model-writtenrun 8821 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Aug 2019 · 3 entries
- Tom Lane opened the thread with #1, proposing to remove pg_pltemplate so external procedural languages could install as conveniently as built-in languages and Python version selection would no longer be hard-wired into that catalog.
- The series in #1 preserved non-superuser installation through a trustable control-file flag and two regex-based configuration settings, ran installation scripts as the bootstrap superuser, and introduced @extowner@ so selected objects could belong to the caller; parameterless CREATE LANGUAGE would become CREATE EXTENSION for compatibility with old dumps.
- The landed work marked the trusted procedural-language extensions as trusted, while broader contrib markings and finer permission controls remained follow-up work discussed in #1 and #34; 50fc694e43 also explicitly left the remaining Python 2-to-3 migration decisions unresolved.
Nov 2019 · 3 entries
- Peter Eisentraut objected to storing permissions in configuration settings and proposed existing role and database privilege mechanisms in #2, while Stephen Frost argued for a grantable database privilege and highlighted the loss of per-extension filtering in #4.
- Tom rejected additional configuration infrastructure in #9 and posted #10, replacing the GUCs with database-owner access or membership in pg_install_trusted_extension and renaming the extension flag to trusted.
- Upstream conflicts and the annual copyright update required rebases in #11 and #12, with only a small documentation rearrangement accompanying them.
Jan 2020 · 6 entries
- The January discussion exposed competing permission models: Robert Haas wanted superusers to retain control over which extensions users could install in #21, while Stephen argued that trusted extensions should fit the existing database privilege hierarchy in #20 and #43.
- The ownership discussion established that privileged extension objects would remain owned by the bootstrap superuser unless the script explicitly reassigned them, limiting what installation alone granted to callers, as Robert and Stephen discussed in #25 and #26.
- Tom deferred a replacement for per-extension filtering in #34, explaining that the GUC design had been rejected and that additional policy controls should not block the catalog removal.
- To break the deadlock, Tom proposed allowing database owners alone to install trusted extensions in #52, but Stephen continued advocating database CREATE privilege in #54 so owners could delegate installation without granting their entire role.
- Tom accepted the CREATE-privilege compromise in #56 and #60, particularly after Stephen pointed out in #59 that database CREATE already covered publications as well as schemas.
- The final thread message, #61, supplied the revised four-part series using database CREATE privilege and announced an imminent push; 50fc694e43 landed the entire series later that day with only an additional catalog version bump.