model-writtenrun 3426 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed cataloging domain NOT NULL constraints alongside table constraints, making them visible in the information schema and manageable through ADD/DROP CONSTRAINT, with separate helper routines for check and not-null constraints #1.
- Aleksander Alekseev found that displaying domains failed because domain constraints have no conkey #2, and Peter added a domain-specific ruleutils.c branch in #6; Peter also clarified that domain nullability remains independent of a table column's own NOT NULL constraint #5.
- After vignesh C reported intermittent regression failures #7, Peter traced them to an existing information-schema locking race with concurrent table drops and planned a separate discussion while rebasing the series #9.
- jian he demonstrated that repeated ADD NOT NULL commands accumulated constraints #10; Tom Lane favored stacking named constraints #12, while Alvaro Herrera explained the inheritance and DROP NOT NULL reasons for permitting only one table constraint and recommended matching that behavior for domains #13.
- Peter considered explicit IF NOT EXISTS syntax #15, but moved the patch forward by adopting jian he's early-return fix from #14 and adding tests for idempotent additions in #18.
- Aleksander tested the final series on two platforms and requested helper-function comments #19, while jian he identified a stale argument comment and raised documentation concerns #20.
- The information-schema test patch from #18 landed unchanged as 9578393bc5, and its catalog implementation landed with light cleanup and a catalog-version bump as e5da0fe3c2; Peter confirmed the commit and retained NOT VALID as unsupported #21.
- Dean Rasheed then showed that CREATE DOMAIN accepted plain NOT NULL while ALTER DOMAIN required an ignored trailing identifier, and recommended consistent syntax plus documentation of the extension #23.
- Peter supplied a separate DomainConstraint grammar patch #24, but questioned domain null semantics under the SQL standard; Tom Lane raised outer-join inconsistencies and suggested reverting the catalog change #25.
- Vik Fearing disputed the interpretation of domain null handling and proposed taking CAST's null short-circuiting to the SQL committee #27 #29, while Tom defended the standard's explicit wording and rejected presenting a change as PostgreSQL's position #31.
- jian he found that the syntax patch left the deparser emitting NOT NULL VALUE #32, and Dean recommended plain NOT NULL, questioned duplicate nullability display, and reported inconsistent acceptance of NO INHERIT for domains #33.
- Alvaro and Dean discussed ways for psql to expose constraint names without excessively wide output, but no display-name solution was included in the submitted patches #34 #35.
- Peter retained the catalog work and reposted the syntax patch with an expanded explanation, noting that reverting would also lose fixes to named domain constraints and ALTER DOMAIN behavior #37.
- After jian he proposed correcting the deparser and filtering psql's Check column #38, Peter incorporated those amendments and new display tests into 9895b35cb8, based on #37.
- The thread ends with all submitted work committed and Peter reporting an additional ORDER BY stabilization for the new display tests #39, although that ordering change is absent from the supplied syntax commit; broader domain-null semantics, constraint-name display, and Dean's NO INHERIT concerns have no recorded resolution here #29 #33 #35.