model-writtenrun 4322 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Jacob Champion proposed documenting why PostgreSQL's SCRAM implementation should be combined with verified TLS or GSS encryption, citing unprotected usernames, low iteration counts, and authentication exchanges that can skip server verification #1.
- Stephen Frost requested clearer mitigation guidance and integration with the existing spoofing discussion, while Michael Paquier explored checking SCRAM usernames; encoding compatibility and protocol concerns kept those implementation ideas outside this documentation patch #2 #3 #4 #7.
- Jacob clarified that the goal was to discourage reliance on SCRAM alone for server authentication, and Jonathan S. Katz supported explicit TLS identity verification and channel binding while leaving possible client iteration-count controls for future development #9 #13.
- Jacob replaced the warning with a network-spoofing paragraph, then Jonathan narrowed the wording to SCRAM and moved it after the existing SSL certificate guidance, retaining the explanation of offline password analysis #14 #15.
- Jacob pointed out that recommending SSL alone weakened the intended server-verification advice, so Jonathan explicitly referenced the preceding anti-spoofing requirements; Michael checked the rendered documentation and Jacob approved the revision #16 #17 #18 #19.
- Daniel Gustafsson approved the final patch but suggested saying an attacker could “potentially” determine the hashed password, and Michael accepted that qualification #21 #22.
- Michael committed the final attached patch with Daniel's one-word qualification as d0f4824a54 and announced its application; Jacob and Jonathan closed the thread with thanks, leaving the documentation work complete and the broader implementation ideas unimplemented here #17 #23 #24 #25.