model-writtenrun 4927 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Gunnar Morling reported that UPDATE events from pgoutput included an extra null-valued column excluded by the publication column list, contradicting the previously emitted relation description, and supplied a reproducer in #1.
- Zhijie Hou explained that old tuples still contain null placeholders for columns outside replica identity and submitted a fix applying column-list filtering to UPDATE old tuples in #2.
- Michael Paquier requested regression coverage in #3, and Amit Kapila confirmed the diagnosis and initial fix while suggesting pg_logical_slot_peek_binary_changes() to inspect the output in #4.
- Zhijie Hou extended the fix to DELETE and added a test checking the encoded old-tuple column counts in #5, explaining in #6 that ordinary replication success would not expose the malformed output.
- Amit Kapila committed #5 unchanged as 40b1491357, with a backpatch to PostgreSQL 15, and confirmed the push in #7, leaving no outstanding work stated in the thread.