model-writtenrun 148 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- shihao zhong reported that btree_gin timestamp index scans missed matching rows when a timestamptz bound fell in the first occurrence of the repeated autumn DST hour, and supplied a round-trip lower-bound adjustment plus regression tests in #1; spring-gap behavior was left outside the patch because ordinary btree indexes exhibited it too.
- Tom Lane argued in #2 that the adjustment assumed too much about timezone conversion and recommended repeatedly comparing against the original bound and subtracting an hour; he also questioned whether date conversion had the same problem.
- shihao zhong adopted the comparison loop in #3, retained the tests, and reported finding no date-conversion failure across backward clock changes in tzdata, explaining that failure would require a rollback exceeding a whole day.
- Tom Lane committed #3 as dbb9bac594, adding positive-infinity handling before the loop, expanding the explanation, and adapting the regression tests to the existing table; the fix was backpatched through PostgreSQL 19.
- In #4, Tom Lane explained the infinity correction and confirmed the push, leaving cvt_timestamptz_date unchanged because no failure case had been demonstrated; the submitted timestamp fix and tests had landed.