model-writtenrun 4606 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Mar 2022 · 3 entries
- Jacob Champion proposed client-side require_auth in #1 to reject unexpected authentication requests and incomplete exchanges, particularly to prevent a server from extracting password material when the client expected certificate authentication.
- Tom Lane proposed comma-separated alternatives in #2, Michael Paquier supported client-side enforcement in #3, and Andrew Dunstan raised combined certificate and password requirements in #4; Tom recommended treating TLS conditions independently in #5.
- Laurenz Albe requested a way to reject methods such as plaintext passwords in #6, and Jacob proposed negation while retaining explicit allowlists and rejecting mixed positive and negative lists in #7.
Jun - Jul 2022 · 5 entries
- Jacob implemented comma-separated alternatives and removed misleading password aliases in #11, while concluding that a separate certificate option would avoid the complexity of a general AND syntax.
- Michael's review in #12 identified parser leaks, non-GSS build failures, translation problems and missing interaction tests, leading to #15, which also removed certificate handling from require_auth and added documentation and PGREQUIREAUTH.
- David G. Johnston challenged the interaction of alternatives and fallback in #16; Jacob retained OR as a simple single-connection policy in #17, then added negation and none in #18 while preserving existing behavior for optionally authenticated, incomplete SCRAM exchanges.
- Jacob introduced sslcertmode with allow, disable and require settings in #19, making certificate policy independent of require_auth and describing require as primarily a troubleshooting aid.
- The next revisions fixed postgres_fdw expected output and Windows configuration in #20 and #22; after discussion of the unwieldy option hint, #28 rebased over its upstream replacement without functional changes.
Sep - Nov 2022 · 6 entries
- Peter Eisentraut questioned naming, future methods and SASL compatibility in #24 and suggested HBA names and separate require/reject settings in #26; Jacob explained in #27 that clients cannot distinguish several HBA methods and that one negatable option is easier to override.
- Peter examined incomplete SCRAM exchanges and the GSS-encryption exception in #29, then accepted preserving both behaviors in #31 after Jacob explained them; #32 replaced the remaining TODOs with those explanations.
- Peter's review in #33 prompted translation-safe error composition and a Meson certificate callback check in #34, and also requested preparation for additional SASL mechanisms.
- Jacob added a separate SASL mechanism-list refinement in #35, but questioned whether its two-tier implementation justified the complexity before a second mechanism existed.
- Aleksander Alekseev reviewed the series in #36, prompting clearer certificate documentation and a callback comment fix in #37; his concern about converting minus one to unsigned was withdrawn in #38.
- The series adopted the new libpq error API in #41, and Aleksander reported that the code was well documented and all tests passed in #44; Jacob still identified the SASL refinement as an unresolved maintenance decision in #45.
Jan - Mar 2023 · 7 entries
- After another rebase in #47, Jacob summarized the two core features and the optional SASL refinement in #48, emphasizing credential protection and control over certificates inherited from a proxy's environment.
- Michael committed the preparatory indentation cleanup b6dfee28f2 to ease work on the affected libpq files, then supplied a revised require_auth patch in #51 with creds support, state reset, memory cleanup and empty-option handling.
- Jacob reviewed Michael's revision in #52, and their exchange in #53 led to #55, which added Windows SSPI and empty-option tests, standardized test descriptions and fixed an errant comma while restoring the full series.
- Michael committed the require_auth portion of #55 as 3a465cc678 with light comment and test-description edits, announcing the landing in #56; Jacob resent the two remaining patches in #57.
- Michael's certificate review in #58 requested accurate default-certificate documentation, reusable error wording, sslcertmode memory cleanup and sslinfo checks proving whether a certificate was sent; Jacob incorporated these in #59.
- The sslmode error-message hunk from #59 landed separately as bcaa1fafc8, followed by the certificate feature as 36f40ce2dc with light documentation, message and test edits.
- Michael reported testing the certificate feature with OpenSSL and applying it in #60, then marked the commitfest entry committed in #62 while leaving the SASL refinement uncommitted because a simpler, less error-prone design was still needed.
Dec 2024 · 1 entry
- In December 2024, Jacob reported a new SASL-separation attempt on the OAuth thread in #63, and Michael directed further discussion there in #64; both core features were landed, with the remaining architectural work continuing elsewhere.