model-writtenrun 7820 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Amul Sul questioned the need for CheckpointLock because holding it throughout a checkpoint suppressed interrupts and delayed global barrier handling needed for his read-write state change work #1.
- Bharath Rupireddy raised possible concurrency with standalone backends and suggested special interrupt handling, but Dilip Kumar explained that standalone operation has no competing checkpointer; Bharath then withdrew his objection #2 #3 #4 #5.
- Robert Haas supported removing the redundant lock, arguing that any newly exposed interrupt-safety problems should be fixed, and posted a complete removal patch after examining which ProcessInterrupts conditions could affect the checkpointer #6 #7.
- Tom Lane confirmed the concurrency rationale and traced the historical changes that centralized checkpoint and recovery-checkpoint execution, supplying context later incorporated into the commit message; Michael Paquier also expressed agreement #8 #9 #10 #11.
- Robert Haas committed #7 with its diff unchanged as d18e75664a, expanded the commit message with historical context, and announced that the removal had landed on master, leaving no outstanding revision in this thread #12.