model-writtenrun 4505 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut began in #1 after FIPS-mode testing of column encryption exposed widespread test failures, proposing SHA-256 replacements for incidental MD5 use while flagging size-sensitive data, SSL certificates, and genuine MD5 tests for special handling.
- Peter separated the genuine MD5 function tests from strings.sql in #2; Michael Paquier approved the slice in #3, and it landed unchanged as 9786b89bd1, confirmed in #4.
- Alvaro Herrera questioned unexpected BRIN row-count warnings in #5, and Michael pursued the concern in #9 and #11; Peter explained in #12 that the original patch changed generated inputs without updating all expected results.
- Peter narrowed the next stage to core regression tests in #6, carefully recalibrating SHA-256-generated data; Michael found a missing largeobject_1.out update in #7, which Peter included in #8.
- Tom Lane supplied a different organization in #13: shared notmd5 helpers preserved MD5's output width, reduced unrelated test changes, and passed his Fedora FIPS testing apart from the intentionally unresolved MD5 and password tests.
- Daniel Gustafsson requested a name explaining the FIPS motivation in #14, Tom suggested fipshash in #15, and Daniel endorsed that name with explanatory comments in #16.
- Peter proposed FAKE_FIPS_MODE for easier local testing in #19, but Daniel warned in #20 that emulation could miss real OpenSSL behavior, so Peter left it available for ad hoc testing in #23.
- The fake-mode review nevertheless exposed misleading cryptohash comments, and Peter committed the independently useful correction, extended across initialization, update, and finalization, as 36ea345f8f, confirming it in #25.
- Peter committed the helper-based core patch from #13 as 208bf364a9, with the agreed fipshash name and light syntax and expected-output adjustments, then deferred the other suites to the next release in #27.
- After that release-cycle gap, Peter returned in #29 with the remaining citext, pgcrypto, authentication, SSL, and alternative-output work; he also included newly reintroduced BRIN MD5 calls as a completeness-only patch whose origin needed separate follow-up.
- Daniel reviewed the renewed series in #30 and requested explanatory comments around MD5 availability probes and consistent SCRAM capitalization.
- Tom found that older OpenSSL reported 'disabled for FIPS' while newer OpenSSL reported 'unsupported' in #31, then reported a newly disabled 3DES case in #32 while confirming that the other tested suites passed.
- The unchanged citext removal from #29 landed first as 3c551ebede, removing a test whose value was dubious because citext does not define its own MD5 function.
- Peter proposed proceeding despite difficulties reproducing the Red Hat environment in #35, and Tom accepted initially supporting only OpenSSL 3.x expected outputs in #36 while insisting that disabled 3DES should be covered.
- Daniel endorsed an alternative results file for disabled ciphers in #39; Tom's revised remainder in #40 supplied that file, isolated the MD5-specific PGP test, and proposed creating password-test roles separately from setting their passwords.
- Peter first committed Tom's PGP test extraction from #40 as 3af0d17ace, then committed the matching MD5 and 3DES alternative outputs as 795592865c.
- The TAP component of #40 landed as 284cbaea7c with formatting and Daniel's requested comments, while the remaining alternative outputs and Tom's password-test delta landed together unchanged as 3c44e7d8d4.
- Peter reported completion in #41; the supplied candidates establish the main landings, although none separately accounts for the completeness-only BRIN cleanup carried in #29 and #40.
- Thomas Munro later asked about regression-role naming warnings in #42, and Tom explained in #43 that the stricter naming policy mattered for installcheck rather than isolated TAP tests, which Thomas accepted in #44.
- At the thread's final message, #45, Tom reopened older-OpenSSL support after new SLES 15 buildfarm animals failed on the previously excluded error wording and proposed additional expected files.
- That proposal subsequently landed on the development branch as c7b0cb367d, with a865b654c7 representing the same change on the stable branch despite its duplicate candidate listing; the commit covers core regression tests and explicitly leaves possible contrib adjustments for further buildfarm evidence.
- Further buildfarm results led Tom to add pgcrypto MD5 and HMAC-MD5 expected files for older OpenSSL's 'No such hash algorithm' errors in e633fa6351, extending the work initiated in #45 without another attached patchset; both the core and pgcrypto follow-ups have now landed, with no further review or outstanding proposal recorded.