[Mailman3-dev] anti-spam related features

[Barry Warsaw]

On Tue, 2004-03-16 at 10:02, Ian A B Eiloart wrote:
> --On lunes, 15 marzo 2004 17:28 -0500 Barry Warsaw <barry at python.org> wrote:
> 
> >
> > I'd also like to see some integration efforts with Spambayes.  Getting a
> > generic interface and U/I for this stuff is going to be tricky.
> 
> I don't know if this is generally the case, but our spam filter 
> (bogofilter) adds a header, which can be easily parsed. A generic interface 
> would just match a specified header with a regular expression to extract 
> some number - that's a site configuration, so U/I isn't that important.

I'm definitely thinking that we'll want spam controls to follow a
similar pattern we developed in the sprint for the flow of options and
user control (yes, Maki, I know I still need to scan those diagrams in!
:)

The idea then is that you might allow each of the site, vhost, list, and
user to define their own thresholds above which the message would be
considered spam.  Certainly, header regexp matching is very doable,
although we have to be careful of giving this control to the users. 
First, evil regexps can be constructed that can tank performance. 
Second, regexps are incomprehensible to most non-programmers (and many
programmers too!).

> You'd then want to convert those numbers to, say, a spamicity percentage.

Note that the ASRG is working on some standards in this area, and though
I'm on the list I haven't been paying very close attention to it.  In
general, if standards are established, Mailman should grok them.

> We should allow people to get as much spam as they want. We should let them 
> choose low thresholds, but not so low that they miss a lot of genuine 
> posts, and maybe there should be no spam filtering on closed lists at all.

OTOH, we have to decide how much spam defenses should be built into
Mailman.  Certainly I believe the MTA should be doing a huge amount (if
not most) of the spam filtering before it even gets to Mailman.  Where
Mailman and list admins need to get into the act is in the murky
"unsure" area, where for example, a message may or may not be spam to a
site-wide filter, but would definitely be on or off topic for a
particular list.

-Barry