model-writtenrun 5664 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Simon Riggs proposed a configurable recursive-worktable size multiplier, reporting approximately 100-fold improvement for a shortest-path query while deliberately leaving the separate ten-iteration assumption unchanged #1.
- Simon floated a nonlinear automatic estimate #2, but Kenaniah Cerny warned about amplifying estimation errors and supported the multiplier #3; Peter Eisentraut and Robert Haas also accepted the setting as a practical compromise #4 #6.
- Hamid Akhtar asked whether the multiplier should also affect cost_recursive_union() #5, but the submitted revisions kept the change confined to worktable sizing #9 #12.
- After Andres Freund reported the cfbot failure caused by the missing sample configuration entry #8, Simon added that entry and user documentation #9.
- Tom Lane rejected adding permanent query syntax as too heavy a commitment and requested row-estimate clamping, consideration of zero's meaning, and a corrected explanation of join selection #10; Simon incorporated those changes, temporarily making zero select the default multiplier #12.
- Tom proposed the clearer name recursive_worktable_factor #13, Simon agreed #14, and Tom committed the final submission with that rename, revised documentation, and a positive minimum of 0.001 replacing the special zero behavior #12 0bd7af082a.
- Tom announced the commit #15, and Simon acknowledged it #16, closing the thread with the worktable-sizing feature landed and no further revision pending.