model-writtenrun 7286 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Bertrand Drouvot proposed normalizing CALL and SET arguments to reduce pg_stat_statements entry churn and LWLock contention in #1, and Jeremy Schneider reported improved throughput and reduced contention when testing the patch in #5.
- Andres Freund argued that procedure calls should remain tracked independently of track_utility while SET should remain subject to it in #3 and #7; Bertrand implemented that distinction in #10.
- Sami Imseih recommended using the presence of JumbleState instead of teaching pg_stat_statements a utility-node whitelist in #11, and Bertrand adopted this extensible check in #12.
- Jeremy Schneider found that RESET ALL crashed in #14, Julien Rouhaud identified its NULL variable name in #17, and Bertrand added the guard alongside Sami's documentation revisions in #18.
- Julien Rouhaud requested normalization of two-phase transaction commands in #19, leading Bertrand to add transaction-ID locations, jumbling, and tests in #21.
- Kotaro Kawamoto reported successful use but noticed case-sensitive grouping of other utility commands in #22; Julien and Bertrand initially treated general utility normalization as separate work in #23 and #24.
- Fujii Masao highlighted cursor-related entry proliferation but recommended handling it separately, and identified the unintended merging of RESET with SET TO DEFAULT in #26; Bertrand distinguished their statement kinds in #29.
- Fujii's further review in #30 led Bertrand to remove redundant CALL and two-phase-commit test combinations and strengthen the post-parse tracking guard in #31.
- Michael Paquier identified the missing SET LOCAL distinction and suggested considering savepoints in #32; Bertrand acknowledged both issues in #34, while the last attachment #33 only rebased the existing implementation.
- Michael questioned the inconsistent treatment and possible overhead of transaction commands in #35, and Tom Lane proposed generating jumbling code for all utility nodes instead of maintaining a selective handwritten implementation in #36.
- Julien Rouhaud supported using existing node annotations in #40, Sami reported withdrawal of the related cursor entry pending consensus in #42, and Michael recognized agreement on the generated approach in #43.
- After the discussion stalled, Michael marked the original patch returned with feedback in #44; the thread ended with his proposal to reuse location fields and add node-level and field-level exclusion annotations in #45, without an attached replacement implementation.
- The later SET-normalization commit dc68515968, dated September 30, 2024 and credited to Greg Sabino Mullane and Michael Paquier, came from #572271 in thread #49989, with only an additional location-field comment; it cites this historical discussion but its custom handler and parser-controlled argument normalization belong to that later thread, superseding the previous attribution here.