model-writtenrun 7201 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed removing distprep after the developer unconference, so release tarballs would stop carrying generated sources and documentation, simplify Meson builds, and improve source traceability; the initial patch made bison, flex, and perl mandatory while leaving documentation-tool behavior open #1.
- Andres Freund argued that removing config/missing and Meson's source-conflict safeguards exceeded the necessary scope #2, so Peter restored them, removed temporary header-check exclusions, and standardized generated-header links on relative paths #3.
- Alvaro Herrera found that cleanup rules moved into subdirectories still used parent-relative filenames #4; Peter corrected those paths and added explicit jsonpath_gram.h generation for header checks #5.
- Tom Lane requested packager feedback and objected to losing the ability to preserve generated parser files during distclean #6; Christoph Berg reported that Debian already built plain Git snapshots successfully #7, while Peter defended distclean restoring the distributed source state and invited a concrete alternative cleanup target #11.
- The packaging discussion distinguished repository-tracked configure from files added only to tarballs, leaving changes to Autoconf policy outside this work #8 #10.
- Peter fixed remaining Cirrus header-check failures by adding parser/gram.h to generated-headers and rebased over intervening wait-event changes #17.
- Michael Paquier found that gettext-files now tried to build without required generated headers and identified obsolete tarball documentation #18; Andres requested splitting mandatory tool requirements from the Makefile changes to ease review #19.
- Peter introduced generated-parser-sources for translation extraction, verified that generated POT files remained identical, corrected the documentation, and posted the two-part series #20; Michael endorsed the fix #21, and Peter confirmed that separately distributed HTML documentation was the intended direction #22.
- Andres tested the final series successfully but questioned whether the maintainer-clean alias preserved extension recursion behavior #24 #26; Peter judged concrete extension breakage unlikely and deferred additional cleanup targets for lack of a specific proposal #27, with Andres agreeing to proceed #28.
- Peter committed both parts of the final series #20 together, with unchanged patch content apart from rebasing, as 721856ff24, and announced completion #29.
- Alvaro subsequently questioned the separate README.git and README's reference to a missing INSTALL file #31; Tom proposed dropping generated plain-text installation instructions and linking to the website #32, which Andrew Dunstan supported #33.
- Tom's proposal was subsequently implemented through the matching patch #549553 in thread #48923 and committed as e2b73f4a4d, removing INSTALL generation, its supporting documentation machinery, and README.git while adding online documentation links; this thread supplied the earlier design discussion rather than the committed attachment.
- Months later, Noah Misch reported that the simplified header-link recipes broke MinGW builds using LN_S='cp -pR' and that restoring the old recipes fixed them #34; Alvaro supported the reversal #35, and Noah confirmed that only two rules needed it because their destination-path semantics differed from the other link commands #37.
- Peter restored those two absolute-path recipes in 58445651db and announced the fix in the final message #38, leaving distprep removal and the INSTALL cleanup landed with the reported MinGW regression resolved.