Scoring unwanted DSNs (bug 4029)

["Jeffrey W. Baker" <jw...@acm.org>]

There's a bug (4029) which is resolved as WONTFIX but which I think
deserves more attention.  The problem is with DSNs and bounce messages
which are in reply to mail which was never sent.  In various places this
is called backscatter, collateral damage, whatever.  I call it
irrelevant daemon mail.

I think there is a pretty good way to solve this problem.  The message
can be initially flagged as a bounce message/DSN, which is pretty easy.
If the message body or attachments makes reference to a message ID which
is in the user's FCC (sent folder) then the mail is tagged with
DSN_RELEVANT.  If the message body or attachments reference a message
ID, but the message ID does _not_ appear in the user's FCC, then the
message gets to be DSN_IRRELEVANT.  Otherwise the message will be
DSN_UNKNOWN.  The user can then apply the desired score to these (my
choice would be -5, +5, and 0 respectively).

The only hitch is how to access the user's FCC.  Personally, I use
dovecot, so I can access my by forking /usr/lib/dovecot/imap and using
this very simple code:

            $c = Mail::IMAPClient->new(
                Socket => *DOVECOT
              , State  => 'Authenticated'
            );

            $c->select($fcc_folder_name);

            foreach my $id (@msgid) {
                my ($res);

                $id =~ s/["'<>\r\n]//g;

                $res = $c->search('HEADER', 'Message-Id', qq/<$id>/);

                if (defined($res) && $res > 0) {
                    # Do something smart here
                    last;
                }
            }

            $c->logout();

Pretty straightforward?

So the main issues are how to allow the user to specify the means of
accessing the FCC (via IMAP would be my first implementation, but I'd
allow for development of other ways ranging all the way down to
grep(1)), and, since I'm completely new to this SpamAssassin thing, how
to safely fork the subprocess as that user.  That's allowed, right?

I do realize that there exist some custom rules for handling virus
reports and the like, but I like the exactness of this implementation a
bit better.  By the way, I've been using this system, via procmail
outside of spamassassin, for years and it works great.

Your thought will be quite welcome.

-jwb