model-writtenrun 6149 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- John Naylor proposed marking the timestamptz variant of date_bin() STABLE because of possible timezone dependence and asked whether PostgreSQL 14 should also receive a catalog-version bump; Tom Lane confirmed that the bump was required #1 #2.
- John committed the patch unchanged as 543f36b43d, with a PostgreSQL 14 backport, and reported completion in #3; Tom subsequently recommended using distinct catalog-version numbers across branches #4.
- Aleksander Alekseev challenged the change because timestamptz_bin() depends only on its inputs, and Tom's experiments confirmed timezone independence while exposing that fixed 24-hour bins can cease to align with local midnight across DST boundaries #5 #7.
- Justin Pryzby's timezone examples prompted Tom to distinguish different displays of the same timestamptz value from changes in the underlying result, also noting that Justin's origin inputs differed #8 #9.
- John demonstrated that casting the inputs to timestamp restores local-midnight alignment, and Tom recommended documenting this unintuitive behavior #10 #11.
- With no support expressed for changing the binning behavior, John chose documentation work and announced that he would restore the function's IMMUTABLE marking barring objections; the supplied material does not confirm that restoration landed #12.
- John submitted a separate documentation patch illustrating DST effects and the timestamp-casting workaround #13.
- Aleksander suggested a simpler general explanation inspired by TimescaleDB, which John initially accepted before deciding that concrete examples were easier to reason about; the thread ends with the documentation patch unresolved and no revised attachment or documentation commit shown #15 #16 #17 #18.