model-writtenrun 1090 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane demonstrated that ABI checking missed changes to SysCacheIdentifier because no exported interface used the enum, proposed a dummy function for stable branches, and raised similar concerns about enum GUCs in #1.
- Michael Paquier supported using the real enum type on HEAD and suggested generating a typedef in #2; Andreas Karlsson implemented that approach in #3 and expanded it to invalidation callbacks and object-address code in #5.
- Zsolt Parragi broadened the discussion to compatibility hazards from public inline functions and macros that ABI checking can miss in #6.
- Michael Paquier reported assertion failures caused by negative cache IDs in #7, then supplied #8 with an explicit SYSCACHEID_INVALID value and named sentinels for objects without syscache entries.
- Andreas Karlsson endorsed the fix and requested separate commits for the invalid value and the type conversion in #9, and Michael Paquier agreed in #10.
- Michael Paquier committed the invalid-value portion of #8 as c06b5b99bb, adding two equivalent sentinel-check cleanups, and committed Andreas Karlsson's typedef/API portion as ee642cccc4 with formatting adjustments.
- Michael Paquier reported the additional invalid-ID cleanups and a quiet buildfarm after both February landings in #11.
- Andres Freund reopened the issue in #12, reporting that the new public-header dependencies increased recompilation after a syscache.h change from 196 files to 632.
- Michael Paquier proposed direct syscache_ids.h includes and a generated include guard in #13; Andres Freund reviewed that repair in #14 while questioning the enum conversion's benefits and proposing named global syscache objects as a longer-term alternative.
- Michael Paquier favored retaining the enum-based ABI checking and applying the narrower header fix for v19 in #15, leaving the broader syscache redesign for future work.
- The header repair from #13 landed as e0fa5bd146 with generated-header formatting edits, confirmed in #16; the HEAD changes were complete, while the original stable-branch dummy-function proposal and broader ABI concerns had no recorded resolution in this thread.