model-writtenrun 3593 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Feb - Apr 2022 · 9 entries
- Jacob Champion proposed C and SQL accessors for authenticated identity so extensions and auditing triggers could distinguish the authenticated client from the database role in #1.
- Michael Paquier questioned the C accessor in #2, Julien Rouhaud confirmed that extensions could rely on Port's layout in #4, and Jacob removed the wrapper in #5.
- Jacob corrected the SQL result from cstring to text in #10; Andres Freund then identified missing parallel-worker state, requested a pg_ prefix, and suggested certificate-DN tests in #11, which Jacob addressed initially with a parallel restriction and expanded tests in #14.
- Stephen Frost supported trigger-based auditing in #15, while Michael Paquier and Peter Eisentraut requested clearer documentation and terminology in #17 and #19; Jacob added documentation and tidied tests in #20.
- Andres Freund argued that a parallel restriction could still allow misleading identity results through nested functions in #24, turning worker synchronization into a prerequisite rather than an optional enhancement.
- Tom Lane recommended separating worker-accessible identity data from Port in #30, and Jacob introduced a second patch with a SharedPort substructure and DSM serialization in #31.
- Tom clarified that the sharable data should leave Port entirely in #32, Andres suggested miscinit.c in #34, and Jacob implemented a separate global structure there in #35.
- Robert Haas questioned the copying overhead in #36, while the GUC alternative raised visibility, mutability, and NULL-versus-empty-string concerns discussed in #40, #42, and #44; the dedicated serialization approach remained.
- Jacob adopted Michael's ParallelProcInfo naming suggestion in #47, then returned after the feature-freeze discussion with a rebase and Windows export annotation in #51.
May - Aug 2022 · 9 entries
- Robert Haas argued for a name describing the data rather than its synchronization mechanism in #58 and #60, leading Jacob to adopt ClientConnectionInfo in #63.
- Bertrand Drouvot introduced a separate SYSTEM_USER effort in #64, and Jacob offered to redirect his work toward it when review stalled in #65.
- Michael Paquier proposed landing ClientConnectionInfo independently of either SQL interface in #67 and #71, Bertrand agreed in #72, and Jacob extracted the infrastructure while dropping the SQL function and its tests in #73.
- Bertrand requested authentication-method propagation for SYSTEM_USER in #74, Jacob added it in #75, and Bertrand reported successful authentication and Kerberos integration tests while identifying a missing struct tag in #76.
- Michael proposed an intermediate serialization structure with explicit string lengths in #77, and Jacob implemented that design and the struct-tag correction in #80.
- Bertrand reviewed the revised serialization in #81, and Michael posted Jacob's patch with those refinements, additional checks, formatting, and a startup memset in #84.
- Bertrand approved the final posting in #85, while Jacob questioned the need for explicit initialization in #86; Michael nevertheless landed d951052a9e from #84, moving the memset to BackendInitialize() as explained in #87.
- Tom Lane confirmed that static initialization already handled EXEC_BACKEND in #88, and Michael removed the redundant memset in 701ac2cb1f, retaining the rest of the infrastructure.
- The supplied commits land authentication identity and method propagation, while the original SQL accessor was set aside in favor of the separate SYSTEM_USER work discussed in #67 and #73.
Feb 2024 · 1 entry
- The thread resumes in February 2024 with Andy Fan's compiler-warning fix in #89, which guards an assertion-only maxsize decrement; the supplied record ends without a response or matching candidate commit for that fix.