model-writtenrun 8900 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Ranier Vilela reported two always-false hash-table checks in buildSubPlanHash(), including a hidden wrong-table reset, and an apparently unreachable publication-actions cleanup in GetRelationPublicationActions() #1.
- Andreas Karlsson supplied a patch that preserves the subplan hash tables for reuse, resets hashnulls correctly, and clears that pointer when no nulls table is needed #2.
- Andreas Karlsson separately proposed removing the relcache cleanup after concluding that it had been unreachable since the function was introduced #3.
- Tom Lane opposed that removal because recursive or reentrant cache accesses could populate the entry, making the cleanup inexpensive protection against a permanent memory leak and consistent with nearby code #4.
- Andreas Karlsson accepted the relcache argument and said he would submit the remaining executor fix to the commitfest, while acknowledging uncertainty about its correctness; the thread ended without a landing announcement #5.
- The executor fix later landed with rescan regression tests as 58c47ccfff, attributed to the newer submission #443846 in thread #42007; the rejected relcache cleanup is absent from that commit.