model-writtenrun 641 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund proposed shortening pg_attribute_always_inline to pg_always_inline because its length caused awkward pgindent formatting; Peter Geoghegan supported the change, and Peter Eisentraut observed that the macro also includes the inline keyword #1 #2 #3.
- Andres initially targeted the PostgreSQL 20 cycle; Peter Geoghegan later suggested the still-shorter pg_mustinline, but Tomas Vondra preferred pg_always_inline and Peter accepted that choice #4 #5 #6 #7.
- When Tomas moved toward committing and backpatching, David Rowley proposed retaining the original macro on released branches to preserve compatibility, and Tom Lane and Peter Geoghegan endorsed that approach #8 #9 #10 #11.
- Tom Lane considered adding only the new macro sufficient for backbranches, but Tomas chose to rename existing uses too, reducing conflicts when backpatching nearby changes, and submitted master and backbranch variants #12 #13 #14.
- Tomas subsequently supplied per-version patches and reported that only three lines in tuplesort.c retained the old spelling; Peter Geoghegan and Andres approved, although Andres would personally have left backbranch callers unchanged #15 #16 #17.
- The thread ended with Peter Geoghegan identifying external packages that depended on the old spelling, reinforcing the compatibility rationale; the patches had approval but Tomas had not yet announced a commit #18.
- The following day, 885dc83380 landed the rename with Andres credited as author and Peter Geoghegan and Tomas credited as reviewers; its diff follows the compatibility-preserving branch variant described in #14, and its message records backpatching through PostgreSQL 14.