model-writtenrun 7957 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed row-type hashing in #1, combining individual field hashes using the array-hashing approach to enable UNION DISTINCT, hash joins, and hash partitioning.
- Andres Freund warned in #2 that fields without hash functions could make previously successful queries fail, and Peter agreed in #3 to integrate the necessary type-cache checks.
- Robert Haas asked about modified operator-class and operator-family definitions in #4; Tom Lane replied in #5 that such modifications could already break many places and did not warrant special handling here.
- Peter's revision #6 checks field hashability before treating a defined composite type as hashable, while retaining the acknowledged possibility of runtime failures for anonymous records.
- Tom reviewed the revised implementation in #7 and requested stronger evidence that tests exercise hashing, including a non-hashable-field case; Peter responded with the independent test patch #8, using money as a non-hashable type and explaining that recursive UNION requires hashing even though its plan does not display that requirement.
- Tom acknowledged that his concerns were addressed in #9, and the separate tests from #8 landed unchanged as 6dd8b00807.
- The feature from #6 then landed as 01e658fa74, with clearer error wording, a catalog-version bump, and expanded test coverage that changes the newly installed expectations to demonstrate record hashing; Peter confirmed both commits in the final message #10, leaving no patch pending.