model-writtenrun 8466 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andrei Lepikhov reported that CREATE INDEX CONCURRENTLY could trigger an assertion when a supposedly immutable predicate obtained a transaction ID, and proposed replacing the assertion with an ERROR plus a regression test in #1; he noted that development already had a transactional-update fix while PostgreSQL 9.6–13 remained affected.
- Michael Paquier proposed backpatching the existing transactional-update fix in #2, and Tom Lane endorsed that approach in #3 because the development version had been in use long enough to inspire confidence.
- Andrei Lepikhov supplied the PostgreSQL 9.6 adaptation in #4, abandoning the earlier error-based approach and replacing in-place catalog updates with transactional heap and index updates.
- The existing fix originating in thread #42893 was backpatched following this discussion for PostgreSQL 13 as acb60edf02, PostgreSQL 12 as e52f7cbec1, PostgreSQL 11 as 08acba558b, and PostgreSQL 10 as 3c465883b8; these implementations use CatalogTupleUpdate() and derive from that other thread.
- The PostgreSQL 9.6 adaptation from #4 landed as d3d0cbeb62 after Michael Paquier corrected its function comment and replaced explicit catalog-index management with CatalogUpdateIndexes(), as explained in #5.
- Michael Paquier extracted the regression test from #1 into 09a69f6e23, adjusted it to expect successful index creation under the transactional fix, and added function cleanup.
- Michael Paquier closed the thread in #5 by confirming that both the backpatch and the separate regression test had been applied, leaving no outstanding patch work.