model-writtenrun 4642 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Antonin Houska proposed publication USAGE privileges because subscribers could bypass one publication's filters by subscribing to another #1; after Euler Taveira and Amit Kapila raised existing privilege restrictions, he clarified that independently administered subscribers could misuse publications already present on the publisher #2 #3 #4.
- Peter Eisentraut questioned whether this required a PostgreSQL 15 change, while Euler treated fine-grained access control as a future feature and recommended documenting the current limitation; Antonin supplied an initial documentation warning and offered to implement privileges for PostgreSQL 16 #6 #7 #8.
- Euler requested explicit wording about missing publication privileges and ordinary paragraph markup #10; Antonin then submitted a separate USAGE implementation and revised the documentation #11 #12.
- Euler approved the revised paragraph but asked that it move to Logical Replication Security, and Antonin made that relocation #14 #16.
- After several months, Peter committed the relocated documentation with light wording improvements as 84387fc889 and requested a refreshed submission of the still-uncommitted privilege implementation #17.
- Antonin rebased the implementation and added tests, but reported that historical decoding snapshots could keep seeing revoked privileges even after a new grant; Amit suggested advancing the replication slot, while Antonin preferred restarting replication after clearing potentially exposed subscriber data #18 #20 #23.
- Peter requested an upgrade strategy, simpler ACL infrastructure and basic GRANT/REVOKE tests; Amit supported default public access, and Antonin confirmed that default while identifying a deeper problem: ordinary SELECT access could bypass publication restrictions #19 #21 #22.
- Antonin considered automatically granting column privileges during subscription creation, but Mark Dilger questioned how subscriber-side commands could grant publisher-side permissions, and Peter argued that object creation and privilege grants should remain separate; Antonin instead proposed bypassing table SELECT and schema USAGE checks when publication privileges authorize access #22 #24 #25 #26 #27.
- Antonin's subsequent archive submission introduced COPY PUBLICATION_NAMES, publisher-controlled filtering, publication ACL display in psql and additional tests, followed by a rebased archive version #28 #29.
- Peter's January review requested a complete design description, consolidation of the five-patch series, fewer walsender-specific branches, direct catalog access instead of SPI, more consistent privilege enforcement and a TOAST table for publication ACLs #30.
- Antonin responded with a two-patch series separating preparatory refactoring from the feature, expanding its explanation and adding publication_security to preserve compatibility with older subscribers #31; he subsequently rebased that series in #33.
- At the thread's end, Peter considered his initial comments largely addressed but said the security-sensitive overrides still required substantial analysis and testing, recommending deferral to PostgreSQL 17; the documentation had landed, some supporting refactorings had been committed separately, and the publication-privilege feature remained uncommitted #34.