model-writtenrun 1350 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Following strong support at the FOSDEM developer meeting, Jelte Fennema-Nio submitted #1 to disable JIT by default, updating its implementation, configuration and documentation to avoid compilation-driven performance cliffs.
- Early respondents supported the change based on JIT problems they had encountered, including Jim Jones #2, Laurenz Albe #3 and Bernd Helmle #4.
- Euler Taveira warned that changing the established default could surprise OLAP users and advocated better documentation or higher activation thresholds in #6, while Pierre Ducroquet proposed disabling or separately costing tuple deforming instead of disabling all JIT in #7.
- Christoph Berg argued that predictable behavior across installations justified disabling JIT in #8, Adrien Nayrat documented cloud-provider defaults in #9, and Andres Freund supported disabling it until improvements such as caching and better generated code addressed growing compilation costs in #11.
- Jelte marked the patch ready for a committer in #12; Tomas Vondra subsequently proposed committing it for PostgreSQL 19 in #13, with Tom Lane supporting action before beta in #14.
- Pierre renewed his tuple-deforming alternative in #15, and Jelte supplied benchmark attachments in #16; Andres countered that expression compilation dominated the cases he had examined in #17, while Tomas cited expensive JIT examples and questioned whether disabling an obscure developer option would sufficiently address the problem in #18.
- Tomas committed 7f8c88c2b8 from #1, retaining the default changes while rewording the commit message and removing the documentation's historical-default parenthesis for consistency with other GUC documentation, as explained in #19.
- After the commit, Thomas Munro raised continued buildfarm JIT coverage in #20; Andrew Dunstan explained that owners could enable it through extra_config and suggested a possible future client default in #21, while a proposed buildfarm indicator remained only a discussion in #22 and #23.
- The default change remained committed at the end of the thread, and Tomas reported enabling LLVM and JIT on five buildfarm animals in #25, following his plan in #24; no further patch or reversal was submitted.