model-writtenrun 8455 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund opened the thread because asynchronous I/O can require logging relation paths inside critical sections, where GetRelationPath() violates allocation restrictions; deferring those messages could postpone them arbitrarily #1.
- Noah Misch favored a temporary memory context around error-report arguments #2, while Thomas Munro cited a caller-provided-buffer implementation #3 and Andy Fan argued for offering both allocating and caller-buffer interfaces #4.
- After several months, Andres proposed returning a fixed-length string inside a struct by value, avoiding allocation, caller-buffer hazards, and non-reentrant static storage; his prototype converted selected callers and exposed an existing warning-path memory leak #5, which Thomas confirmed while endorsing the approach #6.
- Andres investigated extension usage and proposed dropping the old allocating interface #7; Noah supplied supporting PGXN results #8, and Tom Lane agreed that the compatibility break was acceptable while questioning whether caller-supplied storage would better fit existing conventions #9.
- Andres retained return-by-value, explaining its simpler use in logging expressions, and expanded the patch to all callers with RelPathStr, a named process-number length bound, and regression checks; he also introduced a separate patch removing segment-path allocations in md.c #10.
- Thomas explored compile-time digit calculations #11 #13, and Tom advocated simply reserving ten characters #12, but Andres chose the existing exact bound checked by a regression test because the alternatives added complexity #14.
- Discussion of relocating MAX_BACKENDS was moved to another thread #15 #16, while Andres finalized the return-by-value series with only light comment edits and fuller commit messages #17.
- Andres committed both patches from #17 without implementation changes: the relpath API and caller conversion as 37c87e63f9, and the segment-path conversion as ecbff4378b, then announced the push #18.
- In September 2026, Nazir Bilal Yavuz reported two stale comments and supplied a follow-up removing the obsolete palloc description and adding the missing fork suffix to the maximum-path example #19.
- Heikki Linnakangas committed the comment cleanup from #19 unchanged as ffe95d9281, with backpatching through PostgreSQL 18, and announced completion in #20, leaving no outstanding patches in the thread.