect` (or equivalent host action) that recreates the MCP client connection. | Outside this process's control; the sanctioned recovery the gates point operators toward. No in-process code initiates it. | #584, #656, #657 |
| `profile_switch_runtime` | sanctioned_narrow_recovery | Switching the active execution profile at runtime (dynamic-profile mode). | In-process and restart-free: `runtime_switching_supported` is true, so a switch rebinds capability without recreating the process. | #656, #657 |
## Guards enforced in CI
`tests/test_mcp_restart_paths.py` asserts, against the live source tree:
1. **Registry well-formedness** — every path has a valid classification, a
non-empty guard description, references, and locations; ids are unique; all
five classifications are represented.
2. **Unknown restart attempts fail closed**
`assert_restart_attempt_registered()` raises `UnknownRestartPathError` for
any path id not in this inventory, so a novel/unnamed restart primitive
cannot slip through silently.
3. **Daemon never self-replaces**`assert_no_daemon_self_replacement()` scans
the daemon modules for `os.execv`/`os.kill`/`os._exit`/`os.abort` calls
(comment/docstring mentions are ignored) and finds none.
4. **Legacy helper stays removed**`assert_auto_restart_helper_absent()`
confirms `_trigger_mcp_auto_restart` has not returned.
5. **pkill stays forbidden** — a daemon `pkill` command still classifies as
contamination via `runtime_recovery_guard`.
## Residual host behaviors (outside process control)
* `/mcp reconnect` in the IDE/host — the sanctioned recovery for stale-runtime,
transport-flap (#584), and worktree-binding conditions. The daemon can only
emit guidance toward it.
* Host-level process management (the operator relaunching the daemon after a
fail-closed stop, or after resolving merge conflicts).
These are documented rather than code-guarded because the process cannot
observe or gate them from inside itself.
## Rollout
Per #657, guards are introduced flag-free as **regression assertions** (they
codify invariants that already hold) before any hard runtime block is layered
on. When the restart coordinator (#655/#656) lands, registered paths gain a
coordinator token/capability check; unregistered attempts already fail closed
today via `assert_restart_attempt_registered()`.
-1
View File
@@ -69,7 +69,6 @@ that gates each call, not which tools exist.
- `gitea_audit_worktree_cleanup`
- `gitea_authorize_reconciliation_cleanup_phase`
- `gitea_authorize_review_correction`
- `gitea_bootstrap_author_issue_worktree`
- `gitea_capability_stop_terminal_report`
- `gitea_capture_branches_worktree_snapshot`
- `gitea_check_pr_eligibility`
+142 -269
View File
@@ -977,32 +977,6 @@ def _create_issue_bootstrap_assessment(
"""
import create_issue_bootstrap as _cib
import author_issue_bootstrap as _aib
if _aib.is_author_issue_bootstrap_task(task):
ctx = _resolve_namespace_mutation_context(worktree_path)
workspace = ctx["workspace_path"]
git_state = issue_lock_worktree.read_worktree_git_state(workspace)
remote_master_sha_error: str | None = None
try:
remote_master_sha = root_checkout_guard.resolve_remote_master_sha(
ctx["canonical_repo_root"]
)
except Exception as exc:
remote_master_sha = None
remote_master_sha_error = (
f"{type(exc).__name__}: {exc}".strip() or "resolver failed"
)
return _aib.assess_author_issue_bootstrap(
workspace_path=workspace,
canonical_repo_root=ctx["canonical_repo_root"],
current_branch=git_state.get("current_branch"),
head_sha=git_state.get("head_sha"),
porcelain_status=git_state.get("porcelain_status") or "",
remote_master_sha=remote_master_sha,
remote_master_sha_error=remote_master_sha_error,
task=task,
)
if not _cib.is_create_issue_task(task):
return None
@@ -1495,7 +1469,7 @@ def verify_preflight_purity(
dirty_files = sorted(
_parse_porcelain_entries(_get_workspace_porcelain(workspace))
)
if dirty_files and task != "commit_files":
if dirty_files:
raise RuntimeError(
nwb.format_namespace_workspace_binding_error(
role_kind=role,
@@ -9313,7 +9287,6 @@ def gitea_commit_files(
host: str | None = None,
org: str | None = None,
repo: str | None = None,
worktree_path: str | None = None,
) -> dict:
"""Commit changes to multiple files in a Gitea repository in a single atomic commit.
@@ -9326,46 +9299,10 @@ def gitea_commit_files(
host: Override the Gitea host.
org: Override the owner/organization.
repo: Override the repository name.
worktree_path: Optional worktree path for author mutation context.
Returns:
dict with success status and commit/branch information.
"""
if worktree_path is None:
lock_data = issue_lock_store.read_session_issue_lock() or {}
worktree_path = lock_data.get("worktree_path")
if not worktree_path:
try:
prof = get_profile()
uname = prof.get("username") or prof.get("profile_name")
for path in issue_lock_store.iter_lock_files():
lk = issue_lock_store.read_lock_file(path) or {}
claimant = lk.get("claimant") or {}
if lk.get("remote") == remote and (claimant.get("username") == uname or lk.get("profile") == prof.get("profile_name")):
issue_lock_store.bind_session_lock(lk, renewal_sanctioned=True)
worktree_path = lk.get("worktree_path")
break
except Exception:
pass
if worktree_path is None and files:
for f in files:
p = f.get("workspace_path") or f.get("local_path") or ""
if p and os.path.isabs(p):
real_p = os.path.realpath(p)
real_root = os.path.realpath(PROJECT_ROOT)
branches_dir = os.path.join(real_root, "branches")
if real_p.startswith(branches_dir + os.sep):
rel_sub = os.path.relpath(real_p, branches_dir)
wt_folder = rel_sub.split(os.sep)[0]
if wt_folder and wt_folder != "..":
worktree_path = os.path.join(branches_dir, wt_folder)
break
if worktree_path:
os.environ["GITEA_AUTHOR_WORKTREE"] = worktree_path
os.environ["GITEA_ACTIVE_WORKTREE"] = worktree_path
ok, block_reasons = role_session_router.check_author_mutation_after_reviewer_stop(
"commit_files"
)
@@ -9378,7 +9315,7 @@ def gitea_commit_files(
"reasons": block_reasons,
}
blocked = _namespace_mutation_block(
"commit_files", commit="", branch="", remote=remote, worktree_path=worktree_path
"commit_files", commit="", branch="", remote=remote
)
if blocked:
return blocked
@@ -9406,7 +9343,7 @@ def gitea_commit_files(
)
# #735: forward explicit org/repo into shared anti-stomp preflight.
verify_preflight_purity(remote=remote, worktree_path=worktree_path, task="commit_files", org=org, repo=repo)
verify_preflight_purity(remote, task="commit_files", org=org, repo=repo)
processed_files, source_proofs = _prepare_commit_payload_files(files)
h, o, r = _resolve(remote, host, org, repo)
@@ -9683,96 +9620,6 @@ def gitea_publish_unpublished_issue_branch(
}
@mcp.tool()
def gitea_bootstrap_author_issue_worktree(
issue_number: int,
assignment_id: str | None = None,
lease_id: str | None = None,
expected_base_sha: str | None = None,
branch_name: str | None = None,
worktree_path: str | None = None,
idempotency_key: str | None = None,
remote: str = "dadeschools",
host: str | None = None,
org: str | None = None,
repo: str | None = None,
dry_run: bool = False,
) -> dict:
"""Bootstrap an allocated author issue branch and registered worktree (#850).
Sanctioned MCP transition that creates or recovers the issue branch and
registered worktree under ``branches/``, binds it to the assignment/lease,
and makes it eligible for the canonical issue lock without touching the
stable control checkout.
Args:
issue_number: Allocated issue number to bootstrap.
assignment_id: Optional allocation assignment ID.
lease_id: Optional workflow lease ID.
expected_base_sha: Authoritative expected base SHA / concurrency pin.
branch_name: Optional custom branch name (must match issue-<N> pattern).
worktree_path: Optional custom worktree path under branches/.
idempotency_key: Optional key for idempotent replay/resume.
remote: Known instance 'dadeschools' or 'prgs'.
host: Override Gitea host.
org: Override Org.
repo: Override Repo.
dry_run: Report planned transition without mutating repository.
"""
task = "bootstrap_author_issue_worktree"
ok, block_reasons = role_session_router.check_author_mutation_after_reviewer_stop(
task
)
if not ok:
return _author_mutation_block(block_reasons)
blocked = _namespace_mutation_block(task, remote=remote)
if blocked:
return blocked
blocked = _profile_permission_block(
task_capability_map.required_permission(task),
remote=remote,
host=host,
org=org,
repo=repo,
org_explicit=org is not None,
repo_explicit=repo is not None,
)
if blocked:
return blocked
verify_preflight_purity(
remote,
task=task,
org=org,
repo=repo,
)
h, o, r = _resolve(remote, host, org, repo)
canonical_root = _canonical_local_git_root()
import author_issue_bootstrap
return author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=issue_number,
canonical_repo_root=canonical_root,
assignment_id=assignment_id,
lease_id=lease_id,
expected_base_sha=expected_base_sha,
branch_name=branch_name,
worktree_path=worktree_path,
idempotency_key=idempotency_key,
remote=remote,
host=h,
org=o,
repo=r,
active_identity=_active_username(),
active_profile=_active_profile_name(),
owner_session=_current_session_id(),
dry_run=dry_run,
)
# Merge methods supported by the Gitea merge API.
_MERGE_METHODS = ("merge", "squash", "rebase")
@@ -11672,127 +11519,163 @@ def gitea_reconcile_merged_cleanups(
if dry_run:
report["dry_run"] = True
report["executed"] = False
# #851: surface planned lifecycle order so dry-run matches execute.
report["planned_execution_orders"] = {
str(entry.get("pr_number")): entry.get("planned_execution_order") or []
for entry in (report.get("entries") or [])
}
return {"success": True, "performed": False, **report}
verify_preflight_purity(
remote, task="reconcile_merged_cleanups", org=org, repo=repo
)
actions: list[dict] = []
project_root = _canonical_local_git_root()
def _ownership_records_for_branch(
head_branch: str, pr_num_int: int | None
) -> list[dict]:
ownership_bundle = _collect_branch_ownership_records(
remote=remote,
host=h,
org=o,
repo=r,
branch=head_branch,
pr_number=pr_num_int,
project_root=project_root,
auth=auth,
base_api=base,
)
ownership_records = list(ownership_bundle.get("records") or [])
if ownership_bundle.get("inventory_error"):
ownership_records.append(
{
"category": (
branch_cleanup_guard.OWNERSHIP_CATEGORY_INVENTORY_ERROR
),
"status": "unknown",
"remote": remote,
"host": h,
"org": o,
"repo": r,
"branch": head_branch,
"reclaim_allowed": False,
"role": "inventory",
}
)
return ownership_records
def _attempt_owned_remote_delete(
*,
head_branch: str,
pr_num_int: int | None,
after_worktree_removal: bool = False,
) -> dict:
"""Fail-closed remote delete with live ownership reassessment (#851)."""
import urllib.parse
ownership_records = _ownership_records_for_branch(head_branch, pr_num_int)
ownership = branch_cleanup_guard.assess_active_branch_ownership(
remote=remote,
org=o,
repo=r,
branch=head_branch,
host=h,
records=ownership_records,
)
if ownership.get("block"):
return {
"action": "delete_remote_branch",
"branch": head_branch,
"success": False,
"performed": False,
"delete_acknowledged": False,
"verified_absent": False,
"blocker_kind": "active_branch_ownership",
"reasons": ownership.get("reasons") or [],
"blocking_categories": ownership.get("blocking_categories") or [],
"after_worktree_removal": after_worktree_removal,
"ownership_reassessed": after_worktree_removal,
}
encoded = urllib.parse.quote(head_branch, safe="")
url = f"{base}/branches/{encoded}"
with _audited(
"delete_branch",
host=h,
remote=remote,
org=o,
repo=r,
target_branch=head_branch,
request_metadata={
"branch": head_branch,
"source": "reconcile_merged_cleanups",
"ownership_checked": True,
"after_worktree_removal": after_worktree_removal,
},
):
api_request("DELETE", url, auth)
readback = _probe_remote_branch(h, o, r, auth, head_branch)
readback_assessment = branch_cleanup_guard.assess_post_delete_readback(
readback
)
verified = bool(readback_assessment.get("verified_absent"))
return {
"action": "delete_remote_branch",
"branch": head_branch,
"success": bool(readback_assessment.get("ok")),
"performed": True,
"delete_acknowledged": True,
"verified_absent": verified,
"readback": readback_assessment.get("readback"),
"reasons": readback_assessment.get("reasons") or [],
"after_worktree_removal": after_worktree_removal,
"ownership_reassessed": after_worktree_removal,
}
for entry in report.get("entries") or []:
head_branch = entry.get("head_branch") or ""
remote_assessment = entry.get("remote_branch") or {}
local_assessment = entry.get("local_worktree") or {}
pr_num = entry.get("pr_number")
try:
pr_num_int = int(pr_num) if pr_num is not None else None
except (TypeError, ValueError):
pr_num_int = None
if remote_assessment.get("safe_to_delete_remote"):
import urllib.parse
pr_num = entry.get("pr_number")
try:
pr_num_int = int(pr_num) if pr_num is not None else None
except (TypeError, ValueError):
pr_num_int = None
ownership_bundle = _collect_branch_ownership_records(
remote=remote,
host=h,
org=o,
repo=r,
branch=head_branch,
pr_number=pr_num_int,
project_root=_canonical_local_git_root(),
auth=auth,
base_api=base,
)
ownership_records = list(ownership_bundle.get("records") or [])
if ownership_bundle.get("inventory_error"):
ownership_records.append(
{
"category": (
branch_cleanup_guard.OWNERSHIP_CATEGORY_INVENTORY_ERROR
),
"status": "unknown",
"remote": remote,
"host": h,
"org": o,
"repo": r,
"branch": head_branch,
"reclaim_allowed": False,
"role": "inventory",
}
)
ownership = branch_cleanup_guard.assess_active_branch_ownership(
remote=remote,
org=o,
repo=r,
branch=head_branch,
host=h,
records=ownership_records,
)
if ownership.get("block"):
actions.append(
{
"action": "delete_remote_branch",
"branch": head_branch,
"success": False,
"performed": False,
"delete_acknowledged": False,
"verified_absent": False,
"blocker_kind": "active_branch_ownership",
"reasons": ownership.get("reasons") or [],
"blocking_categories": ownership.get(
"blocking_categories"
)
or [],
}
)
continue
encoded = urllib.parse.quote(head_branch, safe="")
url = f"{base}/branches/{encoded}"
with _audited(
"delete_branch",
host=h,
remote=remote,
org=o,
repo=r,
target_branch=head_branch,
request_metadata={
"branch": head_branch,
"source": "reconcile_merged_cleanups",
"ownership_checked": True,
},
):
api_request("DELETE", url, auth)
readback = _probe_remote_branch(h, o, r, auth, head_branch)
readback_assessment = branch_cleanup_guard.assess_post_delete_readback(
readback
)
verified = bool(readback_assessment.get("verified_absent"))
actions.append(
{
"action": "delete_remote_branch",
"branch": head_branch,
"success": bool(readback_assessment.get("ok")),
"performed": True,
"delete_acknowledged": True,
"verified_absent": verified,
"readback": readback_assessment.get("readback"),
"reasons": readback_assessment.get("reasons") or [],
}
)
# #851 lifecycle: when the target worktree is independently safe, remove
# it first so worktree_binding ownership does not permanently strand
# both the worktree and the remote branch. Never skip worktree removal
# merely because remote delete would be blocked by that binding.
# Ownership protection for remote delete remains fail-closed below.
worktree_removed = False
if local_assessment.get("safe_to_remove_worktree"):
result = merged_cleanup_reconcile.remove_local_worktree(
_canonical_local_git_root(),
project_root,
head_branch,
worktree_path=local_assessment.get("worktree_path"),
)
actions.append({"action": "remove_local_worktree", **result})
# Idempotent resume: absent worktree is already gone.
msg = (result.get("message") or "").lower()
worktree_removed = bool(result.get("success")) or (
"not found" in msg
)
if remote_assessment.get("safe_to_delete_remote"):
actions.append(
_attempt_owned_remote_delete(
head_branch=head_branch,
pr_num_int=pr_num_int,
after_worktree_removal=worktree_removed,
)
)
for scratch in report.get("reviewer_scratch_entries") or []:
if not scratch.get("safe_to_remove_worktree"):
continue
result = merged_cleanup_reconcile.remove_reviewer_scratch_worktree(
_canonical_local_git_root(), scratch.get("worktree_path") or ""
project_root, scratch.get("worktree_path") or ""
)
actions.append({"action": "remove_reviewer_scratch_worktree", **result})
@@ -19705,12 +19588,6 @@ def gitea_resolve_task_capability(
task: str,
remote: str = "dadeschools",
host: str | None = None,
org: str | None = None,
repo: str | None = None,
issue_number: int | None = None,
worktree_path: str | None = None,
pr_number: int | None = None,
**kwargs: Any,
) -> dict:
"""Read-only / side-effect free: resolve capability, profile, and namespace for a task.
@@ -19727,14 +19604,13 @@ def gitea_resolve_task_capability(
remote: Known remote instance name.
host: Optional override for the Gitea host.
"""
task_key = task_capability_map._canonical_preflight_task(task)
TASK_MAP = task_capability_map.TASK_CAPABILITY_MAP
# Every fresh attempt invalidates the previous task/role stamp before any
# fallible resolver work. Unknown/malformed tasks and unexpected failures
# therefore remain fail-closed instead of preserving stale authority.
_clear_resolved_capability_stamp()
if task_key not in TASK_MAP:
if task not in TASK_MAP:
# #723: structured fail-closed unknown_task (never raise into internal_error).
profile = get_profile()
h = host or (REMOTES.get(remote, {}).get("host") if remote in REMOTES else None)
@@ -19780,8 +19656,8 @@ def gitea_resolve_task_capability(
result["cleared_stale_denial"] = True
return result
required_permission = task_capability_map.required_permission(task_key)
required_role = task_capability_map.required_role(task_key)
required_permission = task_capability_map.required_permission(task)
required_role = task_capability_map.required_role(task)
role_exclusive_tasks = task_capability_map.ROLE_EXCLUSIVE_TASKS
infra_assessment = role_session_router.assess_infra_stop(PROJECT_ROOT)
@@ -19967,10 +19843,7 @@ def gitea_resolve_task_capability(
available_in_session = allowed_in_current_session
runtime_stale_blocker = False
if (
"PYTEST_CURRENT_TEST" not in os.environ
or "GITEA_FORCE_MCP_RUNTIME_CHECK" in os.environ
) and os.environ.get("GITEA_ALLOW_STALE_RUNTIME") != "1":
if "PYTEST_CURRENT_TEST" not in os.environ or "GITEA_FORCE_MCP_RUNTIME_CHECK" in os.environ:
runtime_reasons = _check_mcp_runtimes_diagnostics(task, matching_profiles)
if runtime_reasons:
restart_required = True
+475
View File
@@ -0,0 +1,475 @@
"""Inventory and fail-closed guards for MCP restart/reload/kill paths (#657).
Single source of truth enumerating every code/script/doc path that can
restart, reload, reconnect, kill, or force-recreate an MCP process. Each path
is classified and linked to the guard that constrains it. The companion
human-readable inventory lives in ``docs/mcp-restart-path-inventory.md`` and is
kept in lock-step with this module by ``tests/test_mcp_restart_paths.py``.
Design intent (aligns with #655 restart-coordinator roadmap):
* **No unguarded full restart.** The in-process MCP daemon
(``gitea_mcp_server.py`` / ``mcp_server.py`` / ``role_session_router.py``)
must never replace or kill its own process — replacing the process after the
host wired up the stdio pipes desyncs the JSON-RPC transport (observed with
Antigravity/Cascade hosts). ``assert_no_daemon_self_replacement`` enforces
this against the live source tree.
* **No legacy auto-restart helper.** ``_trigger_mcp_auto_restart`` was removed
when the stale-runtime resolver became side-effect free (#685);
``assert_auto_restart_helper_absent`` keeps it removed.
* **Unknown restart attempts fail closed.** LLM tools must route any restart
intent through a *registered* path. ``assert_restart_attempt_registered``
raises ``UnknownRestartPathError`` for anything not in this inventory.
* **pkill stays forbidden (#630).** Manual daemon kills are classified as
contamination by :mod:`runtime_recovery_guard`; this module records that path
and the test asserts the classification still holds.
This module performs no restarts, spawns no threads, and touches no config or
process state. It is pure inventory + read-only source assertions.
"""
from __future__ import annotations
import os
from dataclasses import dataclass
from pathlib import Path
from typing import Iterable
# --- Classifications -------------------------------------------------------
#: A narrow, one-shot recovery that is safe by construction (e.g. a CLI wrapper
#: re-execing into the venv interpreter before importing anything, or an
#: in-process profile switch). Never targets the running MCP daemon process.
CLASS_SANCTIONED_NARROW = "sanctioned_narrow_recovery"
#: The path detects a condition that would require a restart, then *fails
#: closed* on mutations and emits restart/reconnect guidance. It never restarts
#: the process itself (recovery is owned by the host/operator).
CLASS_GUARDED_FAIL_CLOSED = "guarded_fail_closed"
#: The path is forbidden. Attempting it is a workflow-safety violation and,
#: where an LLM tool could invoke it, is marked as contamination.
CLASS_FORBIDDEN = "forbidden"
#: A previously-existing unguarded restart primitive that has been deleted. A
#: regression guard keeps it absent.
CLASS_REMOVED = "removed"
#: Behavior that lives in the host/IDE and is outside this process's control
#: (e.g. a manual ``/mcp reconnect``). Documented, not code-guarded here.
CLASS_HOST_RESIDUAL = "host_residual"
VALID_CLASSIFICATIONS = frozenset(
{
CLASS_SANCTIONED_NARROW,
CLASS_GUARDED_FAIL_CLOSED,
CLASS_FORBIDDEN,
CLASS_REMOVED,
CLASS_HOST_RESIDUAL,
}
)
#: The in-process MCP daemon modules. These must never self-replace/self-kill.
DAEMON_MODULES = (
"gitea_mcp_server.py",
"mcp_server.py",
"role_session_router.py",
)
#: The legacy auto-restart helper removed in #685. Must stay removed.
LEGACY_AUTO_RESTART_HELPER = "_trigger_mcp_auto_restart"
#: Call patterns that would let the daemon replace or terminate its own
#: process. Matched as calls (trailing ``(``) so prose/docstring mentions such
#: as "we do NOT os.execv() here" or "never calls ``os._exit``" do not trip the
#: scanner (comment lines are stripped first regardless).
DAEMON_SELF_REPLACEMENT_PRIMITIVES = (
"os.execv(",
"os.execve(",
"os.execvp(",
"os.execvpe(",
"os.kill(",
"os.killpg(",
"os._exit(",
"os.abort(",
)
@dataclass(frozen=True)
class RestartPath:
"""One classified restart/reload/kill path in the inventory."""
path_id: str
title: str
mechanism: str
classification: str
guard: str
locations: tuple[str, ...]
references: tuple[str, ...]
residual_host: bool = False
notes: str = ""
class UnknownRestartPathError(RuntimeError):
"""Raised when a restart attempt is not a registered, classified path."""
# --- The inventory ---------------------------------------------------------
_RESTART_PATHS: tuple[RestartPath, ...] = (
RestartPath(
path_id="cli_venv_bootstrap_execv",
title="CLI wrapper venv re-exec",
mechanism=(
"Standalone CLI scripts re-exec into venv/bin/python3 via os.execv "
"at import top, guarded by `sys.executable != venv_python`."
),
classification=CLASS_SANCTIONED_NARROW,
guard=(
"One-shot, pre-import bootstrap; runs before any MCP transport "
"exists and only when not already on the venv interpreter, so it "
"cannot desync a live daemon. Idempotent guard condition prevents "
"a re-exec loop."
),
locations=(
"create_pr.py",
"create_issue.py",
"close_issue.py",
"merge_pr.py",
"review_pr.py",
"edit_pr.py",
"delete_branch.py",
"mark_issue.py",
"manage_labels.py",
"list_issues.py",
"list_prs.py",
),
references=("#657",),
),
RestartPath(
path_id="daemon_self_replacement",
title="MCP daemon self-replacement",
mechanism=(
"The in-process MCP daemon replacing/terminating its own process "
"(os.execv/os.kill/os._exit) to reload code."
),
classification=CLASS_FORBIDDEN,
guard=(
"Forbidden by design: replacing the process after the host wired "
"up stdio desyncs JSON-RPC (Antigravity/Cascade). Enforced against "
"the source tree by assert_no_daemon_self_replacement()."
),
locations=("gitea_mcp_server.py:~155 (decision comment)",) + DAEMON_MODULES,
references=("#657", "#584"),
),
RestartPath(
path_id="legacy_auto_restart_helper",
title="Legacy _trigger_mcp_auto_restart helper",
mechanism=(
"A helper that actively restarted the MCP server from the "
"read-only resolver path."
),
classification=CLASS_REMOVED,
guard=(
"Removed in #685 when the resolver became side-effect free. Kept "
"absent by assert_auto_restart_helper_absent()."
),
locations=("gitea_mcp_server.py", "mcp_server.py"),
references=("#685", "#657"),
),
RestartPath(
path_id="config_touch_reload",
title="MCP client config-touch reload",
mechanism=(
"Touching (utime) the MCP client config file to make the host "
"reload/recreate the server process."
),
classification=CLASS_REMOVED,
guard=(
"Removed from the resolver in #685: stale-runtime detection is "
"report-only and never mutates client config, spawns threads, or "
"calls os._exit."
),
locations=("gitea_mcp_server.py (resolve_task_capability)",),
references=("#685", "#657"),
),
RestartPath(
path_id="master_advance_auto_restart",
title="Master-advance staleness gate",
mechanism=(
"On-disk master advancing past the running code. The master-parity "
"gate detects it and fails mutations closed with restart guidance."
),
classification=CLASS_GUARDED_FAIL_CLOSED,
guard=(
"Detect + fail closed only; the process never self-restarts. "
"master_parity_gate captures startup parity and blocks mutations "
"while stale, emitting restart/reconnect guidance."
),
locations=(
"master_parity_gate.py",
"gitea_mcp_server.py (gitea_assess_master_parity)",
),
references=("#420", "#591", "#657"),
),
RestartPath(
path_id="stale_runtime_resolver_reconnect",
title="Stale-runtime resolver reconnect guidance",
mechanism=(
"The capability resolver detecting a stale serving process and "
"reporting restart_required/stop_required for a client reconnect."
),
classification=CLASS_GUARDED_FAIL_CLOSED,
guard=(
"Report-only (#685): returns restart_required/stop_required and an "
"exact_safe_next_action pointing at IDE/client reconnect; performs "
"no restart, thread spawn, config touch, or os._exit."
),
locations=("gitea_mcp_server.py (gitea_resolve_task_capability)",),
references=("#685", "#657"),
),
RestartPath(
path_id="manual_daemon_kill",
title="Manual daemon kill (pkill/killall/kill)",
mechanism=(
"Shell kills of the MCP daemon: `pkill -f mcp_server.py`, "
"`killall`, broad `pkill -f python` sweeps, or `kill <pid>` of a "
"daemon pid."
),
classification=CLASS_FORBIDDEN,
guard=(
"Forbidden (#630): runtime_recovery_guard classifies these as "
"contamination and gitea_record_daemon_process_kill_attempt writes "
"a durable marker that fails subsequent mutations closed. Operator "
"maintenance authorization is read only from the environment, not "
"from a tool argument."
),
locations=(
"runtime_recovery_guard.py",
"gitea_mcp_server.py (gitea_record_daemon_process_kill_attempt)",
),
references=("#630", "#657"),
),
RestartPath(
path_id="conflict_marker_infra_stop",
title="Startup conflict-marker infra stop",
mechanism=(
"The daemon entrypoint scans for unresolved merge-conflict markers "
"at startup and stops (sys.exit(1)) if found."
),
classification=CLASS_GUARDED_FAIL_CLOSED,
guard=(
"Fail-closed startup stop, not a restart: the process exits and "
"waits for the operator to resolve conflicts and relaunch. Never "
"self-restarts or loops."
),
locations=("mcp_server.py (check_conflict_markers)",),
references=("#657",),
),
RestartPath(
path_id="ide_client_reconnect",
title="Host/IDE MCP reconnect",
mechanism=(
"A manual `/mcp reconnect` (or equivalent host action) that the "
"IDE performs to recreate the MCP client connection."
),
classification=CLASS_HOST_RESIDUAL,
guard=(
"Outside this process's control. It is the sanctioned recovery the "
"gates point operators toward; documented as residual host "
"behavior. No in-process code initiates it."
),
locations=("host/IDE",),
references=("#584", "#656", "#657"),
residual_host=True,
),
RestartPath(
path_id="profile_switch_runtime",
title="Runtime profile switch",
mechanism=(
"Switching the active execution profile at runtime "
"(dynamic-profile mode)."
),
classification=CLASS_SANCTIONED_NARROW,
guard=(
"In-process and restart-free: runtime_switching_supported is true, "
"so a profile switch rebinds capability without recreating the "
"process. No restart primitive is invoked."
),
locations=("gitea_mcp_server.py (gitea_activate_profile)",),
references=("#656", "#657"),
),
)
_BY_ID: dict[str, RestartPath] = {p.path_id: p for p in _RESTART_PATHS}
# --- Read-only accessors ---------------------------------------------------
def iter_restart_paths() -> tuple[RestartPath, ...]:
"""Return the full inventory as an immutable tuple."""
return _RESTART_PATHS
def restart_path_ids() -> frozenset[str]:
"""Return the set of registered path ids."""
return frozenset(_BY_ID)
def get_restart_path(path_id: str) -> RestartPath:
"""Return the registered path, or raise :class:`UnknownRestartPathError`."""
try:
return _BY_ID[path_id]
except KeyError as exc:
raise UnknownRestartPathError(
f"unknown restart path id {path_id!r}; not in the #657 inventory"
) from exc
def paths_by_classification(classification: str) -> tuple[RestartPath, ...]:
"""Return all registered paths with the given classification."""
if classification not in VALID_CLASSIFICATIONS:
raise ValueError(f"unknown classification {classification!r}")
return tuple(p for p in _RESTART_PATHS if p.classification == classification)
def assert_restart_attempt_registered(path_id: str) -> RestartPath:
"""Fail closed unless ``path_id`` is a registered, classified restart path.
LLM tools that intend to trigger any restart/reload/reconnect must name a
registered path so an unknown/novel restart primitive cannot slip through
silently. Forbidden and removed paths are registered too — this only
asserts the attempt is *known*, not that it is *permitted*; callers must
still honor the classification.
"""
return get_restart_path(path_id)
def assert_registry_wellformed() -> None:
"""Validate the inventory's own invariants (fail closed on drift)."""
seen: set[str] = set()
for path in _RESTART_PATHS:
if path.path_id in seen:
raise ValueError(f"duplicate restart path id {path.path_id!r}")
seen.add(path.path_id)
if path.classification not in VALID_CLASSIFICATIONS:
raise ValueError(
f"{path.path_id!r} has invalid classification "
f"{path.classification!r}"
)
if not path.guard.strip():
raise ValueError(f"{path.path_id!r} is missing a guard description")
if not path.references:
raise ValueError(f"{path.path_id!r} is missing references")
if not path.locations:
raise ValueError(f"{path.path_id!r} is missing locations")
if path.classification == CLASS_HOST_RESIDUAL and not path.residual_host:
raise ValueError(
f"{path.path_id!r} is host_residual but residual_host is False"
)
# --- Source-tree guards ----------------------------------------------------
def _repo_root(root: str | os.PathLike[str] | None = None) -> Path:
if root is not None:
return Path(root)
return Path(__file__).resolve().parent
def _iter_code_lines(text: str) -> Iterable[tuple[int, str]]:
"""Yield (1-based lineno, line) for lines that are not full-line comments."""
for lineno, line in enumerate(text.splitlines(), start=1):
if line.lstrip().startswith("#"):
continue
yield lineno, line
def scan_daemon_self_replacement(
root: str | os.PathLike[str] | None = None,
) -> list[dict[str, object]]:
"""Return violations where a daemon module could self-replace/self-kill.
Scans :data:`DAEMON_MODULES` for calls in
:data:`DAEMON_SELF_REPLACEMENT_PRIMITIVES`. Full-line comments are ignored,
and only call forms (with a trailing ``(``) match, so decision comments and
docstrings that merely mention the primitives do not produce false hits.
"""
repo = _repo_root(root)
violations: list[dict[str, object]] = []
for module in DAEMON_MODULES:
path = repo / module
if not path.exists():
continue
text = path.read_text(encoding="utf-8", errors="replace")
for lineno, line in _iter_code_lines(text):
for primitive in DAEMON_SELF_REPLACEMENT_PRIMITIVES:
if primitive in line:
violations.append(
{
"module": module,
"line": lineno,
"primitive": primitive,
"text": line.strip(),
}
)
return violations
def assert_no_daemon_self_replacement(
root: str | os.PathLike[str] | None = None,
) -> None:
"""Fail closed if any daemon module can restart/kill its own process."""
violations = scan_daemon_self_replacement(root)
if violations:
rendered = "; ".join(
f"{v['module']}:{v['line']} {v['primitive']}" for v in violations
)
raise AssertionError(
"MCP daemon must never self-replace/self-kill (#657); found: "
f"{rendered}"
)
def scan_auto_restart_helper(
root: str | os.PathLike[str] | None = None,
) -> list[dict[str, object]]:
"""Return occurrences of a *definition* of the legacy auto-restart helper."""
repo = _repo_root(root)
needle = f"def {LEGACY_AUTO_RESTART_HELPER}"
hits: list[dict[str, object]] = []
for module in DAEMON_MODULES:
path = repo / module
if not path.exists():
continue
text = path.read_text(encoding="utf-8", errors="replace")
for lineno, line in _iter_code_lines(text):
if needle in line:
hits.append({"module": module, "line": lineno})
return hits
def assert_auto_restart_helper_absent(
root: str | os.PathLike[str] | None = None,
) -> None:
"""Fail closed if the removed ``_trigger_mcp_auto_restart`` reappears."""
hits = scan_auto_restart_helper(root)
if hits:
rendered = "; ".join(f"{h['module']}:{h['line']}" for h in hits)
raise AssertionError(
f"{LEGACY_AUTO_RESTART_HELPER} was removed in #685 and must not "
f"return (#657); found definition at: {rendered}"
)
-2
View File
@@ -73,8 +73,6 @@ AUTHOR_TASKS = frozenset({
"claim_issue",
"create_branch",
"push_branch",
"bootstrap_author_issue_worktree",
"gitea_bootstrap_author_issue_worktree",
"create_pr",
"comment_pr",
"address_pr_change_requests",
-13
View File
@@ -69,14 +69,6 @@ TASK_CAPABILITY_MAP: dict[str, dict[str, str]] = {
"permission": "gitea.branch.create",
"role": "author",
},
"bootstrap_author_issue_worktree": {
"permission": "gitea.branch.create",
"role": "author",
},
"gitea_bootstrap_author_issue_worktree": {
"permission": "gitea.branch.create",
"role": "author",
},
"push_branch": {
"permission": "gitea.branch.push",
"role": "author",
@@ -496,8 +488,6 @@ TASK_CAPABILITY_MAP: dict[str, dict[str, str]] = {
# merger lease (#763).
_PREFLIGHT_TASK_TRANSITIONS = frozenset({
("review_pr", "acquire_reviewer_pr_lease"),
("work_issue", "bootstrap_author_issue_worktree"),
("bootstrap_author_issue_worktree", "lock_issue"),
})
@@ -544,8 +534,6 @@ ROLE_EXCLUSIVE_TASKS: frozenset[str] = frozenset(
"gitea_release_merger_pr_lease",
"create_branch",
"push_branch",
"bootstrap_author_issue_worktree",
"gitea_bootstrap_author_issue_worktree",
"publish_unpublished_branch",
"create_pr",
"commit_files",
@@ -571,7 +559,6 @@ ISSUE_MUTATION_TOOL_TASKS: dict[str, str] = {
"gitea_set_issue_labels": "set_issue_labels",
"gitea_cleanup_terminal_pr_labels": "cleanup_terminal_pr_labels",
"gitea_create_label": "create_label",
"gitea_bootstrap_author_issue_worktree": "bootstrap_author_issue_worktree",
"gitea_commit_files": "commit_files",
}
-483
View File
@@ -1,483 +0,0 @@
"""Regression test suite for native author issue worktree bootstrap (#850)."""
from __future__ import annotations
import json
import os
import shutil
import subprocess
import tempfile
import unittest
from unittest import mock
import author_issue_bootstrap
import task_capability_map
def _concurrent_bootstrap_worker(args: tuple[str, int, str, str, str, str]) -> dict:
repo_dir, issue_num, key, lock_dir, journal_dir, master_sha = args
os.environ["GITEA_BOOTSTRAP_JOURNAL_DIR"] = journal_dir
return author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=issue_num,
canonical_repo_root=repo_dir,
expected_base_sha=master_sha,
idempotency_key=key,
lock_dir=lock_dir,
owner_session="session-concurrent-test",
active_identity="jcwalker3",
active_profile="prgs-author",
)
class TestAuthorIssueBootstrap(unittest.TestCase):
"""Test suite covering AC1-AC10 and comment #14959 specification."""
def setUp(self):
self.tmp_dir = tempfile.mkdtemp(prefix="test_bootstrap_")
self.repo_dir = os.path.join(self.tmp_dir, "repo")
os.makedirs(self.repo_dir)
# Initialize synthetic git repo
subprocess.run(["git", "init", "-b", "master"], cwd=self.repo_dir, check=True, capture_output=True)
subprocess.run(["git", "config", "user.name", "Test User"], cwd=self.repo_dir, check=True)
subprocess.run(["git", "config", "user.email", "[email protected]"], cwd=self.repo_dir, check=True)
readme = os.path.join(self.repo_dir, "README.md")
with open(readme, "w", encoding="utf-8") as f:
f.write("# Test Repo\n")
subprocess.run(["git", "add", "README.md"], cwd=self.repo_dir, check=True, capture_output=True)
subprocess.run(["git", "commit", "-m", "initial commit"], cwd=self.repo_dir, check=True, capture_output=True)
rev_res = subprocess.run(["git", "rev-parse", "HEAD"], cwd=self.repo_dir, capture_output=True, text=True, check=True)
self.master_sha = rev_res.stdout.strip()
self.branches_dir = os.path.join(self.repo_dir, "branches")
os.makedirs(self.branches_dir, exist_ok=True)
self.lock_dir = os.path.join(self.tmp_dir, "locks")
os.makedirs(self.lock_dir, exist_ok=True)
self.journal_dir = os.path.join(self.tmp_dir, "journals")
os.makedirs(self.journal_dir, exist_ok=True)
os.environ["GITEA_BOOTSTRAP_JOURNAL_DIR"] = self.journal_dir
def tearDown(self):
os.environ.pop("GITEA_BOOTSTRAP_JOURNAL_DIR", None)
shutil.rmtree(self.tmp_dir, ignore_errors=True)
def test_bootstrap_success_path(self):
"""AC1/AC3/AC8: Successful bootstrap creates branch, worktree, registration, and lock proof."""
key = "test_key_success_1"
res = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
assignment_id="asn-12345",
lease_id="lease-67890",
expected_base_sha=self.master_sha,
idempotency_key=key,
remote="prgs",
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
self.assertTrue(res.get("success"), f"Bootstrap failed: {res}")
self.assertFalse(res.get("replayed"))
self.assertEqual(res.get("issue_number"), 850)
self.assertEqual(res.get("base_sha"), self.master_sha)
self.assertIn("branches/fix-issue-850-native-mcp-bootstrap", res.get("worktree_path"))
# Verify worktree directory exists and is registered
worktree_path = res["worktree_path"]
self.assertTrue(os.path.isdir(worktree_path))
wt_list = subprocess.run(["git", "-C", self.repo_dir, "worktree", "list"], capture_output=True, text=True, check=True)
self.assertIn(worktree_path, wt_list.stdout)
# Verify phase journal written
journal = author_issue_bootstrap.load_phase_journal(key, journal_dir=self.lock_dir)
self.assertIsNotNone(journal)
self.assertTrue(journal.get("completed"))
self.assertEqual(journal.get("current_phase"), author_issue_bootstrap.PHASE_7_TRANSITION_COMPLETED)
def test_idempotent_replay(self):
"""Item 2: Replaying with identical key returns cached transition without duplicate creation."""
key = "test_key_idempotent_1"
res1 = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
idempotency_key=key,
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
self.assertTrue(res1["success"], f"res1 failed: {res1}")
self.assertFalse(res1.get("replayed"))
# Second call
res2 = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
idempotency_key=key,
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
self.assertTrue(res2["success"], f"res2 failed: {res2}")
self.assertTrue(res2.get("replayed"))
self.assertEqual(res1["worktree_path"], res2["worktree_path"])
def test_stale_concurrency_pin_refusal(self):
"""Item 3: Mismatched expected base SHA fails closed without silent rebasing."""
stale_sha = "0000000000000000000000000000000000000000"
res = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
expected_base_sha=stale_sha,
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
self.assertFalse(res["success"])
self.assertEqual(res.get("reason_code"), "stale_concurrency_pin")
self.assertIn("exact_next_action", res)
def test_path_outside_branches_root_refusal(self):
"""Item 6: Worktree path outside branches/ root is refused."""
outside_path = os.path.join(self.tmp_dir, "outside_worktree")
res = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
worktree_path=outside_path,
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
self.assertFalse(res["success"])
self.assertEqual(res.get("reason_code"), "path_outside_canonical_branches_root")
def test_preexisting_dirty_worktree_preservation(self):
"""Item 6: Preexisting dirty worktree fails closed and is NOT modified or cleaned."""
branch = "fix/issue-850-dirty-test"
wt_path = os.path.join(self.branches_dir, "fix-issue-850-dirty-test")
subprocess.run(["git", "-C", self.repo_dir, "worktree", "add", "-b", branch, wt_path], check=True, capture_output=True)
# Create dirty untracked file
dirty_file = os.path.join(wt_path, "dirty.txt")
with open(dirty_file, "w") as f:
f.write("dirty edits\n")
res = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
branch_name=branch,
worktree_path=wt_path,
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
self.assertFalse(res["success"])
self.assertEqual(res.get("reason_code"), "preexisting_dirty_worktree")
# Prove dirty file is preserved byte-for-byte
self.assertTrue(os.path.exists(dirty_file))
with open(dirty_file, "r") as f:
self.assertEqual(f.read(), "dirty edits\n")
def test_compensating_recovery_on_failed_phase(self):
"""AC4/Item 4: Failure during transition rolls back ONLY newly created artifacts."""
key = "test_key_recovery_1"
# Simulate partial progress in journal
journal = {
"idempotency_key": key,
"issue_number": 850,
"branch_name": "fix/issue-850-recovery-test",
"worktree_path": os.path.join(self.branches_dir, "fix-issue-850-recovery-test"),
"artifacts_created": {
"branch_created": True,
"worktree_dir_created": True,
"worktree_registered": True,
"lock_created": False,
},
"failure_reason": "simulated lock failure",
"current_phase": author_issue_bootstrap.PHASE_5_REGISTRATION_VERIFIED,
"completed": False,
}
# Create the branch and worktree manually to simulate partial state
subprocess.run(["git", "-C", self.repo_dir, "branch", journal["branch_name"]], check=True, capture_output=True)
subprocess.run(["git", "-C", self.repo_dir, "worktree", "add", journal["worktree_path"], journal["branch_name"]], check=True, capture_output=True)
# Run compensating recovery
rec = author_issue_bootstrap.run_compensating_recovery(journal, self.repo_dir)
self.assertTrue(rec["executed"])
self.assertIn(f"worktree_path:{journal['worktree_path']}", rec["rolled_back"])
self.assertIn(f"branch:{journal['branch_name']}", rec["rolled_back"])
# Prove worktree directory and branch were rolled back
self.assertFalse(os.path.exists(journal["worktree_path"]))
branch_check = subprocess.run(["git", "-C", self.repo_dir, "rev-parse", "--verify", journal["branch_name"]], capture_output=True, text=True, check=False)
self.assertNotEqual(branch_check.returncode, 0)
def test_cross_process_concurrency(self):
"""Review #525 Finding 1: Genuine cross-process concurrency locking prevents corruption."""
import concurrent.futures
key = "test_concurrent_key_850"
args = (self.repo_dir, 850, key, self.lock_dir, self.journal_dir, self.master_sha)
with concurrent.futures.ProcessPoolExecutor(max_workers=2) as executor:
fut1 = executor.submit(_concurrent_bootstrap_worker, args)
fut2 = executor.submit(_concurrent_bootstrap_worker, args)
res1 = fut1.result(timeout=10)
res2 = fut2.result(timeout=10)
self.assertTrue(res1["success"], f"res1 failed: {res1}")
self.assertTrue(res2["success"], f"res2 failed: {res2}")
# One process performs creation, the other process receives idempotent replay
replayed_count = sum(1 for r in (res1, res2) if r.get("replayed"))
created_count = sum(1 for r in (res1, res2) if not r.get("replayed"))
self.assertEqual(replayed_count, 1)
self.assertEqual(created_count, 1)
self.assertEqual(res1["worktree_path"], res2["worktree_path"])
def test_interrupted_replay_preserves_artifacts_created_provenance(self):
"""Review #525 Finding 2: Replaying incomplete journal preserves creation provenance monotonically."""
key = "test_key_interrupted_replay_1"
branch = "fix/issue-850-interrupted-replay"
wt_path = os.path.join(self.branches_dir, "fix-issue-850-interrupted-replay")
# Simulate Phase 2/3 completion where branch and worktree directory were created by this transition
journal = {
"idempotency_key": key,
"issue_number": 850,
"branch_name": branch,
"worktree_path": wt_path,
"active_identity": "jcwalker3",
"active_profile": "prgs-author",
"remote": "prgs",
"org": "Scaled-Tech-Consulting",
"repo": "Gitea-Tools",
"phases": {
author_issue_bootstrap.PHASE_1_REQUEST_ACCEPTED: {"status": "completed"},
author_issue_bootstrap.PHASE_2_BRANCH_CONFIRMED: {"status": "completed", "created": True},
},
"artifacts_created": {
"branch_created": True,
"worktree_dir_created": True,
"worktree_registered": True,
"lock_created": False,
},
"current_phase": author_issue_bootstrap.PHASE_3_PATH_RESERVED,
"completed": False,
}
# Pre-create the branch and worktree on disk to simulate partial state after crash
subprocess.run(["git", "-C", self.repo_dir, "branch", branch, self.master_sha], check=True, capture_output=True)
subprocess.run(["git", "-C", self.repo_dir, "worktree", "add", wt_path, branch], check=True, capture_output=True)
author_issue_bootstrap.save_phase_journal(journal, journal_dir=self.lock_dir)
# Now resume/replay the transition but simulate lock binding failure during Phase 6
with mock.patch("issue_lock_store.bind_session_lock", side_effect=RuntimeError("Lock failure test")):
res = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
branch_name=branch,
worktree_path=wt_path,
idempotency_key=key,
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
self.assertFalse(res["success"])
self.assertEqual(res.get("reason_code"), "issue_lock_acquisition_failed")
# Verify that compensating recovery correctly deleted transition-created branch & worktree
# because creation provenance was preserved across replay (NOT downgraded to False!)
self.assertFalse(os.path.exists(wt_path))
branch_check = subprocess.run(["git", "-C", self.repo_dir, "rev-parse", "--verify", branch], capture_output=True, text=True, check=False)
self.assertNotEqual(branch_check.returncode, 0)
def test_transition_created_only_compensation(self):
"""Review #525 Finding 4: Preexisting branch is NOT deleted by compensation when only worktree was transition-created."""
key = "test_key_preexisting_branch_compensation"
preexisting_branch = "fix/issue-850-preexisting"
wt_path = os.path.join(self.branches_dir, "fix-issue-850-preexisting")
# Create branch BEFORE bootstrap (preexisting branch)
subprocess.run(["git", "-C", self.repo_dir, "branch", preexisting_branch, self.master_sha], check=True, capture_output=True)
# Call bootstrap with simulated failure during Phase 6 (lock binding)
with mock.patch("issue_lock_store.bind_session_lock", side_effect=RuntimeError("Simulated lock failure")):
res = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
branch_name=preexisting_branch,
worktree_path=wt_path,
idempotency_key=key,
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
self.assertFalse(res["success"])
# Worktree dir was created by transition -> removed by compensation
self.assertFalse(os.path.exists(wt_path))
# Preexisting branch was NOT created by transition -> MUST BE PRESERVED!
branch_check = subprocess.run(["git", "-C", self.repo_dir, "rev-parse", "--verify", preexisting_branch], capture_output=True, text=True, check=False)
self.assertEqual(branch_check.returncode, 0, "Preexisting branch was deleted by mistake!")
def test_incompatible_idempotency_replay_refusal(self):
"""Review #525 Finding 4: Replaying key with incompatible parameters returns refusal."""
key = "test_key_incompatible_replay"
res1 = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
branch_name="fix/issue-850-param-a",
idempotency_key=key,
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
self.assertTrue(res1["success"])
# Second call with different branch_name
res2 = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
branch_name="fix/issue-850-param-b",
idempotency_key=key,
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
self.assertFalse(res2["success"])
self.assertEqual(res2.get("reason_code"), "incompatible_idempotency_replay")
def test_exact_next_action_satisfiable_via_mcp(self):
"""Review #525 Finding 4: exact_next_action provides satisfiable MCP actions, not shell commands."""
key = "test_key_next_action_mcp"
stale_sha = "0000000000000000000000000000000000000000"
res = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
expected_base_sha=stale_sha,
lock_dir=self.lock_dir,
owner_session="session-test-1234",
)
next_action = res.get("exact_next_action", "")
self.assertNotIn("scripts/worktree-start", next_action)
self.assertNotIn("git worktree add", next_action)
self.assertNotIn("bash", next_action.lower())
def test_missing_owner_session_refusal(self):
"""Finding D: Missing owner_session context fails closed with typed refusal and zero mutation."""
res = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
owner_session=None,
lock_dir=self.lock_dir,
)
self.assertFalse(res["success"])
self.assertEqual(res.get("reason_code"), "missing_owner_session")
self.assertIn("exact_next_action", res)
def test_symlink_lock_file_refusal(self):
"""Finding C: BootstrapTransitionLock refuses to follow symlinks."""
key = "test_symlink_lock_key"
safe_key = "".join(c if c.isalnum() or c in ("-", "_", ".") else "_" for c in key)
lock_path = os.path.join(self.lock_dir, f"{safe_key}.lock")
target_file = os.path.join(self.tmp_dir, "fake_target")
with open(target_file, "w") as f:
f.write("target")
os.symlink(target_file, lock_path)
with self.assertRaises(RuntimeError) as ctx:
with author_issue_bootstrap.BootstrapTransitionLock(key, journal_dir=self.lock_dir):
pass
self.assertIn("symlink", str(ctx.exception).lower())
def test_lock_directory_escape_refusal(self):
"""Finding C: BootstrapTransitionLock refuses keys that escape lock directory."""
with mock.patch("os.path.abspath", return_value="/tmp/outside/evil_key.lock"):
with self.assertRaises(RuntimeError) as ctx:
author_issue_bootstrap.BootstrapTransitionLock("key", journal_dir=self.lock_dir)
self.assertIn("escapes", str(ctx.exception).lower())
def test_missing_active_identity_refusal(self):
"""F-5: Missing active_identity parameter fails closed."""
res = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
owner_session="session-test-1234",
active_identity=None,
active_profile="prgs-author",
lock_dir=self.lock_dir,
)
self.assertFalse(res["success"])
self.assertEqual(res.get("reason_code"), "missing_active_identity")
def test_missing_active_profile_refusal(self):
"""F-5: Missing active_profile parameter fails closed."""
res = author_issue_bootstrap.bootstrap_author_issue_worktree(
issue_number=850,
canonical_repo_root=self.repo_dir,
owner_session="session-test-1234",
active_identity="jcwalker3",
active_profile=None,
lock_dir=self.lock_dir,
)
self.assertFalse(res["success"])
self.assertEqual(res.get("reason_code"), "missing_active_profile")
def test_dirty_worktree_preserved_during_recovery(self):
"""F-4: Compensating recovery does not delete dirty worktree."""
branch = "fix/issue-850-rec-dirty"
wt_path = os.path.join(self.branches_dir, "fix-issue-850-rec-dirty")
subprocess.run(["git", "-C", self.repo_dir, "worktree", "add", "-b", branch, wt_path], check=True, capture_output=True)
dirty_file = os.path.join(wt_path, "dirty.txt")
with open(dirty_file, "w") as f:
f.write("uncommitted work")
journal = {
"idempotency_key": "test_dirty_rec",
"issue_number": 850,
"branch_name": branch,
"worktree_path": wt_path,
"artifacts_created": {
"worktree_dir_created": True,
"worktree_registered": True,
},
"failure_reason": "test dirty recovery",
}
rec = author_issue_bootstrap.run_compensating_recovery(journal, self.repo_dir, journal_dir=self.lock_dir)
self.assertTrue(os.path.exists(wt_path))
self.assertIn(f"worktree_path_preserved_dirty:{wt_path}", rec["rolled_back"])
def test_branch_with_commits_preserved_during_recovery(self):
"""F-4: Compensating recovery does not delete branch with author commits."""
branch = "fix/issue-850-rec-commits"
subprocess.run(["git", "-C", self.repo_dir, "branch", branch, self.master_sha], check=True, capture_output=True)
# Add a commit on the branch
wt_path = os.path.join(self.branches_dir, "fix-issue-850-rec-commits")
subprocess.run(["git", "-C", self.repo_dir, "worktree", "add", wt_path, branch], check=True, capture_output=True)
cfile = os.path.join(wt_path, "commit.txt")
with open(cfile, "w") as f:
f.write("author commit")
subprocess.run(["git", "-C", wt_path, "add", "commit.txt"], check=True, capture_output=True)
subprocess.run(["git", "-C", wt_path, "commit", "-m", "author commit"], check=True, capture_output=True)
subprocess.run(["git", "-C", self.repo_dir, "worktree", "remove", "--force", wt_path], check=True, capture_output=True)
journal = {
"idempotency_key": "test_commits_rec",
"issue_number": 850,
"branch_name": branch,
"resolved_base_sha": self.master_sha,
"artifacts_created": {
"branch_created": True,
},
"failure_reason": "test commit branch recovery",
}
rec = author_issue_bootstrap.run_compensating_recovery(journal, self.repo_dir, journal_dir=self.lock_dir)
branch_check = subprocess.run(["git", "-C", self.repo_dir, "rev-parse", "--verify", branch], capture_output=True, text=True, check=False)
self.assertEqual(branch_check.returncode, 0, "Branch with commits was deleted!")
self.assertIn(f"branch_preserved_commits:{branch}", rec["rolled_back"])
def test_task_capability_map_integration(self):
"""Verify task_capability_map has bootstrap_author_issue_worktree configured correctly."""
self.assertEqual(task_capability_map.required_role("bootstrap_author_issue_worktree"), "author")
self.assertEqual(task_capability_map.required_permission("bootstrap_author_issue_worktree"), "gitea.branch.create")
self.assertTrue(task_capability_map.preflight_task_matches("work_issue", "bootstrap_author_issue_worktree"))
self.assertTrue(task_capability_map.preflight_task_matches("bootstrap_author_issue_worktree", "lock_issue"))
if __name__ == "__main__":
unittest.main()
-15
View File
@@ -28,21 +28,6 @@ class TestPathUnderBranches(unittest.TestCase):
amw.is_path_under_branches("/repo/other-checkout", self.ROOT)
)
def test_unrelated_branches_dir_fails(self):
self.assertFalse(
amw.is_path_under_branches("/tmp/branches/evil", self.ROOT)
)
def test_prefix_confusion_fails(self):
self.assertFalse(
amw.is_path_under_branches(f"{self.ROOT}/branches-other/foo", self.ROOT)
)
def test_traversal_fails(self):
self.assertFalse(
amw.is_path_under_branches(f"{self.ROOT}/branches/../evil", self.ROOT)
)
class TestAssessAuthorMutationWorktree(unittest.TestCase):
ROOT = "/repo/Gitea-Tools"
+146
View File
@@ -0,0 +1,146 @@
"""Tests for the MCP restart-path inventory and guards (#657).
Covers:
* the registry is well-formed and every path is classified;
* unknown restart attempts fail closed (AC "fail closed on unknown restart");
* the previously-unguarded full-restart primitives stay guarded/absent
against the real source tree (AC "tests for at least one previously
unguarded path");
* pkill of the daemon is still classified as contamination (#630, AC3);
* the inventory doc and module stay in lock-step.
"""
import os
import tempfile
import unittest
from pathlib import Path
import mcp_restart_paths as rp
import runtime_recovery_guard
REPO_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
DOC_PATH = os.path.join(REPO_ROOT, "docs", "mcp-restart-path-inventory.md")
class TestRegistryWellformed(unittest.TestCase):
def test_registry_is_wellformed(self):
# Must not raise.
rp.assert_registry_wellformed()
def test_every_path_has_valid_classification(self):
for path in rp.iter_restart_paths():
self.assertIn(path.classification, rp.VALID_CLASSIFICATIONS)
self.assertTrue(path.guard.strip(), path.path_id)
self.assertTrue(path.references, path.path_id)
self.assertTrue(path.locations, path.path_id)
def test_ids_are_unique(self):
ids = [p.path_id for p in rp.iter_restart_paths()]
self.assertEqual(len(ids), len(set(ids)))
def test_covers_every_classification(self):
present = {p.classification for p in rp.iter_restart_paths()}
self.assertEqual(present, set(rp.VALID_CLASSIFICATIONS))
class TestUnknownAttemptFailsClosed(unittest.TestCase):
def test_unknown_path_raises(self):
with self.assertRaises(rp.UnknownRestartPathError):
rp.assert_restart_attempt_registered("totally_novel_restart_hack")
def test_get_unknown_raises(self):
with self.assertRaises(rp.UnknownRestartPathError):
rp.get_restart_path("nope")
def test_registered_attempt_returns_path(self):
path = rp.assert_restart_attempt_registered("manual_daemon_kill")
self.assertEqual(path.classification, rp.CLASS_FORBIDDEN)
class TestDaemonNeverSelfReplaces(unittest.TestCase):
"""Previously-unguarded full-restart primitive: daemon self-replacement."""
def test_no_self_replacement_in_source(self):
# The live daemon modules must contain no os.execv/os.kill/os._exit
# self-restart call. Must not raise.
rp.assert_no_daemon_self_replacement(REPO_ROOT)
def test_scanner_flags_injected_violation(self):
# Guard the guard: prove the scanner catches a real self-replace call.
with tempfile.TemporaryDirectory() as tmp:
bad = Path(tmp) / "gitea_mcp_server.py"
bad.write_text(
"import os\n"
"def restart():\n"
" os.execv('/usr/bin/python', ['python'])\n",
encoding="utf-8",
)
found = rp.scan_daemon_self_replacement(tmp)
self.assertTrue(found)
with self.assertRaises(AssertionError):
rp.assert_no_daemon_self_replacement(tmp)
def test_scanner_ignores_comment_and_docstring_mentions(self):
with tempfile.TemporaryDirectory() as tmp:
ok = Path(tmp) / "gitea_mcp_server.py"
ok.write_text(
"import os\n"
"# NOT os.execv() to re-point the interpreter here.\n"
'"""Never calls os._exit to restart."""\n'
"value = 1\n",
encoding="utf-8",
)
self.assertEqual(rp.scan_daemon_self_replacement(tmp), [])
class TestLegacyAutoRestartHelperRemoved(unittest.TestCase):
"""Previously-unguarded full-restart path: _trigger_mcp_auto_restart."""
def test_helper_absent_in_source(self):
# Must not raise: helper was removed in #685.
rp.assert_auto_restart_helper_absent(REPO_ROOT)
def test_scanner_flags_reintroduced_helper(self):
with tempfile.TemporaryDirectory() as tmp:
bad = Path(tmp) / "mcp_server.py"
bad.write_text(
"def _trigger_mcp_auto_restart():\n return True\n",
encoding="utf-8",
)
with self.assertRaises(AssertionError):
rp.assert_auto_restart_helper_absent(tmp)
class TestPkillStaysForbidden(unittest.TestCase):
"""AC3: pkill of the daemon remains forbidden/contaminating (#630)."""
def test_manual_daemon_kill_registered_as_forbidden(self):
path = rp.get_restart_path("manual_daemon_kill")
self.assertEqual(path.classification, rp.CLASS_FORBIDDEN)
def test_pkill_classified_as_contamination(self):
assessment = runtime_recovery_guard.assess_recovery_command(
"pkill -f mcp_server.py"
)
self.assertTrue(assessment["contaminated"])
def test_read_only_probe_not_contamination(self):
assessment = runtime_recovery_guard.assess_recovery_command(
"ps aux | grep mcp_server"
)
self.assertFalse(assessment["contaminated"])
class TestInventoryDocInSync(unittest.TestCase):
def test_doc_exists(self):
self.assertTrue(os.path.exists(DOC_PATH), DOC_PATH)
def test_doc_mentions_every_path_id(self):
with open(DOC_PATH, encoding="utf-8") as handle:
doc = handle.read()
for path in rp.iter_restart_paths():
self.assertIn(path.path_id, doc, f"doc missing {path.path_id}")
if __name__ == "__main__":
unittest.main()
@@ -139,8 +139,6 @@ EXPECTED_ROLE_EXCLUSIVE_TASKS = frozenset(
"gitea_release_merger_pr_lease",
"create_branch",
"push_branch",
"bootstrap_author_issue_worktree",
"gitea_bootstrap_author_issue_worktree",
# #812 AC20: publishing an unpublished local head is author-only for the
# same reason every other push is — it writes a branch to the remote.
"publish_unpublished_branch",