> >0.000 0 2 0 non-token data: bayes db version > >0.000 0 1 0 non-token data: nspam > >0.000 0 0 0 non-token data: nham > >0.000 0 116 0 non-token data: ntokens > > ---snip--- > > > >Its as if the only updates were from auto-learn. What am I missing? > > You have per user bayes databases perhaps ? And you're running sa-learn as > root, and not the user that spamassassin/spamc is called under. You might > need to specifiy the location of your bayes database. > > I use a global bayes database with the database path/file specified in > local.cf so no matter what user I call sa-learn as its accessing the same > database...
Very interesting. I think you are onto something. I am running spamd in a separate server. I point my .procmailrc at the spamd server. This way I can test 2.60 without unleashing it on my users. I do my sa-learn updating via ssh, something like: scp my_pine_folder spamd_server: ssh spamd_server /usr/local/bin/sa-learn --ham --mbox ....... ssh and rsh don't always establish the environment properly. Evidence of this is that if I run my proc by hand, on spamd_server, everything works fine. If I run it via ssh, it doesn't. What is odd is that the time stamp on the bayes files is updated, and rerunning sa-learn clearly indicates that the messages have been learned, but the counts don't update. So right now my guess is that something in --rebuild is more path sensitive than --ham or --spam. I've added an explicit "bayes_path" to my user_prefs, and I've added a couple of "sa-learn --dump magic" statements to my proc. We'll see where it goes. -- Jack Gostl [EMAIL PROTECTED] ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk