model-writtenrun 6035 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Neil Chen diagnosed pg_dump's use of the global default ACL when dumping schema-local default privileges and proposed passing NULL for their object type in #1; Boris P. Korzun reported that the fix worked in #3.
- After months without a commit, Boris P. Korzun revived the issue in #4, and Masahiko Sawada confirmed that supported branches remained affected and requested regression coverage in #6.
- Masahiko Sawada took over updating the patch in #8, preserving Neil Chen's fix while rebasing it and adding a function-privilege regression test; Tom Lane's request to explain the sequence object-type conversion in #10 was incorporated in #11.
- Michael Paquier questioned changes to both function and table privilege dumps in #12; Nathan Bossart in #14 and Masahiko Sawada in #15 explained that the patch restored the intended function grant and removed an unnecessary table revocation.
- Nathan Bossart identified the dependence on acldefault() being STRICT in #13 and requested documentation in #16; Masahiko Sawada added that explanation in #17, favoring the smaller existing approach over caller-specific expression construction.
- Tom Lane explained in #18 that a separate planned refactoring would remove the STRICT dependency, while Nathan Bossart suggested separating the comment's two special cases and matching the test's GRANT wording in #19; the committed test retained GRANT EXECUTE as input and GRANT ALL as expected output.
- Tom Lane proposed using an empty baseline for global entries too in #21, but Nathan Bossart demonstrated that this would lose a global function-privilege revocation in #22; Tom withdrew the alternative and confirmed the original distinction between global and schema-local entries in #23.
- Tom Lane committed #17 as 2acc84c6fd with a clearer comment, an equivalent cast adjustment, and test formatting, crediting Neil Chen and Masahiko Sawada and backpatching the fix to all supported branches.
- Tom Lane announced the push in #24, explaining that the fix unblocked his other pg_dump work, and Nathan Bossart acknowledged it in #25; the fix and regression test were fully landed with no remaining patch in this thread.