Repository navigation
fix: fail account sync when backup fails - #127
Conversation
|
🦞👀 Pull request received. I will update this pull request when review starts. |
|
Codex review: needs maintainer review before merge. Reviewed August 18, 2026, 7:40 PM ET / 23:40 UTC. ClawSweeper reviewWhat this changesThe PR makes scheduled account sync report backup-export failures and adds timestamp-aware reconciliation for imported profile identities, snapshots, affiliations, and bio entities. Merge readinessNo actionable patch defect was found. This owner-authored PR is cleanly mergeable with passing CI and targeted regression coverage; it remains open for routine maintainer merge review because it intentionally changes scheduled-job failure signaling and backup-import conflict behavior. Likely related people: steipete (feature owner and recent area contributor, high confidence). Priority: P2 Review scores
Verification
Live VerificationCommand: Result: FAIL (failed) — execution before step 1 Assertions:
How this fits togetherScheduled account sync refreshes local X data and then runs configured backup synchronization, returning an audit result to launchd or other schedulers. Backup import merges portable JSONL profile data into SQLite, where profile identity and history rows feed search and downstream reads. flowchart LR
A[Scheduled account sync] --> B[Live account refresh]
B --> C[Backup auto-sync]
C --> D[Job audit and exit status]
E[Backup JSONL profiles] --> F[Profile recency and identity checks]
F --> G[SQLite profile merge]
G --> H[Search and profile reads]
Before merge
Agent review detailsSecurityNone. Review metrics
Merge-risk optionsMaintainer options:
Technical reviewBest possible solution: Keep backup failures visible to schedulers while preserving local profile state unless newer imported snapshot evidence and proven identity links justify an update. Do we have a high-confidence way to reproduce the issue? Yes—on the pre-PR path, account sync calculated success from sync steps alone despite a failed backup result; the focused job and CLI regressions exercise that path, and import tests cover the affected profile states. Is this the best way to solve the issue? Yes—the patch keeps the failure signal at the job boundary and uses existing snapshot history for import decisions rather than introducing a separate backup mode or data store. AGENTS.md: not found in the target repository. Codex review notes: model internal, reasoning high; reviewed against 1b94a3fb8dc8. LabelsLabel justifications:
EvidenceWhat I checked:
Likely related people:
Rating scale
Overall follows the weaker of proof and patch quality. Workflow
HistoryReview history (3 earlier review cycles) |
4ef618b to
e33b509
Compare
Summary
profile_merows into the canonical numeric profile after imported references existProof
Live discovery
Production rehearsals completed all six data steps and correctly exited 1 while surfacing two backup-import bugs: stale handle ownership and a legacy
profile_mecollision with the canonical numeric selected account. In both cases the scheduler stayed unloaded, no lock remained, and the clean backup checkout was unchanged. This PR fixes the status propagation and both import paths before scheduling is re-enabled.