model-writtenrun 8524 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund identified that PG_TEST_INITDB_EXTRA_OPTS disables initdb-template reuse, reported substantial extra CPU costs, and explored ways to configure CI servers without repeatedly running initdb in #1.
- Tom Lane proposed recording the options used to create the template and reusing it when they match in #2; Jelte Fennema-Nio implemented that mechanism in #3, reporting improved Linux CI timings but no clear macOS improvement.
- Andres Freund argued that template-option matching was unnecessarily complicated for CI's server-only settings and proposed a separate environment variable in #4; Nazir Bilal Yavuz implemented PG_TEST_SERVER_POSTGRES_OPTS in #5, adding configuration-file handling, documentation and CI changes.
- Jelte Fennema-Nio accepted the alternative while noting that it would not accelerate actual initdb options in #6, Jakub Wartak demonstrated consistent Linux gains through repeated comparative runs in #7, and Greg Sabino Mullane clarified that the existing initdb option variable remained available in #8.
- Andrey Borodin reviewed Nazir's implementation and separately proposed splitting Linux ASAN tests across two runners in #9, using manual round-robin selection because the Linux image lacked Meson's native slicing support.
- Nazir Bilal Yavuz's review in #10 led Andrey Borodin to sort test names to avoid missing or duplicating tests, extend slicing to MinGW, and share collision-free log artifact naming with Visual Studio in #11; further review in #12 led to removing unnecessary per-slice ccache keys in #13.
- After Jelte Fennema-Nio highlighted the prolonged delay in #14, Heikki Linnakangas announced the server-settings patch's landing in #15; 6b0896f367 applies #5 unchanged apart from rebasing.
- Andres Freund requested backpatching to reduce differences before backporting CI in #16, and Nazir Bilal Yavuz supplied branch-specific versions in #17, removing references to the unavailable PG_TEST_INITDB_EXTRA_OPTS variable on PostgreSQL 15–16.
- At the thread's end, Heikki Linnakangas explained that adding a user-visible testing option had made him hesitate to backpatch, although he would not object, in #18; the server-settings change had landed on the development branch, while backpatching remained undecided and no landing of the separate runner-splitting patch in #13 was recorded.