model-writtenrun 2454 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Viktor Holmberg reported that prepared calls to isn_weak() retained stale results because the function was marked IMMUTABLE, initially suggesting STABLE as a workaround in #1; Daniel Gustafsson proposed VOLATILE in #2, and Viktor adopted that workaround for both overloads in #3.
- Daniel Gustafsson called for backpatching in #4, while Tom Lane highlighted contradictory documentation and missing weak-mode tests in #5; Viktor then explained that the session flag ignored transactions and could leak through connection pools in #6.
- Tom Lane proposed making weak mode a GUC in #7 and explained that RESET ALL would address the pooling hazard and that an extension upgrade script was required in #9, following Viktor's compatibility concerns in #8.
- Viktor accidentally attached an unrelated generic-subscripting patch in #10; after Tom identified the mistake in #11, Viktor supplied the actual ISN patch in #12, implementing isn.weak while retaining the existing functions for compatibility.
- Tom clarified that the git commit log serves as the change history in #13, then committed #12 as 4489044239 with corrections to use set_config_option(), reserve the configuration prefix, install the upgrade script through Meson, expand tests and refine documentation.
- In #14, Tom explained that the committed setter is VOLATILE and PARALLEL UNSAFE while the getter is STABLE and PARALLEL SAFE, matching set_config() and current_setting(), and that direct assignment to the underlying variable would fail to roll back on transaction abort.
- Viktor accepted the corrections and thanked Tom in #15, leaving the ISN change committed with no further patch pending; the earlier backpatch request has no recorded completion in this thread.