* fix: resolve remaining 11 CI test failures (#2982)
Two categories of fixes:
1. Flock contention in test setup helpers: bdCreate, bdCreateSilent,
bdUpdate, bdDelete, bdClose, and bdDepAdd now retry up to 5 times
with exponential backoff when they encounter "one writer at a time"
flock errors. This handles transient contention when parallel tests
within the same CI shard compete for the embedded Dolt lock during
their setup phase.
2. macOS path symlink mismatch: TestContext* tests in
explicit_db_nodb_test.go now use filepath.EvalSymlinks() when
comparing beads_dir paths, fixing failures where macOS resolves
/var/folders to /private/var/folders.
Fixes#2982
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: increase flock retry budget and add retry to read helpers
The initial 5-retry/100ms-base backoff was insufficient for CI
contention. Refactored to a shared bdRunWithFlockRetry helper with
10 retries and 500ms base backoff (~30s total budget).
Also added flock retry to bdDiff, bdDiffJSON, bdShowRaw, and
bdShowDetails which were also failing on CI due to flock contention
during parallel test execution.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: close store before subprocess to release flock
The root cause of the persistent flock failures was openStore()
deferring Close() to t.Cleanup, keeping the exclusive flock held
while bd subprocess commands tried to acquire it. No amount of
retry can help when the same test holds the lock.
- getCommitHash: open/close store inline instead of via openStore
- show_json_includes_comments: close store after AddIssueComment
- show_as_of: close store after GetCurrentCommit
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>