model-writtenrun 3956 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jun - Jul 2023 · 3 entries
- Vik Fearing proposed SQL-standard AT LOCAL using the session's timezone, including use inside views, and supplied an implementation based on nested current_setting and timezone calls in #1.
- Laurenz Albe questioned whether timezone resolution happened at parse time in #2; Vik demonstrated execution-time behavior with a prepared statement in #3, while Alvaro Herrera raised the related partition-pruning concern in #5, without a failing example subsequently appearing.
- Daniel Gustafsson fixed a compiler-dependent switch-scope error and reflowed comments in #7, and Cary Huang later reported successful functional and CI testing of that version in #9.
Sep - Oct 2023 · 12 entries
- Michael Paquier argued that nested calls complicated deparsing and doubled executor overhead in #11 and #13, prompting Vik's dedicated unary timezone functions in #14; Michael measured an improvement from roughly 4.3 to 3.6 seconds in #15.
- Vik corrected the deparsing tests in #16, and Michael expanded the documentation and requested missing time-with-time-zone support in #17; Vik supplied that support, stable function declarations and corresponding tests in #18.
- Michael first extracted the independent timezone(zone, time) documentation paragraph from #18 into 7e1f544827, then landed AT LOCAL as 97957fdbaa with test presentation edits and a catalog bump, explaining both steps in #19.
- Thomas Munro reported malformed timetz results from the AIX xlc buildfarm animal in #22, and Tom Lane traced the likely problem to existing timezone-wrap arithmetic in #23, rather than the new syntax itself.
- Tom discovered that the feature's tests had exposed previously untested timetz_zone code and that timetz_izone remained uncovered in #25; Noah Misch then showed that upgrading xlc still produced the failure in #35.
- Tom proposed extending interval-timezone coverage in #39 and implemented it without an attached patch as 2f04720307; he confirmed the same failure in timetz_izone in #41.
- After reproducing the problem with the buildfarm configuration, Tom reported that inserting debugging output suppressed the failure while a local-variable rewrite did not in #42, strengthening the compiler-bug diagnosis.
- Michael proposed replacing the positive-wrap loop with guarded modulo arithmetic in #43; Tom retained the negative-wrap loop because C99 remainder semantics differ for negative inputs and supplied a tested fix with additional coverage in #45.
- Tom committed #45 unchanged as 19fa977311, then installed combined test coverage on older branches as 05b4614dec, deliberately separating the tests from the repair to observe failures before back-patching the code.
- The workaround remained necessary even with xlclang: Tom reported the same pre-fix arithmetic failure there in #51, while discussion shifted toward whether maintaining these compiler targets was worthwhile.
- Noah proposed dropping xlc support for PostgreSQL 17 onward while retaining AIX with GCC and older-branch xlc testing in #56, receiving support from Robert Haas, Tom Lane and Andres Freund in #57, #58 and #59.
- The feature and follow-up fixes had landed by the thread's end, but platform policy remained open: Tom's final message #60 raised dropping AIX entirely because of unresolved alignment assumptions and little apparent end-user demand.