model-writtenrun 9685 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed suppressing post-bootstrap TOAST creation for system catalogs so ALTER TABLE could set options such as autovacuum parameters without lock errors or unwanted TOAST tables, and included a catalog TOAST regression test in #1.
- Andres Freund opposed silently bypassing TOAST requirements and preferred adding necessary TOAST tables in #2; Peter subsequently explained that the separate catalog-TOAST work would still leave pg_attribute and his original use case unresolved in #5.
- Michael Paquier found the proposed regression test superseded by existing coverage and requested updated comments and tests in #6; Peter explained that testing required a server-start setting and therefore a separate test setup in #7.
- Andres suggested either moving the error below the no-op checks or placing catalog exclusions in needs_toast_table() in #11, while Tom Lane favored the minimal error-check relocation in #12; Peter selected the needs_toast_table() approach because merely relocating the error would not fix pg_attribute in #13.
- Andres questioned unchecked callers and removal of the shared-table error in #14; Peter answered that needs_toast_table() runs regardless of the check argument and would make that error redundant in #18.
- Alvaro Herrera emphasized the practical need for catalog autovacuum settings and moving pg_largeobject between tablespaces in #15 and #17, while Andres clarified that his objection concerned inconsistent TOAST requirements rather than all catalog DDL in #16.
- After several months, Kyotaro Horiguchi proposed compression-only storage and explicit catalog-column annotations in #19, then added user-visible SET STORAGE compress support in #20; Chris Travers supported exposing and documenting the feature in #21.
- Peter suggested using existing MAIN storage and generating catalog settings automatically in #22; John Naylor demonstrated table-level annotations in #23, but found that bootstrap generation could only cover pg_class and pg_attribute and suggested generated initdb SQL as another possibility.
- Peter committed #13 unchanged apart from rebasing as 590a87025b and announced in #24 that the remaining storage discussions were independent of this fix.
- At the thread's close, Kyotaro clarified that MAIN storage still permits TOAST-table creation whereas his proposed compression-only storage would prevent it, but acknowledged that Peter's committed patch solved the reported problem in #25; the alternative storage and annotation patches remained uncommitted here.