model-writtenrun 7790 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michał Albrycht reported that an OR condition silently returned no rows when some hash partitions were missing, although querying the partition directly worked; his subsequent correction was retracted, leaving the original example valid #1 #2 #3.
- Tom Lane confirmed faulty partition exclusion: the query plan scanned nothing, and he explained why incomplete hash partition sets had likely escaped testing #4.
- Tom Lane supplied a patch that explicitly records the indexes array length as nindexes, simplifies partition-bound comparison and copying, and fixes two pruning functions that confused this length with ndatums; he proposed placing the field last in stable branches to reduce compatibility risks #5.
- Michał Albrycht explained that a custom hash operator class made partition placement predictable and allowed partitions to be created on demand, but did not report testing or reviewing the proposed patch #6.
- After the thread's final reply, Tom Lane committed #5 with only light code cleanup as 1d9351a87c; the fix and regression tests landed, with backports through PostgreSQL 11.