model-writtenrun 7697 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Pavel Boev reported that placing NULL before or after GENERATED BY DEFAULT AS IDENTITY produced different column nullability, demonstrating an order-dependent bug in #1.
- Vik Fearing confirmed the bug and submitted #2, rejecting explicit NULL in either order, documenting identity columns as NOT NULL, and testing both conflicting and compatible declarations.
- Amit Langote identified a trailing blank line in the tests in #4, and Tom Lane proposed treating IDENTITY as an implicit explicit NOT NULL declaration to reuse existing conflict handling in #5.
- Tom Lane committed Vik's patch with that parser simplification, revised diagnostics, and small documentation and test edits as f52c5d6749, back-patching the fix through PostgreSQL 10; Vik acknowledged the work in #6.
- After the minor release, Shay Rojansky reported that the release notes mentioned only GENERATED ALWAYS even though GENERATED BY DEFAULT also rejected NULL, and questioned whether that restriction was intended in #7.
- Tom Lane demonstrated the existing NOT NULL behavior in #8, then confirmed that the SQL standard requires it for both identity variants in #9; he corrected the release-note wording in e5f9821569 without another attached patch.
- The thread ended with both fixes committed, while Shay's proposal to permit nullable BY DEFAULT identity columns as a PostgreSQL extension in #10 and his request for explicit documentation of the NULL incompatibility in #11 remained unanswered.