model-writtenrun 7335 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andrew Dunstan opened the thread after migrating the two buildfarm animals running TAP tests under MSYS to a Perl setup that no longer needed perl2host, proposing an inline one-line disabling patch followed by complete removal if nothing broke #1.
- Andres Freund supported disabling perl2host but requested a diagnostic for MSYS Perl targeting native Windows; he opposed rejecting MSYS Perl universally, while Andrew favored checking the actual Perl invoked by prove in config/check_modules.pl #2 #4 #6.
- The preliminary inline change landed as f1ac4a74de, making perl2host a no-op so the buildfarm could expose any remaining dependency before its removal #1.
- Andres explained that MSYSTEM could disagree with the configured host_os, so Andrew agreed to export host_os for the Perl checking script, and Andres accepted that design #11 #12 #13.
- Andrew attached the configure-time compatibility check, allowing MSYS Perl only for an MSYS target #14.
- Andrew's experiment with an MSYS-targeted server failed during bootstrap because shared memory was unavailable; after identifying the absence of cygserver, he declined further work on that target and considered buildfarm coverage impractical #15 #16.
- With no failures from disabling perl2host, Andrew posted and reported successful testing of two cleanup patches: complete removal of perl2host and its callers, followed by removal of most remaining MSYS-specific TAP processing #17.
- Andrew additionally identified removable skip logic in src/bin/pg_dump/t/010_dump_connstr.pl, but that change is absent from the supplied candidate commits #18.
- Andres raised the risk that future backpatched fixes would break older branches without the cleanup; Andrew committed to backpatching it despite TAP-test differences, and Andres agreed #19 #20 #21.
- The compatibility check from #14 landed unchanged apart from rebasing as cf12541f2b; the first cleanup patch from #17 landed unchanged as 95d981338b, and the second landed as 1c6d462939 with the pg_ctl test using command_like instead of command_like_safe.
- Although thread #45395 supplied earlier discussion motivating removal, 95d981338b comes from the later matching attachment in #17, superseding the other thread's attribution to discussion alone.
- Both cleanup changes were backpatched through release 10, and all attached work had landed when Thomas Munro closed the thread by welcoming the improvement; his response came after the commits and therefore supplies no pre-commit review credit 95d981338b 1c6d462939 #22.