Hello Karl Berry!

Karl Berry wrote in
 <[email protected]>:
 |Hi - in junk.c on the heirloom branch of s-nail is some interesting code to
 |do both the two-token Bayes spam classification via the
 |chained-junk-tokens option (as well as the usual single token), as
 |documented in mailx.1 there. (Distributed as mailx in CentOS7.)
 |
 |The main thing I'm wondering is whether there is any command to purge
 |the junkdb in some way. I could not find anything about this, either in
 |the code or the doc, but I know I could easily be overlooking something.
 |Otherwise, growing without bound does not seem good ...
 |
 |The other thing I'm wondering is why this code is apparently moribund?
 |Seems it hasn't been touched for ~10 years.
 |https://sourceforge.net/p/s-nail/code/ci/heirloom/tree/
 |I couldn't find anywhere (e.g., someone on github) that the code had
 |been taken up and currently maintained, although it's not the easiest
 |thing to search for.

Yes, the junk code was one of the first things i have removed once
i effectively forked the project, if i recall correctly.  Whereas
Gunnar Ritter implemented some really interesting or neat things
(local IMAP cache, MD5 token fingerprint based not full-text junk
DB storage) i am the opinion that this MUA has to (have the
possibility to) integrate into an environment.

The junk database could not become shared with other programs, so
if people switch to say mutt, or simply want to preprocess their
mails with procmail or fdm, then these DBs were separate, which
counteracts the idea of fighting spam (i mean, i think this is one
of the great things with Google spam fighting, that so many people
anticipate and that the checksum of billions of mails is inspected
as they enter the Google big data mail system all over the world).

Also, storing only (parts of) fingerprints is neat, but simply
placing the database on an encrypted storage is what i prefer.
With the bogofilter i use i _could_ look at statistics etc.  I can
export and import data, shall i ever feel the desire to use
a different spam detector.  All this was not possible.

I do not know, actually, how good bogofilter really deals with
HTML (and even, at least in parts, XML, if you have to deal with
senders which use Microsoft Word i think) mails, and attachment
content, and how it handles cases where there are super
complicated tables and kilobytes of style and other directives
which then possibly even get replicated in the actual content, if
there is any :), all this is complicated.  But i rather use
a dedicated program to address this problem.

 |The s-nail master branch has plugins to allow using "outside" spam
 |filters, the usual spamassassin and bogofilter being mentioned, but
 |these do not support multiple-token Bayes as far as I know. In fact, I
 |do not know of any practical system (as opposed to academic examples or
 |defunct code, namely dspam) that supports multiple-token Bayes. If there
 |is anything out there, I'd love to know.

I have no idea, actually.  The paper is no longer available under
the linked URL.  I think i recall that i read it once, but this is
fuzzy.  Anyhow, i personally think deep analysis is the only real
option, after text normalization.  I do not think bogofilter deals
with this very well, i learn it a long time, and it handles most
perfect, but each and every day i have to add some, and it is
likely they are designed to contain a lot of good stuff,
explicitly.  I hit the "S"pam button and hope we get it over time.

 |Thanks for any information, and all your work. 

Oh, so much to do still!  But i will come to roff, one day!
Ciao from Germany!

 |--best, karl (karl at freefriends dot org).
 --End of <[email protected]>

--steffen
|
|Der Kragenbaer,                The moon bear,
|der holt sich munter           he cheerfully and one by one
|einen nach dem anderen runter  wa.ks himself off
|(By Robert Gernhardt)

Reply via email to