model-writtenrun 9814 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Alvaro Herrera proposed removing unnecessary relfilenode assignments in #1 after finding them on partitioned indexes and other relation kinds without storage, while leaving pg_upgrade behavior unverified.
- Andres Freund requested regression coverage in #2, and Michael Paquier recommended a shared macro and questioned how heap_create handled caller-supplied relfilenodes in #3.
- Alvaro's revision in #4 centralized the storage checks, incorporated Ashutosh's macro under the name RELKIND_CAN_HAVE_STORAGE and added a catalog sanity test; he explained why caller-supplied relfilenodes must remain supported and noted that the test ran too early to exercise partitioned relations comprehensively.
- Alvaro rebased the patch over intervening changes in #5, adapting to partitioned-table tablespace handling and the macro definition already present in the base.
- Kyotaro Horiguchi reported passing regression tests in #6 and clarified that actual files had already been suppressed: the patch removed false catalog and relcache relfilenodes; Michael acknowledged index-relfilenode reuse and suggested optional boolean and test-comment cleanups in #7.
- Kyotaro advocated RELKIND_HAS_STORAGE because mapped relations also have storage in #6, and Amit Langote reinforced that reasoning while explaining the tablespace assertion in #8.
- Peter Eisentraut requested a rename patch in #9, Alvaro volunteered in #10, and the thread's final message, #11, had Michael request a separate rename commit that also covered copy.c.
- Alvaro committed the discussion-derived macro rename separately as df5be63763, then landed #5 as 807ae415c5 with the new name, a tablespace comment and a catalog-version bump; both intended changes landed, while the optional boolean removal and additional test comment were omitted.