model-writtenrun 7326 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Mar 2022 ยท 15 entries
- The thread began with Robert Haas's announcement of the basebackup_to_shell module, which sends backup archives and manifests to an administrator-configured command; Andres Freund requested basic regression coverage, leading Robert to attach a TAP test in #4 after the exchange in #1, #2, and #3.
- Robert found that Windows CI did not automatically run contrib TAP tests, so Thomas Munro supplied an explicit CI invocation in #8 and identified broken Windows path escaping in its output in #10.
- Difficulty identifying similarly named TAP tests also prompted a separate improvement: Robert requested directory names in test output in #6, Andres illustrated banners in #13, and Robert and Daniel Gustafsson endorsed them in #14 and #15 before Andres attached coverage for TAP, regression, and isolation tests in #17.
- Robert's next test version escaped Windows paths and added missing required_role documentation after Joe Conway reported the omission elsewhere; Thomas resent that series for another Windows run in #22 and #23.
- The next Windows run timed out, as Thomas reported in #24; Andres and Andrew Dunstan explained that type con reads the console rather than standard input in #26 and #27, leading to the gzip --fast implementation attached in #30 after Andrew's suggestion in #29.
- Andres confirmed that the gzip version passed Windows CI in #32 and suggested simplifying its path separators, but Robert retained the established platform-specific style and left further cleanup to someone with a local Windows environment in #38.
- For the test-output banners, Tom Lane requested separate check and installcheck labels and favored changing only the development branch in #33; Andres proposed the resulting wording in #35, Tom accepted it in #37, and Thomas supported the improvement while favoring broader backpatching in #41.
- Robert landed the documentation component of #30 unchanged as 26a0c025e2 and its TAP-test component unchanged as 027fa0fd72, leaving the temporary CI workaround outside those commits; the documentation commit also links to Joe Conway's omission report in thread #45017, which supplied no matching documentation patch.
- Nathan Bossart then supplied a .gitignore for generated test directories in #46, and Tom committed it unchanged as f8e0d900af, explaining in #47 that it would prevent accidental commits of test output.
- Tom reported Windows buildfarm SSPI failures in the landed test in #48; Andres explained in #49 that CI used Unix sockets and therefore bypassed the authentication setup exposing the problem, while Andrew confirmed that switching two buildfarm animals to Unix sockets made them pass in #53.
- Robert relayed Neha Sharma's pointer to auth_extra in #54, and discussion established that pg_regress --create-role configures authentication without actually creating the database role, so the test must also retain CREATE USER, as Robert demonstrated in #57.
- Andrew attached an authentication fix in #59, but Robert found a Perl syntax error and an incorrect replacement of CREATE USER with ALTER USER in #60; Tom subsequently agreed with retaining role creation alongside auth_extra in #61.
- Robert committed the corrected authentication setup, preserving CREATE USER and adding explanatory comments, as fea1cc49e4 and announced it in #63; Andrew noted in #64 that the next non-Unix-socket buildfarm run would provide confirmation.
- Andres landed the banner patch from #17 with the agreed check/install-check wording as 28bdfa2adf and confirmed the push in the thread's final message, #66.
- All five candidate changes had landed by the end of the thread, while final buildfarm confirmation of the authentication fix was still pending and Andrew had deferred cleanup of the confusing authentication infrastructure until after feature freeze in #64 and #65.