model-writtenrun 5250 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed allowing pfree(NULL), removing the resulting redundant caller checks, and making separate cast, PQfreemem(), and FreeDir() cleanups in #1, while noting that the FreeDir() change was debatable because its underlying closedir() rejects NULL.
- Tom Lane opposed changing pfree() because new code and widespread edits would create back-patching hazards, while accepting the smaller cleanup patches in #2; Andres Freund added concerns about branch overhead and suggested a pointer-clearing helper as an alternative in #3.
- Ranier Vilela reviewed the PQfreemem() cleanup and submitted a replacement adding three missed cases in pg_recvlogical.c in #4.
- David Rowley measured approximately 2.6% slowdown from the proposed pfree() NULL check and suggested a dedicated inline function or macro for nullable pointers in #5.
- David Rowley raised a possible list_free_deep() problem involving NIL sublists in #6, then withdrew it in #7 because list elements need not have node tags and cannot generally be identified as sublists.
- Peter Eisentraut committed the unchanged cast-removal component of #1 as 45987aae26 and its unchanged PQfreemem() component as e890ce7a4f; the latter excludes the additional cases from #4, with no reason stated in the thread.
- The pfree(NULL) behavior change and dependent caller cleanup were abandoned, and 805a397db4 instead documented why repalloc() and pfree() reject NULL, incorporating the discussion's performance concerns.
- Peter Eisentraut closed the thread in #8 by reporting that the secondary patches and documentation update were committed; the supplied candidates verify the cast and PQfreemem() cleanups and documentation, but do not establish the FreeDir() patch's individual landing.