model-writtenrun 2276 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane proposed discounting metapages from genericcostestimate() because counting them could double estimated page access for tiny indexes, while deferring upper-tree-page accounting as difficult and unlikely to matter much; he targeted PostgreSQL 19 without backpatching because the new GenericCosts field changes ABI #1.
- Tom supplied the initially omitted attachment in #2, implementing per-access-method non-leaf-page counts and adjusting existing regression tests to preserve their intended plans.
- Alvaro Herrera supported the idea without a detailed review and suggested changing 'leafs' to 'leaves' in #3, which Tom agreed to do in #4.
- After a long gap, Henson Choi reviewed the implementation, reported passing regression tests, and supplied a supplementary test demonstrating preference for a tiny partial index over a full index in #5.
- Tom committed 733f20df53 from #2 with the wording correction and a rewritten positive test, explaining in #6 that reusing onek2 avoided creating a sizable new table and that a transaction isolated the competing index; the final message confirms the work was pushed.