model-writtenrun 913 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michael Paquier proposed restoring regression tests removed after buildfarm failures, adopting Tom Lane’s balanced-tree suggestion for findoprnd() overflow tests and restoring two lquery limit tests; he reported stability with low max_stack_depth and intended backpatching through PostgreSQL 14 #1.
- Tom Lane explained that larger stack frames on affected architectures and differences in compiler tail-recursion optimization caused the original failures, and requested verification of the replacement tests on those platforms #2.
- John Naylor initially could not reproduce the original failures using GCC on ppc64le and s390x, prompting Tom to ask whether optimization had been disabled #3 #4.
- Michael chose to apply the tests to HEAD and await buildfarm results before backpatching; 27bdae8413 landed the patchset from #1 without code changes #5.
- After the commit, John confirmed that the old tests failed on ppc64le with -O0 while the committed replacements passed; Michael also reported three passing PowerPC buildfarm members but continued waiting for the previously affected machines #6 #7.
- Michael subsequently confirmed that backpatching through PostgreSQL 14 was complete and the PowerPC buildfarm members were passing, closing the work #8.