model-writtenrun 5966 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Martín Marqués opened the thread after discovering that pg_monitor could not read pg_replication_origin_status, then proposed selectively bypassing the hardcoded superuser check for status/progress functions in #1 and #2.
- Michael Paquier proposed replacing the hardcoded check with function ACLs in #3, and Martín implemented that approach with PUBLIC execution revocations and monitoring grants in #5.
- Michael requested separate patches for the ACL conversion and new role privileges in #7, while Kyotaro Horiguchi and Stephen Frost favored assigning monitoring access through pg_read_all_stats in #8 and #10; Martín produced the corresponding four-part series in #11.
- Kyotaro verified that all affected functions had execution revoked but found the missing EXECUTE grant for the view's backing function in #12, leading Martín to add it and revise the documentation in #13.
- Kyotaro proposed grouping grants with revocations and supplied supplemental view documentation in #14; Martín adjusted placement and function documentation in #15, and Kyotaro supplied a tested revision fixing an SGML closing tag in #17, although the separate catalogs.sgml addition was not incorporated.
- Masahiko Sawada questioned why ordinary users could not access a filtered version of the view, as with other statistics views, in #16; Kyotaro shared that concern but recognized Martín's preference to preserve existing behavior in #18.
- Michael narrowed the immediate proposal to the ACL conversion and its documentation in #19, deferring automatic pg_read_all_stats access because that part remained unsettled and could subsequently be implemented with grants.
- Masahiko questioned requiring grants on both the view and its undocumented backing function in #20, but accepted Michael's explanation of comparable existing views in #21 and #22; Michael also withdrew his proposed extension to replication-slot functions because their REPLICATION privilege checks complicated it.
- Kyotaro approved the narrowed patch while continuing to support default monitoring access in #24, and Michael committed #19 as cc072641d4, adding permission-denial regression tests and announcing the landing in #26.
- Tom Lane identified the missing catalog-version bump in #27, explaining that existing clusters could retain incompatible catalog ACLs and fail regression tests; Michael fixed it directly from the discussion in 46241b28d8 and acknowledged the correction in #28.
- Daniel Gustafsson marked the commitfest entry committed in #29, and Michael agreed in #30 because discussion had stalled, while explicitly noting that automatic pg_read_all_stats privileges remained unfinished.
- After more than a year, Martín revived the uncommitted monitoring-access portion in #31, attaching the view and function grants with updated documentation; the thread ends with that remainder awaiting review and his question about adding regression tests still open.