model-writtenrun 7217 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Alvaro Herrera proposed using catalogued not-null constraints in information_schema and attached separate tests adapted from Peter Eisentraut's work, while identifying unstable result ordering and unwanted partition foreign-key rows in #1.
- Tom Lane rejected tests that inspected the entire regression database because they would depend on unrelated tests and require continual expected-output changes in #2; Alvaro agreed and proposed isolated test objects in #3.
- Alvaro found that domain not-null constraints had never been represented adequately and proposed handling them separately in #3; Peter confirmed that omission and supported separate work in #11.
- The missing table identity in CHECK_CONSTRAINTS prompted a standards debate: Vik Fearing favored schema-wide constraint-name uniqueness, David G. Johnston defended the existing compromise, and Tom Lane opposed compatibility-breaking enforcement or changing standard view columns, leaving the naming policy unchanged #4 #5 #6 #8 #10 #13.
- Alvaro submitted the view changes without tests in #12, explicitly deferring both testing improvements and domain support, then committed them as 3af7217942 with only a pg_catalog qualification added and announced the landing in #14.
- Peter reported that the not-null check_clause incorrectly included CHECK and surrounding parentheses, supplied the corrected expression inline in #15, and committed it with a catalog-version bump as a0a5e0feb3, confirmed in #16.
- Further testing led Peter to propose a separate ordinary CHECK constraint cleanup in #18: using pg_get_expr() would remove redundant parentheses and prevent inappropriate suffixes such as NOT VALID from appearing in check_clause.
- Peter committed that cleanup as e59fcbd712, adding only a catalog-version bump to #18, and announced completion in #19; all three implementation changes had landed, while isolated tests, domain not-null support, and the partition foreign-key display concern remained unresolved in this thread.