model-writtenrun 4919 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- The thread opened with Noah Misch's announcement that public-schema CREATE privileges were revoked from PUBLIC and ownership transferred to pg_database_owner, while upgrades would preserve existing ownership and permissions #1.
- Tom Lane reported that skipping the tablespace regression test broke subsequent tests because it also skipped their required public-schema GRANT, and proposed either a separate setup script or initialization through pg_regress #2.
- Noah supported a dedicated setup script, and agreed with Tom that it could run before tablespace despite losing a little coverage of default permissions there; other suites retained that coverage #3 #4 #5.
- Tom implemented that agreement in 944dc45d1b, moving the GRANT into a new test_setup script scheduled first and reverting the earlier tablespace-test adjustments; broader consolidation of shared test objects remained future work.
- Nearly a year later, Robert Haas proposed describing current user-private-schema setup before upgrade-specific REVOKE instructions; after Justin Pryzby noticed that the attachment contained only a diffstat, Robert supplied the missing patch #7 #8 #9.
- Noah objected that unconditional REVOKE guidance avoided silent security mistakes and also covered permissions changed by another administrator, while Robert maintained that the existing presentation was confusing #10 #11 #12 #13.
- Tom proposed retaining Robert's paragraph separation while explicitly requiring schemas without public CREATE privileges, and questioned whether schema ownership needed an example; Isaac Morland and David G. Johnston recommended AUTHORIZATION guidance #14 #15 #16.
- Robert accepted Tom's direction, and Noah refined it to require that no schemas, including the per-user schemas, have public CREATE privileges, placing that requirement first #17 #18.
- Alvaro Herrera endorsed the revised wording, and the thread ended with Tom promising to commit it #19 #20.
- Tom subsequently committed the documentation rewrite as ef2d7c6f0b, retaining the organization from #9 with the agreed security wording, an explicit AUTHORIZATION example, and editorial cleanup; both concrete fixes discussed in the thread landed.