fix(update): clear a stale .git/index.lock before Update Code (#801)

* fix(update): clear a stale .git/index.lock before Update Code

A git that subprocess.run's timeout kills (SIGKILL) cannot remove
.git/index.lock, so every later update failed with "Unable to create
.../.git/index.lock: File exists" until the file was removed by hand.

Update Code and the automatic update's rollback now remove the lock when
no git process is running in the checkout and the file is more than a
few seconds old. While another git is running, Update Code says so
instead of failing; when the lock could not be checked, the error names
the rm command that fixes it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SvFtgthKUNEPhJu6RuX9Sk

* fix(update): count a git that names the checkout in its arguments as running

git --git-dir/--work-tree can work on the checkout from any working
directory, so the cwd check alone could take such a git's live lock for
a stale one.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SvFtgthKUNEPhJu6RuX9Sk

* fix(update): resolve a git's relative path arguments against its cwd

`git -C LEDMatrix ...` or a relative --git-dir started outside the
checkout works on it too; resolve each argument against the process's
working directory before deciding a lock is stale, and count unreadable
arguments as naming the checkout.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SvFtgthKUNEPhJu6RuX9Sk

---------

Co-authored-by: Claude <noreply@anthropic.com>
This commit is contained in:
Chuck
2026-10-26 20:13:40 -04:00
committed by GitHub
co-authored by Claude Opus 5.5
parent 4c883ec2c2
commit fa51ebab5f
6 changed files with 389 additions and 2 deletions
+23
View File
@@ -9,9 +9,11 @@ Git runs for real against a throwaway origin/device clone pair; systemd, sudo
and the health check service are faked.
"""
import json
import os
import shutil
import subprocess
import sys
import time
from datetime import datetime, timezone
from pathlib import Path
from unittest.mock import MagicMock, patch
@@ -579,6 +581,27 @@ class TestUpdateIsVerified:
assert repo.head() == old
assert 'rolled back' in result['core_message']
@pytest.mark.skipif(not Path('/proc/self/comm').exists(), reason='needs /proc')
def test_a_pull_killed_by_its_timeout_is_still_rolled_back(self, tmp_path):
"""subprocess.run SIGKILLs a timed-out git, which leaves .git/index.lock
behind; the rollback's reset would fail on it."""
repo = Repo(tmp_path)
old = repo.head()
repo.publish()
pull = real_pull(repo.device)
lock = repo.device / '.git' / 'index.lock'
def killed(**kwargs):
pull()
lock.write_text('')
stale = time.time() - 60
os.utime(lock, (stale, stale))
raise subprocess.TimeoutExpired(['git', 'pull'], 60)
result = Harness(tmp_path, repo, core_update=killed).updater.run()
assert repo.head() == old
assert 'rolled back' in result['core_message']
assert not lock.exists()
@pytest.mark.parametrize('how', ['partial_pull', 'check_never_started'])
def test_a_failed_rollback_leaves_plugins_and_the_display_alone(self, tmp_path, how):
"""update_core returns rollback_failed itself on two paths. The device