model-writtenrun 9723 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane proposed replacing IndexPath's three related lists with IndexClause nodes, recording lossiness and commuting clauses up front to avoid repeated work and prepare for extension-provided index conditions; he reported a 30% planning-time improvement for a RowCompareExpr example and deferred further indxpath.c cleanup #1.
- Andres Freund gave the patch a quick skim and asked whether the new structure could be populated lazily #2; Tom explained that recording information immediately was central to avoiding repeated discovery and that commute_restrictinfo() reduced the extra allocation cost #3.
- Tom also explained that omitting RestrictInfo wrappers would impair extended-statistics handling and selectivity caching #3; Alvaro Herrera suggested allowing extended statistics to construct wrappers only when needed #4.
- In the final thread message, Tom rejected delayed wrapper construction because of mutation risks around GEQO and lost sharing of cost and selectivity estimates, retaining the posted design #5.
- The submitted refactoring #1 subsequently landed unchanged apart from rebasing as 1a8d5afb0d, with further index-condition expansion cleanup explicitly left for follow-on work.