Kevin A. McGrail wrote:
I'm working on packaging 3.4.1-rc1 and fixed a few bugs but It's not ready for announcement of voting. I will need to review if the rc1 is "burned" and we move on to rc2 but wanted to get this mailed out so people can look at things.
I'd like to get Bug 7106 resolved somehow, but haven't found time to work on it yet. Have some details in the pipeline to be committed (few small optimizations, some comment updates) - will commit these tomorrow. Another thing I have ready now (just needs some polishing) is adding message digests (hashes) of all MIME parts as tokens to Bayes. It's a relatively small change, will open a PR tomorrow. Currently Bayes is only seeing tokens from textual parts and ignores attachments, like pictures, icons, doc, pdf, zip, ... By adding hashes of all MIME parts, Bayes can correlate with spammy/hammy non-textual attachments too. This is based on an idea by Andreas Schulze, who first experimented with it in Amavis. Having it running here for two days now, these hash tokens are beginning to show among the top-5 hammy or spammy tokens (as reported by HAMMYTOKENS and SPAMMYTOKENS macros). Mark
