model-writtenrun 340 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Nathan Bossart reported that REPACK (CONCURRENTLY) failed to check whether the table access method supported logical decoding and proposed restricting it to heap for v19 #1; after Bharath Rupireddy questioned why concurrent repack needed a special restriction, Nathan explained its dependence on logical decoding #2 #3.
- Masahiko Sawada explained that unsupported access methods could silently lose concurrent changes and favored a heap restriction because introducing a capability callback was inappropriate at this stage #4; Bharath agreed, leaving broader support for possible future work driven by field reports #5.
- Bharath submitted the heap-OID guard, then added documentation and a regression test using an alternate access method backed by the heap handler #6 #7 #8.
- Masahiko observed that the test's alternate access method should actually work, so he supplied a revision comparing the table-AM routine pointer instead of its OID and removed the test #9; Bharath accepted the correction, reported successful tests and pgindent, and agreed that core lacked a suitable non-heap access method for testing the restriction #10.
- Bharath's final revision changed the diagnostic from a hint to detail and aligned its wording and documentation #11; Masahiko committed it unchanged as e8636f2bed, with a v19 backport, and confirmed completion in the thread's last message #12.