[Mimedefang] Can I Use Spamassassin Score Instead of MIMEDefang Score?

[Kelson Vibber]

At 11:08 AM 9/10/2003, Jason Granat wrote:
>So I tried Chris Meyer's suggestion and also found a section in the 
>README.SPAMASSASSIN file about how to get MD to use SA.  MD is scoring 
>higher now but still averages a few points lower than SA.  Is there any 
>way to just force MD to use the SA score instead of it's own score? SA's 
>scores have been working so much better for so long.  That brings up 
>another question, could the scoring be different due to spam learned or 
>does MD take that into account since it's told to use SA?

1. MD *does* use the SA score.  Always.  It calls SA, SA gives it a score, 
and MD uses that.  At the time MD runs, the only instance of SA that has 
looked at the message is the one MD calls, and that's where it gets the 
score.  Unless you've got multiple milters set up, the stand-alone 
SpamAssassin doesn't even get the message until MIMEDefang is already done 
with it.  There's no way to force MD to use something that doesn't exist.

2. Compare what tests get hit.  Have MD log the value of $names or add it 
to a header.  That way you can see if SA-by-itself called Razor but 
MD-calling-SA didn't, or if some RBLs didn't get tripped in some, etc.

3. If none of the network tests show up in the MD+SA scores, put 
$SALocalTestsOnly=0 in the beginning of your mimedefang-filter (and make 
sure you don't have $SALocalTestsOnly=1 later on)

4. If you're trying to use Bayes, remember that you can only use a single 
sitewide database.  This will often be less effective than per-user Bayes 
databases, but better than running SA without Bayes at all.  You'll need to 
configure SA to use a single database that the MD user can write 
to.  Search the archives for more info on that.

5. Older versions of MD do not reconstruct the latest set of Received 
headers when passing the message to SA, so SA will miss some 
information.  I forget what version fixed this, but it's probably best to 
grab the latest.

The most important bits are #1 and #2.  Until you can compare which tests 
are getting hit by each instance of SA, everything else is speculation.


Kelson Vibber
SpeedGate Communications <www.speed.net>