model-writtenrun 9698 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Ashutosh Sharma reported that ON SELECT rules could create zero-column views while CREATE VIEW rejected them, and proposed rejecting rule-based conversion too in #1.
- Rushabh Lathia found that the proposed relnatts check ignored dropped columns and suggested clearer error wording in #2, then demonstrated the resulting dump/restore failure in #4.
- Andres Freund argued that zero-column views should be allowed consistently with tables and SELECT in #3 and #6; Tom Lane agreed in #8, and Ashutosh replaced the restrictive patch with removal of CREATE VIEW's check and regression coverage in #9.
- Tom raised backpatching and related materialized-view and SELECT INTO behavior in #10; Ashutosh confirmed those operations already worked and supported backpatching in #11, then added tests for them in #12.
- Andres requested tests selecting no columns from a table that has columns and persistent, distinctly named views for pg_upgrade's dump/restore coverage in #13; Ashutosh incorporated those changes in #14.
- Tom committed a32ca78836 from #14, preserving the backend fix but replacing the tests with key-dependent views that also exercise pg_dump's postponed view-rule path, and backpatched the dump/restore bug fix to all supported branches; his announcement in #15 explained the additional coverage.
- Ashutosh acknowledged the test revisions and commit in #16, leaving the fix landed with no further work requested.