model-writtenrun 9194 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michael Paquier proposed tracking executed isolation-test steps and warning about unused definitions, attaching cleanup of ten unused steps across six specs #1; Tom Lane questioned whether these were missing tests or development leftovers #2, and Michael attributed several cases to copying queries while developing tests #4.
- Melanie Plageman requested checking unused steps in dry-run mode #3, leading Michael to move the check into run_testspec() and add dry-run tracking while preserving normal-run warnings #5.
- Alvaro Herrera explained that dry-run originally generated permutations for authors to trim, but changes to deadlock handling made that workflow impractical #6 #9; Tom insisted normal test runs should still warn #8, and Michael explained the revised behavior while considering renamed permutation-printing options #10.
- Melanie acknowledged misunderstanding dry-run and supported renaming it #12, but Michael ultimately chose removal and submitted two patches: remove dry-run first, then add unused-step detection and clean up the specs #14.
- A parallel discussion explored Greenplum's isolation2 framework, especially explicit blocking control #15; Michael suggested adapting features as existing-framework commands and questioned adding Python as a test dependency #17, while Melanie agreed the features did not require Python #18.
- Asim described blocking-step syntax in a separate fault-injection patch series #20, and Alvaro argued that blocking expectations must belong to individual permutations because the same step can block in one ordering and not another #21.
- Michael committed the first patch from #14 unchanged as 9903338b5e and the second as 989d23b04b, adding obsolete-option usage-text cleanup and whitespace formatting to the latter; he confirmed both commits and reported handling logical-decoding isolation-test issues #22.
- Asim subsequently reported revising the separate blocking-syntax patch #23, and Michael redirected that discussion to its own thread #24, leaving both patches submitted here committed.