Hi all, Following up on Stage 1a (PR #3310). I realize the diff looks heavier than the core change is: the source is ~900 lines, but the unit tests plus the by-name mapping to upstream test vectors roughly double that, which makes it a daunting review. I'd like to cut that load down. A few options, happy to do any or all:
1. Split Stage 1a into three small, standalone PRs by module, in dependency order: - mysql_auth_packet (~220 lines) - length-encoded int/string and packet-header helpers, no crypto. Reviewable in minutes and a good first merge. - mysql_auth_scramble (~320 lines) - the clean-room scramble/auth crypto. This is the license-sensitive part flagged on #2093, so it gets its own focused review. - mysql_auth_handshake (~360 lines) - depends on packet. 2. Move the 75-test by-name upstream mapping table out of the PR description into a short README under test/mysql_auth/, leaving just summary counts in the PR. The table is provenance documentation, not code, and right now it makes the PR read like a compliance audit. 3. Add an integration test that brings up a real MySQL server in CI and exercises the handshake end to end against it. That gives far more confidence in wire-correctness than the hand-written known-answer vectors do, and it lets me trim the exhaustive unit vectors down to a representative core. Is there one part that's the actual bottleneck for reviewers - test volume, the clean-room provenance, or CI time - so I can prioritize? And the earlier ask still stands: would a PMC member be able to assign a Stage 1 shepherd from the rotation (wwbmmm the natural fit, but anyone is welcome)? I'll do whatever lowers the load on reviewers. Thanks, Varun Raj --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
