model-writtenrun 1513 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Paul A Jungwirth proposed using SupportRequestSimplify to inline dynamically generated queries from non-SQL set-returning functions, motivated by temporal joins and predicate pushdown, with a small proof of concept lacking tests or documentation #1.
- Heikki Linnakangas identified that inserting a Query where an expression was expected was invalid #2, and Tom Lane required a separate support request to preserve the existing protocol #3; Paul consequently introduced SupportRequestInlineSRF with documentation and PL/pgSQL regression examples for calls in FROM #4.
- Tom argued that SQL-specific restrictions and result processing should be separated from support-driven replacement, requested access to RangeTblFunction and PlannerInfo, and required clearer documentation and graceful failure handling #5; after a lengthy gap, Paul returned with separate SQL and support helpers beneath a shared wrapper #6.
- Tom's next review requested shared memory management, common volatile-argument and configuration checks, early SELECT validation, support for non-set-returning functions where valid, and HeapTuple metadata #7; Paul implemented those changes and explained why dynamically chosen tables, columns and joins justified the feature #8.
- Tom committed #8 as b140c8d7a3, renaming the request to SupportRequestInlineInFrom to reflect its applicability beyond set-returning functions, simplifying the implementation and documentation, trimming regression cases, and adjusting error handling and null checks; he announced the landing in #9.
- After the commit, Pavel Stehule questioned whether the regression function should raise an exception if executed directly #10, but Tom explained that inlining is never guaranteed and a working fallback remains necessary #11; Pavel clarified that his concern was the example's clarity #12.
- Man Zeng proposed a separate regression enhancement demonstrating fallback to the original PL/pgSQL body and a Function Scan when inlining fails #13; Pavel supported it but requested clearer warnings #14.
- Man proposed removing redundant TEXT checks and issuing argument-specific null warnings #15, received Pavel's approval #16, and folded those changes into the revised fallback-test patch #17.
- Tom questioned the additional tests because they added no line coverage for inline_function_in_from #18; Man defended their explanatory value but accepted that they could be omitted #19, leaving the core feature committed and the fallback-test enhancement unresolved.