model-writtenrun 4795 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andrew Dunstan opened the thread by questioning Meson's platform-specific library directory and compiler definitions appearing in pg_config's linker flags, and Michael Paquier confirmed that custom compiler flags also appeared there #1 #2.
- Andres Freund explained Meson's platform defaults and inclusion of compiler flags during linking, while Peter Eisentraut argued that pg_config should still report compiler and linker flags separately; the thread shows no resulting flag-reporting fix #3 #4 #6.
- Andrew Dunstan and Tom Lane favored preserving PostgreSQL's familiar installation layout, while Andres Freund defended platform-specific library directories and explained that forcing lib would be simple but undesirable for parallel library variants #5 #9 #10.
- Discussion of automatically appended postgresql directories exposed a separate concern: Andres wanted explicit paths respected, while Tom emphasized keeping backend extensions separate from ordinary libraries, a distinction Andres supported #12 #13 #14.
- In January, Peter called for equivalent installation layouts across build systems, supported by Robert Haas and Tom Lane, while Andres continued defending platform defaults; Peter then clarified that his immediate concern was the existing postgresql-suffix logic ignoring the prefix #15 #17 #18 #19 #20.
- Peter's first patch checked complete installation paths before appending postgresql, and Andres accepted the fix while suggesting that the prefix be checked once and reused #20 #21.
- Peter subsequently proposed making every installation-directory variable absolute, believing this would also fix handling of absolute directory options, and supplied a broader patch updating path configuration and rpaths #22 #23.
- Andres explained that Meson's path-joining operator already handles absolute paths and raised a Windows drive-relative-path concern, leading Peter to withdraw the broader rewrite and return to his original patch #24 #25.
- Andres reiterated his reusable-prefix-check suggestion, which Peter incorporated into 8f6858064b from #20; Peter's final message confirmed the commit, leaving the broader installation-layout and flag-reporting debates unresolved in this thread #26 #27.