model-writtenrun 7268 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tomas Vondra opened the thread with a reproducer showing that HOT updates could leave BRIN summaries stale and make index scans miss rows, explaining that BRIN did not consult visibility information as the original optimization assumed #1.
- Matthias van de Meent proposed distinguishing summarizing indexes from indexes that block HOT, allowing HOT chains while still forwarding changed values to summarizing indexes; Tomas considered this viable but too complicated for the late PostgreSQL 15 development stage #2 #3.
- Matthias supplied an implementation with amsummarizing, three-way index-update signaling, and a regression test; he noted its ABI changes and left updating only the individually affected summarizing indexes as a future optimization #4.
- Andres Freund and Robert Haas favored reverting for PostgreSQL 15 because new infrastructure needed more analysis and testing, while agreeing that the underlying optimization could be salvaged #5 #6.
- Tomas posted a revert and asked whether to retain an unused access-method field for ABI compatibility; Michael Paquier supported removing it during beta #7 #8.
- The revert landed unchanged as e3fcca0d0d from #7, and Tomas announced that the optimization would be retried for PostgreSQL 16 #9.
- After an eight-month gap, Matthias revived the optimization by combining the original implementation with his repair, adding comments and regression coverage for PostgreSQL 16 #10.
- Tomas questioned temporarily repurposing a tuple-header bit and requested interface and documentation cleanup; Matthias replaced that signaling with an output parameter and allowed summarizing-index predicate columns to participate in HOT updates through the summarized-attribute bitmap #11 #12.
- Tomas supplied a relcache simplification and asked about predicate regression coverage; Matthias incorporated the simplification and corrected the description of an existing predicate test #13 #14.
- After Matthias marked the patch ready, Tomas supplied a final review series addressing comments, documentation, assertions, and formatting; Matthias agreed and proposed clearer assertion and bitmap-comment wording #15 #16 #17.
- Tomas committed the series from #16 as 19d8e2308b, with light assertion, comment, enum, and commit-message cleanup, then announced the successful landing #18; this thread supplies the latest matching patchset, superseding the earlier implementation in thread #44477.
- The thread ended with thanks from Matthias and Josef Šimánek after the repaired optimization landed, with no attached patches still awaiting action #19 #20.