Created attachment 1869 [details]
Spamd debug log.

Debug dump from "spamd -D -c -m5 -H -x -q". Includes startup and reciept of 2
messages, one of which enters/updates a bunch of tokens under the root user, as
apposed to the user spamc was called with.

Got the per-user bayes stuff to start showing up in the database by hacking the
handle_user_sql function in spamd.

sub handle_user_sql {
  my ($username) = @_;
  unless($spamtest->load_scoreonly_sql(@_)) {
    $spamtest->signal_user_changed ({ username => $username,
                                      user_dir => "" });
    return 0;
  }
  $spamtest->signal_user_changed ({ username => $username,
                                    user_dir => "" });
  return 1;
}

(Note: I don't claim to be a developer, but the hack seems to work)

This however, brings up a bigger issue, when combined with Auto-Learn, in that
it will auto-learn and plug stuff into the database for every incoming email
when used from something like qmail-scanner on the front end system-wide. So
even non-existent accounts get plugged into the bayes database. There should
probably be some check with either a vpopmail lookup, or a system lookup to
verify an account before logging bayes stuff maybe?

Also, there's currently no way I can see to get ham into the bayes database on a
per user basis. It looks like to fully make use of the bayes database stuff a
flag is needed on sa-learn to give it a user.

Subject: Re:  Spamd not using per-user Bayes settings

On Thu, Mar 25, 2004 at 02:28:26PM -0800, bugzilla-daemon@bugzilla.spamassassin.org wrote:
> 
> ------- Additional Comments From jeno@finali.com  2004-03-25 14:28 -------
> Got the per-user bayes stuff to start showing up in the database by hacking the
> handle_user_sql function in spamd.
> 
> sub handle_user_sql {
>   my ($username) = @_;
>   unless($spamtest->load_scoreonly_sql(@_)) {
>     $spamtest->signal_user_changed ({ username => $username,
>                                       user_dir => "" });
>     return 0;
>   }
>   $spamtest->signal_user_changed ({ username => $username,
>                                     user_dir => "" });
>   return 1;
> }
> 

Yeah, but the real question is why was it not already signaling that
the user had changed?

Anyone know?

> This however, brings up a bigger issue, when combined with Auto-Learn, in that
> it will auto-learn and plug stuff into the database for every incoming email
> when used from something like qmail-scanner on the front end system-wide. So
> even non-existent accounts get plugged into the bayes database. There should
> probably be some check with either a vpopmail lookup, or a system lookup to
> verify an account before logging bayes stuff maybe?

Hmmm...this is an interesting problem, can you open up another bug so
we can track it separately.

> Also, there's currently no way I can see to get ham into the bayes database on a
> per user basis. It looks like to fully make use of the bayes database stuff a
> flag is needed on sa-learn to give it a user.

Yeah, I think there is a bug somewhere, you can cheat by setting
bayes_sql_override_username to the username you wish to train for.

Michael

Went ahead and added call to signal_user_changed for handle_user_sql and
handle_user_ldap.  I couldn't find/think of a reason why we shouldn't call them.

Committed revision 9785.