model-writtenrun 9013 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tim Clarke reported repeated logical-replication bitmapset errors on PostgreSQL 10.7 in #1, then traced their appearance to adding primary keys and refreshing the subscription; rebuilding the subscriber restored replication but left the bug unresolved #6.
- Tom Lane requested a reproducible case #7, while Alvaro Herrera considered system-column replica identities #4 and Peter Eisentraut suspected missing attribute-number offsets but could not reproduce the failure #9.
- After a six-month gap, Jehan-Guillaume de Rorthais supplied a reproducer affecting PostgreSQL 10, 11, and 12 #11, then traced the error to an unmapped subscriber identity column represented by -1 and posted a tested inline guard before the bitmap lookup #12; he also noted an assertion-related crash that he intended to report separately.
- Peter Eisentraut packaged that guard with a simplified regression test in #13, targeting the upcoming minor releases; the fix marks the relation non-updatable when a subscriber identity column has no publisher mapping, allowing inserts to proceed without the bitmap error.
- Jehan-Guillaume de Rorthais questioned subscription disable/enable operations in the test #14, and Andres Freund objected to creating another subscriber and duplicating setup #15; Peter Eisentraut retained the code fix, simplified the scenario, and moved it into the existing schema-difference test in #16.
- Peter Eisentraut committed #16 unchanged as 1c60e40ad5 and announced that it had been backpatched #19; the thread ended with Jehan-Guillaume de Rorthais's acknowledgment and no further requested changes #20.