I wrote this a while ago, for dspam v. 2.10, and things have changed a bit
since then.  In particular, the ./configure options
--enable-delivery-to-stdout --enable-spam-delivery no longer exist, and
you need to change --deliver-spam to --deliver=innocent,spam when invoking
it.

I'm still getting well over 99.9% accuracy with my current setup which is
why I'm loathe to change it.  When I do I'll rewrite this.  In the
meantime, email me if you have any trouble following this.

-- Dickon Hood, 20041114.



Using dspam for a single user
=============================

All errors in this documnt are mine, Dickon Hood <dickon@fluff.org>, and
please bear in mind I wrote it from memory a couple of days after the fact
in a situation marginally different to the one here, due to our setup at
work.  Any questions, let me know.


dspam is aimed at large-scale implementations, but that doesn't mean it
has to be installed like that.  If you're not root on your machine, but
need some good spamfiltering, dspam as a single-user is a good thing.

It's not that difficult.

First, you'll need to choose your DB format.  I'm always *very* wary of
BerkeleyDB, as I've had a bunch of DBM files get corrupted, and programs
using them sit and chew all the RAM and swap in the machine, and leave it
very unhealthy.  Not much of a problem on a workstation (as you can
usually spot it going mad and kill it before it's done too much damage),
but on an unattended server, it's an issue.

If you have a proper database around, I suggest you use it.  I went to the
effort of compiling and installing MySQL (because I could, because it isn't
that hard, and because I've not done it before and rarely have an opportunity
to fiddle with databases).

Compiling dspam isn't difficult.  Amongst other options, I chose:

./configure --with-userdir=$HOME/.dspam --with-userdir-owner=none \
--with-userdir-group=none --with-dspam-mode=none --with-dspam-owner=none \
--prefix=$HOME --enable-spam-delivery --enable-delivery-to-stdout

but you may also want to set --enable-whitelist, --enable-experimental,
and suchlike.  If you have ~/bin and ~/lib directories, you may as well
set prefix to $HOME as it'll put all binaries and libraries into bin and
lib under it.  Your userdir may as well be ~/.dspam or something similar
as you'll be the only user using it.

Build it with 'make && make install'.  On Solaris with MySQL this may fail
when building libdspam.la (IIRC) as it fails to link against the
libmysqlclient.a properly, and spits out some 5600-odd (!) errors at you.
'make 2>&1 | head' will give you the commandline that fails.  You'll need
to look for the './buildtool' line, and C&P it into the shell.  Change the
'-L/path/to/mysql/lib -lmysqlclient' bits into an explicit
'/path/to/mysql/lib/libmysqlclient.a'.  Google was spectacularly unhelpful
when trying to debug this; I hope this helps someone...

If you're using MySQL, you'll need to setup a MySQL user, and set a
password on your root account.  Start mysql, login as the root user, and
type:

CREATE DATABASE dspamdb;
GRANT ALL PRIVILEGES ON dspamdb.* TO dspam IDENTIFIED BY 'password';
GRANT ALL PRIVILEGES ON dspamdb.* TO dspam@localhost IDENTIFIED BY 'password';
USE dspamdb;
SOURCE /path/to/dspam/source/tools.mysql_drv/mysql_objects.sql.space.optimized;

to create the database and install the schema.  Create your mysql.data
file in ~/.dspam/ as detailed in the docs; use the username 'dspam' and
whatever you used as the password.

If you're not using MySQL, then all that was irrelevant to you.

Feed your dspam a mail to check it works.  Hopefully it should.  Run
dspam_stats to check that it's learned a spam / ham.

To actually use it, you'll need to feed it a corpus of ham, a corpus of
spam, and you'll need to be able to spawn it automatically on receipt of
an email.  I run it from procmail, delivering to stdout, and I filter spam
based on the headers dspam outputs.

Launch procmail by putting:

"|/path/to/procmail"

in your ~/.forward file.  Stick the following in your ~/.procmailrc file:

------8<------

# Begin spam treatment.
:0fw
| dspam --user username --stdout --deliver-spam

:0:
* ^X-DSPAM-Result: spam
spam.inbox
# End spam treatment.

------>8------

and you'll find your spam is delivered to a mailbox called 'spam'.

To train a spam, pipe the spam through 'dspam --addspam', and a
false-positive through 'dspam --falsepositive'.  You can set up
keybindings in mutt if you so wish, but as it's one of the very few things
I need pipes for, I generally just use its use of getline and hit
uparrow-return rather a lot.

With this setup, you don't need the quarantine (with attendant webserver),
you don't need the added aliases for spam / ham misclassification (per
user), but you do have to know how to pipe mail through it.  This probably
won't be much of a problem for most people.

And that should hopefully be it.  It's fairly simple in concept, you just
need to think a bit.

Any questions, drop me an email.