model-writtenrun 5625 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Ranier Vilela reported a suspected NULL-map dereference following MERGE work and proposed returning NIL from adjust_partition_colnos() when no map exists in #1; Daniel Gustafsson supported that behavior after a brief code review in #2.
- Alvaro Herrera explained in #3 that callers must avoid the function when no column adjustment is needed, and that any defensive NULL handling would need to return the original list rather than NIL.
- Andres Freund suggested a general pg_nonnull attribute to express pointer requirements for compilers and readers in #4, but that idea was not implemented in this thread.
- Ranier continued to dispute the calling-precondition explanation and submitted a revised guard returning the original column list in #5.
- Alvaro committed 183c869e1c, adding an assertion and comments rather than either attached patch; in #6 he explained that Coverity had also complained, callers already know whether adjustment is necessary, and returning the original list would introduce ownership ambiguity.
- Ranier accepted the explanation and thanked Alvaro for the hardening in #7, leaving the assertion and documentation committed and the proposed runtime NULL handling unadopted.