model-writtenrun 8462 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane reported cross-version upgrade failures caused by changed array-function signatures, judged the usability benefit worth the compatibility cost, and initially suggested filtering the affected aggregate from upgrade tests #1.
- After checking the test's purpose, Tom chose array_larger as a replacement because the test only needed to exercise arrays of records, and supplied development-branch and back-branch patches #2.
- Tom committed the array_larger replacement together with an additional first_el aggregate compatibility fix using a SQL transition-function wrapper in 97f73a978f; the latest matching submission was #1093567 in thread #220392, so that thread receives the commit attribution rather than the earlier patch #2.
- Justin Pryzby later reported real PostgreSQL 14 beta upgrade and restore failures involving user-defined aggregates, requested an incompatibility notice, and proposed wording listing the affected functions; Tom agreed documentation was needed #3 #4 #5.
- After clarifying that the failures involved aggregates using built-in transition functions, Bruce Momjian posted a separate release-note item in the incompatibilities section #7 #8 #9.
- Tom explained that custom operators could also break and requested both a general signature-change warning and concrete examples; Bruce incorporated those suggestions across the documentation revisions #10 #11 #12 #14.
- Justin approved the revised explanation but questioned whether dropping objects before upgrading was appropriate advice for dump/restore, and Bruce asked how users should handle restore errors and dependent objects; that workflow question remained unresolved in the thread #13 #15.
- Bruce committed the final release-note attachment #14 unchanged as 25dfb5a831 and announced its application #16.
- A year later, Andrey Borodin reported the same problem in his service and proposed a backpatchable pg_upgrade check, then supplied a draft that detected affected aggregates and operators by function name and reported testing it against PostgreSQL 13 #17 #18.
- Justin pointed to an existing patch in another thread and noted difficulties resolving function signatures on older servers, while Tom warned that name-only matching could select user-defined functions and recommended server-version-dependent regprocedure OID lists #19 #20.
- Andrey withdrew his draft in favor of reviewing Justin's existing work, while suggesting function-name matching restricted to built-in OIDs; the test and documentation fixes had landed, but the pg_upgrade check continued elsewhere with its identification strategy still under discussion #21.