[TriLUG] Nifty procmail trick for today

[Jason White]

* Mark Shuford (davemarcus at pobox.com) [030518 23:15]:
> On Sun, 18 May 2003 18:57:16 -0700 (PDT)
> Turnpike Man <turnpike420 at yahoo.com> wrote:

> > I haven't learned it yet, but I believe SA has Hayes learning system which
> > could learn from your SPAM and HAM to make filtering even better.  Check our
> > archives, the discussions were fairly recent... last 1-2 months I think?  Sorry

> Would that be Bayes rather than Hayes. I've been using Spam Bayes,
> rather than Spam Assassin, for the last couple/three months.  It is
> purely Bayesian, as opposed (as I understand it) to Spam Assassin,
> which uses a bunch of different things, internal and external.

> I wasn't really happy with the idea of keeping up with SPAMsource
> lists, either dynamical/real-time or with periodically gathered
> static list. I also didn't really dig the idea of having to consume
> more bandwidth to do all the dynamic things, such a checking black
> hole lists and relay-host list, yada yada. 

> I've been happy with Spam Bayes. Having not use Spam Assassin I do
> not know if it would be even more happy-making.

> Has anyone out there use both packages? Any comments on how they
> compare?

I started off with SpamAssassin (SA), but I agree with Mark -- I
didn't like all the internal/external stuff.  But, the biggest problem
I had with SA was that it thrashed my mail server when a message came
in with a 20MB attachment.  I know there is probably some
configuration changes you can make to force SA not to search through
attachments, but I didn't keep it long enough to find out.  

I switched to SpamProbe (which is also 100% Bayesian, AFAIK), and I've
been happy for several months.  (It does require a lot of spam and ham
to feed it, initially).

Jason