model-writtenrun 8911 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Jehan-Guillaume de Rorthais reported a logical-replication worker crash with assertions enabled and supplied a reproducer in #1; Tom Lane suggested freed-memory clobbering was exposing the fault in #2, while Kyotaro Horiguchi could not reproduce it in #3.
- Jehan confirmed that memory clobbering exposed the crash and reported Valgrind evidence of an uninitialized attribute-map value in #4, also identifying a separate allocation that used sizeof(int) instead of sizeof(AttrNumber).
- Peter Eisentraut fixed the allocation size in d40abd5fcf and announced it in #5, while asking Jehan to continue investigating the crash.
- After confirming the crash still reproduced under sufficient write activity in #6, Jehan traced it to relation locking processing pending invalidations after the attribute-map validity check and posted a fix that rebuilt the map after opening the relation in #7.
- Amit Kapila found that removing locking from RangeVarGetRelid could lock the wrong relation in #8; Vignesh C restored that locking while retaining the delayed map rebuild in #9, and Jehan acknowledged the correction in #12.
- Amit documented reliable debugger-assisted reproduction and polished the patch in #10; Peter agreed to omit an automated regression test because the race was difficult to reproduce reliably and flagged commit-message wording in #11.
- Jehan proposed clearer commit-message wording in #13, which Amit incorporated without code changes in #14.
- Vignesh supplied and tested the PostgreSQL 10/11 adaptation using heap_open in #15, explicitly retaining Amit's preceding version for PostgreSQL 12 and development.
- Amit committed #14 with only comment punctuation changed as 04c8a69c0c, backpatched the fix through PostgreSQL 10, and announced completion in #16.
- The thread ended with Jehan's thanks in #17 and both fixes landed; his separate question about freeing attrmap before rebuilding it in #7 received no explicit resolution in the thread.