Bayes problem with MailScanner

[Jan-Peter Koopmann]

Hi Julian,

> Does it change if you do a
>          sa-learn --rebuild

Nope but we are getting close. To be exact: I have it working here now
but I need you to verify something for me.

I noticed that I have four databases in my /var/spool/spamassassin file

bayes_seen
bayes_seen.db
bayes_toks
bayes_toks.db

I always ignored the ones without .db. The strange thing is that
sa-learn and check_bayes_db always worked on the files without .db but
spamassasin called via MailScanner always used the ones with .db. Since
the one with .db only greq via auto-learn though, there were only 53
spams in there indeed and bayes did not work.

I linked the files now and suddenly it works. I get BAYES scores in the
headers and everything. I looked through the SA source code and could
not find any reason for this behaviour. Especially why spamassassin -t
uses bayes_toks and MailScanner uses bayes_toks.db but hey... I am no
perl guru.

So please have a look at your database dirs and see if you have
"duplicates" as well. This would explain it then.

Regards,
  JP