model-writtenrun 8347 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane opened the discussion by demonstrating that posixrules-based DST handling fails after 2038, noting upstream's intention to retire the feature and considering removal of its loading logic while retaining POSIX time-zone specifications #1.
- Nearly a year later, Tom revived the issue because upstream planned to stop installing posixrules by default, exposing installations using system tzdata to changed behavior; he proposed documenting the fallback and reconsidering removal #2.
- Tom supplied a documentation patch explaining the full POSIX syntax in an appendix #3; Robert Haas identified confusing explanation order and proposed replacement wording #4 #6, which landed with other light clarifications in a3235a53ae.
- Experiments removing posixrules exposed regression-test failures, initially leading Tom to propose replacing the tests' PST8PDT setting with America/Los_Angeles #8.
- Tom then traced the failures to the nonstandard CST7CDT specification, confirmed that PST8PDT correctly loads a named zone, and retained POSIX parser coverage by proposing explicit historical DST rules instead #10; 2c8ef9363d implemented that fix and accompanying documentation corrections across all supported branches.
- Peter Eisentraut favored removing support in master and possibly v13 while preserving released branches #9; Tom posted a complete removal patch that retained vestigial upstream structure to minimize maintenance differences and argued for predictable behavior across platforms #11.
- Robert Haas opposed forcing a potentially breaking change into released branches #12, while Tom argued that system tzdata updates could otherwise change behavior without PostgreSQL release notes #13.
- Investigation showed that Fedora/RHEL explicitly requested posixrules generation, reducing the immediate risk there #16, but upstream estimates of adoption elsewhere kept Tom concerned about older PostgreSQL branches #17.
- Peter argued that long-term-support distributions could preserve the feature #18; Tom remained concerned about platform-dependent semantics and the burden of retaining code that upstream might later remove #19.
- At the final message, Tom accepted limiting removal to master and v13 despite his remaining concerns about older branches #20; ea57e531b9 subsequently landed the removal patch #11 unchanged, completing the submitted work while leaving posixrules support in older releases.