model-writtenrun 2725 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Nikolay Samokhvalov proposed renaming EXPLAIN ANALYZE to EXECUTE, making VERBOSE include BUFFERS and SETTINGS, and allowing simpler unparenthesized syntax in #1.
- Tom Lane and Robert Haas explained that EXECUTE already identifies prepared-statement execution, extending unparenthesized options requires reserved keywords, and coupling VERBOSE to other options creates compatibility and precedence problems, so those proposals did not advance into patches #2 #4 #7.
- Nikolay argued that buffer counts help explain performance independently of timing, and David G. Johnston advocated enabling them specifically with ANALYZE while explicitly disabling them in regression tests #8 #10.
- David Rowley demonstrated that narrower approach with a patch changing EXPLAIN's default and adjusting tests across 23 files, while noting that enabling BUFFERS without ANALYZE would cause substantially more test churn #13.
- Robert argued that changing the BUFFERS default was more achievable than EXPLAIN ALL, and Guillaume Lelarge independently submitted a small implementation; Robert then requested an audit of buffer handling and a corresponding auto_explain default change #17 #19 #20.
- Guillaume expanded his patch with tests, documentation, and auto_explain changes, while David clarified that explicit BUFFERS without ANALYZE must remain useful for measuring planner buffer accesses #21 #23.
- Greg Sabino Mullane's review led to clearer default-setting wording and examples showing dirtied and written buffers, which Guillaume incorporated in successive revisions #22 #25 #26 #27.
- Support accumulated, but Robert clarified that he was neutral and Tom opposed the additional clutter more than the compatibility change; Michael Christofides checked documentation coverage and argued that buffer information particularly helps less experienced users #32 #36 #39.
- David combined and polished the implementations, repaired additional documentation and tests, and split auto_explain into a separate patch because changing logging defaults needed its own justification and release-note visibility #41.
- David committed the first patch from #41 unchanged in substance as c2a4078eba, crediting Guillaume and himself, while explicitly leaving the auto_explain patch uncommitted pending further discussion #43.
- Buildfarm failures exposed two missed select_into test statements whose planning could access buffers, prompting 9fa1aaa652 to add BUFFERS OFF and update expected output; David subsequently reported these missed-test fixes in #44.
- Testing with forced relation-cache and catalog-cache release exposed more omissions in materialized-view tests and the misc_functions helper, fixed separately by 9df2a4b931, while David continued a full test run for remaining failures #44.
- Jian he identified outdated materialized-view examples in rules.sgml, and Guillaume supplied a separate documentation fix containing buffer output and refreshed plan details #46 #47.
- Jelte Fennema-Nio reported the remaining pg_stat_statements failure, Guillaume pointed to a fix he had sent on another list, and David committed the level_tracking test corrections as 89988ac589 after his full run found no further failures #48 #49 #50.
- Michael supported consistent auto_explain defaults but acknowledged upgrade overhead, and his testing of the rules.sgml examples suggested explicitly disabling BUFFERS there might be clearer because file_fdw scans did not report execution buffers #51.
- Jelte later highlighted a separate proposal for actual filesystem I/O metrics as related work, without adding a patchset to this thread #52.
- Fujii Masao revived the unresolved auto_explain default change in May, arguing for consistency and noting that its sample log output also needed updating; David remained neutral and measured a small throughput reduction with buffer logging enabled #53 #54.
- At the thread's end, Jelte and Fujii supported enabling auto_explain.log_buffers because users could disable it, but no landing of that patch was reported; the main EXPLAIN change and three regression fixes had landed, while the rules.sgml follow-up also lacked a recorded resolution #55 #56.