model-writtenrun 7750 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Arthur Nascimento reported that COMMIT AND CHAIN failed to start a new transaction when unreleased savepoints existed, initially encountering the problem with psql's ON_ERROR_ROLLBACK; he also described a local psql fix in #1 and supplied it in #2.
- Fujii Masao identified missing chaining logic in the server's TBLOCK_SUBCOMMIT path and posted a fix with regression tests in #3; Arthur confirmed that the patch solved his problem in #4.
- After finding no indication that the existing behavior was intentional, Fujii proposed committing both Arthur's psql fix and his server patch in #5.
- Vik Fearing explained the standard's savepoint-destruction requirement in #6; after Fujii requested clarification in #7, Vik confirmed in #8 that AND CHAIN must immediately start a new transaction.
- The server patch from #3 landed unchanged as 8a55cb5ba9, while Arthur's separately supplied psql fix landed as fe06819f10 to prevent releasing a temporary savepoint already destroyed by COMMIT AND CHAIN.
- Fujii announced that both patches had been pushed in #9, closing the thread with both fixes committed and designated for backpatching to PostgreSQL 12.