model-writtenrun 5802 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed explicit UNIQUE NULLS DISTINCT and UNIQUE NULLS NOT DISTINCT options in #1, following the SQL:202x draft while preserving PostgreSQL's default behavior and adding analogous CREATE UNIQUE INDEX syntax.
- Marko Tiikkaja suggested choosing null treatment separately for each indexed column or expression in #2, but Peter rejected that extension in #3 because its bookkeeping outweighed its utility and would make index syntax inconsistent with constraint syntax.
- After several months, Peter returned with #4, rebasing the patch and completing pg_dump compatibility handling and psql test coverage; Zhihong Yu proposed positive flag names in #5, but the final implementation retained negative flags so false preserved existing behavior.
- Maxim Orlov reviewed #4 and reported passing tests on Linux and macOS in #7, finding the B-tree implementation reasonable and requesting a fresh rebase before marking it ready.
- Peter Geoghegan proposed handling the option when constructing insertion scan keys in #8, with a wording correction in #9, and Pavel Borisov agreed in #10; Peter implemented that change in #11 while retaining the anynullkeys name because alternatives seemed less clear.
- Maxim resent the unchanged final patch to restart cfbot in #12, and Pavel explicitly approved that version's code and explanatory comment after green CI results in #13.
- Peter committed the patch resent in #12 as 94aa7cc5f7, adding only a catalog-version bump beyond the submitted diff, and confirmed completion in #14; the proposed feature was fully landed with no remaining patch work identified.