model-writtenrun 6300 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Arne Roland proposed recursive trigger renaming because renaming a partitioned table's trigger left its child triggers unchanged, initially requiring matching names and parent links and rejecting discrepancies #1.
- Arne promptly restored the existing public renametrig interface by introducing a recursive helper, then corrected its comments #2 #3.
- Alvaro Herrera recommended identifying child triggers through tgparentid and initially proposed ON ONLY to preserve independent renaming, while also raising the alternative of forbidding clone-trigger renames #5.
- After a gap, Arne added ON ONLY support but questioned overwriting independently chosen child names; Alvaro argued that recursion should follow parent links regardless of names and requested multilevel-partition and legacy-inheritance tests #7 #8 #9 #11.
- Arne adopted parent-ID matching, added notices for differing names and SQL regression tests, and investigated detached-partition behavior #12 #13.
- Work resumed in June with agreement that the behavior change belonged in PostgreSQL 15 rather than stable branches; Arne refreshed the patch in June and July, adjusting detach tests and simplifying the implementation #15 #16 #18 #21.
- Alvaro rewrote the implementation to lock descendants upfront and separate single-trigger updates from partition traversal, reporting that Arne's tests passed and initially removing differing-name notices #22.
- Arne argued that notices mattered because trigger names determine firing order and reported a compilation failure caused by the assertion-variable annotation; Alvaro corrected the declaration, restored notices, improved name-conflict errors, and added statement-trigger tests #23 #24 #26 #27 #28.
- Examining pg_dump changed the design: Alvaro concluded that independently renamed clone triggers could not be preserved by dumps, and Arne agreed to discard ONLY support while suggesting upgrade-time normalization of existing inconsistencies #29 #30.
- Alvaro posted the final design with unconditional partition recursion, no ONLY syntax, and an error for direct clone-trigger renames, together with documentation and revised tests #33.
- Arne also proposed restrictions on enabling or disabling individual clone triggers, but Alvaro pointed to existing fixes for that separate behavior and defended per-partition enablement as a valid use case, leaving it outside this patch #32 #34 #35 #36.
- Alvaro committed #33 with minor wording and comment corrections as 80ba4bb383, retaining co-author credit for Arne and reviewer credit for Zhihong Yu; the behavior change was not backpatched #37.
- Immediately after landing, Alvaro reported a locale-sensitive sorting failure in the regression tests and said he would fix it, although this thread supplies no attached correction for that issue #38.
- Tom Lane relayed Coverity's complaint about the new counter/assertion and questioned its design; Alvaro responded with 21b3aa9c8f, removing the assertion and stopping after the first matching child trigger #39 #40.
- The thread ended with both candidate changes landed and no replacement patch pending, while Tom and Arne continued discussing whether duplicate child-trigger links warranted a runtime corruption error; a proposed uniqueness constraint was not pursued because it would require unsupported partial catalog indexes #40 #41 #42.