I'm worried. Whenever I feed a message with "autolearn=spam" or "autolearn=ham" to sa-learn --forget, I get "Forgot tokens from 0 message(s) (1 message(s) examined)" back. That's bad, because it means that the net effect of re-learning a spam incorrectly learnt as ham is one spam occurrence and one ham occurrence of each token, instead of just one spam occurrence.
Indeed, when I did "spamassassin -D bayes < testmessage" the debug output reported learning from a different "@sa_generated" message ID than "sa-learn -D bayes --forget" said it was trying to forget (but didn't find). AFAICT from reading the source, get_msg() in Mail::SpamAssassin::Bayes is used in both cases. So why does it make up different IDs? -- Magnus Holmgren [EMAIL PROTECTED] (No Cc of list mail needed, thanks)
pgpXUlYVd2XFV.pgp
Description: PGP signature