model-writtenrun 8551 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Fujii Masao proposed #1 to prevent descriptor and memory leaks after LZ4 finalization or final-write failures, also saving the pathname before cleanup to prevent use-after-free during error reporting.
- Chao Li questioned why the new cleanup bypassed rename, unlink, and sync in #2; Fujii explained in #3 that preserving existing failure behavior kept the fix focused on releasing resources.
- Michael Paquier supported backpatching in #4 because repeated failures in service-like operation could accumulate file descriptors.
- Chao identified temporary files left behind when CLOSE_UNLINK encounters an LZ4 failure in #5, and Xuneng Zhou approved the resource-leak fix while also questioning skipped unlink operations in #6.
- Fujii proposed handling unlink-on-error separately, including close() and gzclose() failures, in #7; Chao offered to prepare that follow-up in #8, Fujii encouraged him in #9, and Chao accepted in #10.
- Fujii committed #1 without code changes as 0cb41ee807, with backpatches through PostgreSQL 15, and announced the push in #11; the broader unlink cleanup remained separate follow-up work with no patch posted in this thread.