model-writtenrun 3480 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed replacing make's distribution-copying script with git archive in #1, preserving gzip and bzip2 outputs while making release archives reproducible and directly traceable to Git; the Meson equivalent used the custom name pgdist because dist was reserved.
- Junwang Zhao reported archives appearing in the source directory and suggested commit identifiers in filenames in #2; Peter left filename customization manual in #3, while Tristan Partin supplied a tested absolute-output-path fix and identified missing-program and dirty-check dependency problems in #4.
- Peter explained in #6 that standard meson dist introduced reproducibility and compatibility problems through compression and archive repacking; Tristan pursued upstream improvements in #8, and Peter supported avoiding unnecessary repacking in #9.
- Peter incorporated path fixes and make vpath support, added cross-platform distribution testing, and discovered that the FreeBSD CI copy omitted .git in #10; after Tristan identified a missing-bzip2 configuration failure in #11, Peter changed it to fail only when distribution generation was requested in #12.
- Eli Schwartz questioned the archive-metadata and reproducibility requirements in #14 and challenged treating reproducibility as a replacement for signing in #16; Peter defended independently reproducible source archives in #15, and Robert Haas supported keeping that limited improvement separate from signing-policy changes in #17.
- Peter added core.autocrlf=false in #18 after Windows CI produced converted line endings, reporting matching cross-platform archives except for a suspected old-Git gzip difference on Linux; after Tristan questioned upstream applicability in #19, Peter explained in #20 that the same configuration could make standard Meson archives unusable on Unix.
- Peter removed the buggy dirty-tree checks in #21, citing Meson problems and unexplained Linux CI failures, and split the FreeBSD copy correction into its own patch; continuous distribution testing remained optional because its runtime and artifact-storage costs still needed consideration.
- Tristan reviewed the split series in #22, recommended disabling Git-dependent targets when Git was absent and preventing accidental standard meson dist use, and approved the FreeBSD correction; Peter committed that correction unchanged as fc2d260c7e.
- Peter posted the remaining series in #23 with the Meson safeguards, retaining command-local line-ending control and allowing git archive itself to report a missing repository; Tristan endorsed this version with minor cleanup suggestions in #24.
- Peter committed archive generation from #23 as 619bc23a1a, incorporating the cleanup plus phony make archive targets, retention of distdir-location, and gzip level -9, and announced completion in #25; the accompanying CI distribution-testing patch was not included.
- Andres Freund reported a Meson minimum-version warning after the commit in #26; Peter proposed guarding the dist script by Meson version in #27, Andres confirmed that approach in #28, and Tristan offered a literal-program-name alternative while favoring the version guard in #29, but no applied fix is shown in this thread.
- At the last message, Tristan reported an upstream Meson improvement merged for version 1.5 and planned further work on archive repacking in #30; both PostgreSQL changes had landed, while the broader CI testing patch and the reported Meson compatibility warning had no recorded resolution here.