model-writtenrun 865 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Nikita Kalinin reported that DO_NOT_SCAN((a)) with an EXISTS query triggered an identifier-target assertion in pg_plan_advice, providing a reproducer and backtrace in #1.
- Pierre Forstmann could not reproduce the crash and questioned the reported server version in #2; after Michael Paquier clarified module availability in #3, Nikita confirmed testing the development branch in #4, and Tender Wang reproduced the failure without optimization in #6, answering Tom Lane's diagnostic suggestion in #5.
- Tender Wang first proposed unpacking ordered-list targets into individual identifiers in #7, then abandoned that approach because it changed the displayed advice and instead proposed removing the assertion in #8.
- Ayush Tiwari endorsed preserving the target tree in #9, explaining that matching and hash indexing already traverse descendants and recommending a corresponding comment update.
- Tender Wang submitted the sole attached patch in #10, removing the assertion, updating its comment, and adding syntax and scan tests that treated grouped DO_NOT_SCAN targets as valid.
- Robert Haas explained in #11 that DO_NOT_SCAN was intended to accept only simple targets and announced b1901e2895, which fixes scanner/parser classification and tests rejection of grouped targets; the bug was fixed, and the attached assertion-removal approach was superseded.