model-writtenrun 2864 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Nathan Bossart proposed retiring MD5 passwords over releases 18–21, progressing from deprecation notices through restrictions on creation and authentication to rejecting upgrades with MD5 passwords, citing weak hashing and pass-the-hash attacks in #1.
- Jelte Fennema-Nio favored rejecting incompatible upgrades before applications broke, Jonathan S. Katz and Jacob Champion advocated an accelerated warning-based transition, and Heikki Linnakangas questioned intermediate restrictions that would prevent dump-and-restore workflows #2 #4 #5 #6.
- Nathan narrowed the immediate work to deprecation notices and configurable warnings for release 18 in #9; Tom Lane subsequently opposed direct pg_authid updates as a compatibility workaround and favored warnings followed by eventual outright removal, with Daniel Gustafsson and Andrew Dunstan supporting that direction #14 #15 #18.
- Nathan's first patch added documentation notices and CREATE ROLE/ALTER ROLE warnings controlled by md5_password_warnings, deliberately omitting authentication-time warnings as too noisy; subsequent revisions documented the setting, simplified passwordcheck tests, and rebased the patch #19 #20 #21.
- Greg Sabino Mullane approved the patch after applying it, and Jim Nasby requested another warning in the pg_hba.conf documentation, which Nathan added in the final submission #22 #23 #25.
- Cleartext-password deprecation remained outside the patch: Tom explained that PAM, BSD, LDAP, and RADIUS also require cleartext password transmission, and Nathan deferred that question to another thread #24 #25.
- After compiling and checking the final patch, Greg questioned whether its migration hint promised more documentation than existed; Nathan favored retaining the hint and expanding the documentation later, and Greg agreed #27 #28 #29 #30 #31.
- Nathan committed #25 unchanged apart from rebasing as db6a4a985b and announced completion in #32, leaving expanded SCRAM migration guidance as follow-up work and the eventual MD5 removal schedule unsettled.