dspam auto-learn with amavisd-new 2.8.0

[Georg Lindner]

Am 01.10.2012 17:21, schrieb Georg Lindner:
> We recently upgraded from amavisd-new 2.6.1 to version 2.8.0, and are now trying to get dspam integration (including autolearn) working together with the new amavis setup.
>
> Our dspam entry in the amavis.conf:
>
> ['DSPAM',  'Amavis::SpamControl::ExtProg', 'dspamc',
>    [ qw(--client --stdout --deliver=innocent,spam --user), $daemon_user ],
>    learn_ham  => [ qw(--class=innocent --source=error--user), $daemon_user ],
>    learn_spam => [ qw(--class=spam --source=error --user), $daemon_user ],
>    mail_body_size_limit => 65000, score_factor => 1.00
> ]
>
> According to the amavis log the auto-learn process for dspam is called:
> Sep 26 23:38:33 srv1 amavis[5475]: (05475-05) DSPAM result: Spam, score=1.000, sig=506375d921989068461704
> Sep 26 23:38:36 srv1 amavis[5475]: (05475-05) SpamControl: scanner DSPAM, auto-learn as ham / 0.389 (was: spam / 1.000)
>
> The dspam auto-learn is based on the X-DSPAM-Signature header. But it looks like all the X-DSPAM-* headers are not passed over the the autolearn process, so the messages are not reclassified.
> Neither as spam nor as ham.
>
> When a message is deliverd to the user later on the X-DSPAM-* headers are present.
>
> Is there a way to pass the "X-DSPAM-Signature" header to the auto learning?
> Either in-line in the message header, or as a parameter --signature=
>
> Thanks in advance
> Georg
>

As I found no resolution to the above problem I am now using a custom
hook to train dspam. This works quite nice but now I am facing the next
problem...

How can I disable the amavis built-in autolearn mechanism for dspam?
I've tried to remove learn_ham and learn_spam from the configuration,
but apparently it doesn't disable anything.

Any ideas?

Georg