model-writtenrun 4831 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Kotaro Kawamoto opened the thread with #1, proposing fixes for missing MATCHED/NOT MATCHED suggestions after WHEN and premature USING suggestions after an unfinished AS alias.
- Shinya Kato requested consolidation of equivalent completion branches and caught a compilation error and whitespace problems; Kawamoto incorporated those reviews and added DO NOTHING completion across #3, #5, and #6.
- Kawamoto expanded alias and join-condition handling in #8, then adopted Fujii Masao's request to complete all selectable relations after USING in #10 and Shinya Kato's UPDATE SET suggestion in #12.
- Fujii Masao took over patch preparation in #13, fixing repeated USING suggestions, moving a misplaced LOCK TABLE rule, adding comments, and completing MERGE INTO as a unit.
- Álvaro Herrera observed in #14 that several rules lacked a MERGE-context restriction, prompting Fujii Masao to retain the existing fixes as one patch and add a separate experimental PartialMatches helper in #15.
- Álvaro Herrera endorsed backpatching the first patch and applying the helper only to master in #16; Fujii Masao committed the first patch unchanged as 9a6915257d and backpatched it to PostgreSQL 15, leaving the helper pending.
- After a January application-failure report, Dean Rasheed explained in #18 that the first patch had already landed and resubmitted the unchanged second patch alone.
- Dean Rasheed's review in #18 questioned whether the extra guard was necessary, suggested the name MidMatches and clearer documentation, and proposed simplifying both the matching loop and the attribute-completion rules.
- The remaining patch received no revision after that review: Gregory Stark flagged the inactivity in #19, and Daniel Gustafsson closed it as Returned with Feedback in #21, leaving the basic MERGE improvements committed and the general matching helper unfinished.