model-writtenrun 7637 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed SQL:2016 JOIN USING aliases in #1, initially treating the feature as a grammar extension that could reuse ordinary join-alias behavior.
- Thomas Munro examined the syntax and diagnostics in #2, while Fabien COELHO reported successful build, regression and documentation checks but demonstrated that hiding the input-table aliases made some columns inaccessible in #3.
- After uncertainty about the standard, Fabien COELHO clarified in #7 that the new alias must expose only USING columns while preserving input-table aliases, prompting Peter Eisentraut's substantially expanded implementation in #8.
- Vik Fearing observed in #9 that common columns remained accessible through their input-table names contrary to his reading of the standard; Peter Eisentraut retained PostgreSQL's existing behavior for consistency and rebased onto the parser namespace refactoring in #11.
- Wolfgang Walther argued in #13 that retaining input-table references matters for distinguishing unmatched outer-join rows; Peter Eisentraut later treated the outer-join documentation question as independent of the new alias syntax in #16.
- Following a long review gap and build failures, Peter Eisentraut refreshed the patch in #15, rebasing it and simplifying the grammar to carry the USING columns and alias together through join_qual.
- Peter Eisentraut requested Tom Lane's parser expertise in #18, and Tom's review in #20 replaced the single-namespace-item approach with separate alias metadata and namespace items, splitting the work into an infrastructure patch and the feature patch.
- Tom Lane also found that whole-row references incorrectly included every join column, fixed them by constructing a RowExpr containing only USING columns, and added tests in #20; he deferred improvements to view deparsing and hidden-alias error hints as cosmetic follow-up work.
- Peter Eisentraut committed the infrastructure portion of #20 as 66392d3965, crediting Tom Lane and moving feature-specific comments out of the prerequisite.
- Peter Eisentraut committed the feature portion of #20 as 055fee7eb4, with light documentation and comment edits plus a regression test recording the hidden-alias diagnostic.
- The thread ended with Peter Eisentraut confirming both patches were committed in #21, while leaving better use of aliases during view deparsing and improved hidden-alias diagnostics for future work.