model-writtenrun 1113 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Shihao Zhong reported that database-wide VACUUM could reuse one table's TRUNCATE and INDEX_CLEANUP settings for later tables, and proposed passing parameters by value with regression tests in #1.
- Nathan Bossart suggested extending the protection to vacuum() or explicitly copying parameters in #2, prompting Shihao's broader by-value patch in #3.
- Nathan then favored a smaller backpatch and supplied explicit copies in #4, also identifying that TOAST relations could inherit settings modified for their main table.
- Michael Paquier endorsed the small fix for stable branches but requested comprehensive input-only parameter handling for the next development release in #5, which Shihao implemented in #6.
- Junwang Zhao advocated const pointers and consistent type spelling in #7 and #8; Shihao adopted the spelling change in #9, while Michael retained by-value arguments and added const qualifiers in #11.
- Nathan found that Michael's version still copied TOAST parameters too late in #12, leading Michael to split the work into Nathan's backpatchable fix and a development-only refactor, with expanded TOAST tests, in #13.
- Nathan supported checking parameter values with injection points in #14, avoiding tests that depended on physical truncation or sufficient index data.
- Michael committed the bug-fix portion of #13 as 661643deda, retaining Nathan's backend fix but replacing the tests with injection-point coverage; he reported backpatching the fix through PostgreSQL 13 and the tests through PostgreSQL 17 in #15.
- After Shihao asked about the original regression tests in #17, Michael explained in #18 that their physical-effect checks could be unstable or expensive and that they lacked TOAST coverage.
- Michael posted the remaining refactor in #15, Nathan reviewed it favorably in #16, and Michael landed it unchanged as 2252fcd427, announcing completion in #20.
- Nearly nine months later, Andres Freund reopened the design discussion in #21 because the refactor pulled vacuum.h and its dependency tree into widely included headers; he advocated const pointers with local copies where necessary in #23.
- Nathan supplied the header-cleanup series in #24, and Andres's review in #25 led to a consistently used params_local copy in #26.
- Michael accepted that direction but requested a warning against using the original pointer in #27; Nathan instead simplified the patch in #28 by retaining pass-by-value only for vacuum_rel().
- Michael preferred uniform const-pointer interfaces in #29, while Nathan argued in #30 that passing the mutable argument by value avoided both an explicit copy and an easily missed warning comment.
- Nathan rebased that design in #31 and committed it unchanged as 771fe0948c, confirming the landing in #32; all submitted work had landed, with vacuum_rel() retaining the debated by-value exception.