model-writtenrun 9250 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane opened the thread because existing tests never verified actual notification delivery, proposing isolationtester support instead of TAP tests or a new psql formatting feature in #1.
- Andres Freund supported isolationtester and suggested replacing sender PIDs with session names in #2, which Tom accepted in #3.
- While implementing that work, Tom discovered that isolationtester globally suppressed NOTICE messages and posted #4 to remove the suppression, exposing notices already written into two tests.
- Slow-machine testing revealed unstable notice ordering, so Tom proposed test-local suppression and shorter TOAST output in #6; ebd4992821 landed those adjustments along with stdout notice reporting.
- Tom posted the separate LISTEN/NOTIFY implementation in #9, identifying both sending and receiving sessions and testing delivery, deduplication, and the existing lack of deduplication across subtransactions.
- Peter Geoghegan considered the locking function's deliberately misleading IMMUTABLE marking in #8, and Tom subsequently supplied the notice-restoration patch and slow-machine failure output in #10.
- Andres identified the race between the separate lock-check connection and consumption of pending notices in #11; Tom's additional input recheck in #13 produced reliable passes on the slow machine.
- Andres examined the recheck's timing guarantees in #14, and Tom rejected a more complicated input-draining loop as unnecessary for these tests in #15; 30717637c1 combined the fix with #10, restoring the previously suppressed notices.
- Andres questioned notification-reporting timing in #16, while Tom explained in #17 that the tested transaction behavior avoided the analogous race and deferred spontaneous notification reporting across idle connections.
- The LISTEN/NOTIFY work from #9 landed as b10f40bf0e with additional test cases and revised payload strings, retaining the posted isolationtester implementation.
- By the final message, all submitted implementation work had landed; Tom confirmed the isolation tests were committed and left open a psql change to print 'from self' for self-notifications so psql's own notification code could receive regression coverage in #18.