[Mimedefang] Spamassassin public corpus not scanning

[Nels Lindquist]

On 1 Mar 2004 at 10:40, Gwendolynn ferch Elydyr wrote:

> When I feed spamassassin's public corpus through, I'm seeing a lot of
> messages which are clearly spam being passed through without being marked
> as spam, although the header shows mimedefang processing.  Has
> anybody else had this experience?

AFAIK, if the MIMEDefang header has being added, then milter 
processing has taken place.  Why are you feeding SA's public corpus 
through?  If it's for Bayes training, it'd be better to use sa-learn 
on the commandline.  If you're just trying to verify spamassassin is 
working, try using the GTUBE test.
 
> I've got the filter timeouts set fairly high:
> 
> 	T=C:5m;S:5m;R:5m;E:5m

Indeed; the MIMEDefang recommended settings are T=S:60s;R:60s;E:5m.  
Are you seeing any log messages which lead you to believe the filter 
is timing out?  Are you feeding messages through fast enough to 
exhaust the maximum available mimedefang slaves?

> ... but I do have sendmail set to pass messages onwards when it doesn't
> hear from the filter.  The number of messages which aren't marked as spam
> doesn't seem to have any relationship to the rate at which the messages
> are arrived, either.

Okay, so some of the messages *are* marked?  In that case, why not 
add a header which includes the SA tests which were triggered by the 
message, regardless of its spam status?  Then you can figure out why 
a particular message isn't marked as spam.

Something like this should work:

> my ($header) = "X-Spam-Status";
> my ($value) = $hits >= $req ? 
>     "Yes, hits=$hits required=$req tests=$tests" : 
>     "No, hits=$hits required=$req tests=$tests";
> action_change_header($header, $value);

----
Nels Lindquist <*>
Information Systems Manager
Morningstar Air Express Inc.