Dovecot + KMail + Bogofilter

[David Relson]

On Mon, 4 Jun 2007 21:41:07 +0100
Anne Wilson wrote:

> I seem to have a problem with my imap Inbox.  If bogofilter has given
> a false positive I would expect to run the Mark as Not Spam filter (I
> run from a toolbar icon) then move it to Inbox.
> 
> At the moment, on this box (Mandriva Spring) moving it causes it to
> disappear, but it can be recovered with Edit > Undo.
> 
> If I attempt to copy it to Inbox instead, the same thing appears to
> happen, but Edit > Undo results in two copies in my local wastebin.
> 
> I can copy or move to another folder.
> 
> I suspect that this is a bogofilter problem.  I see the Inbox message
> count increase, then immediately decrease again.  It seems to be
> ignoring the fact that I had marked it Not Spam, and instantly marks
> it as Spam again.  If the appropriate place for the message is the
> Inbox, I'm stymied.
> 
> I don't recall seeing this behaviour before, so it may be confined to
> the version supplied with Mandriva Spring, or indeed to the package
> supplied.
> 
> Is there any way I can tackle this?
> 
> If it is likely to be a bogofilter bug, where should it be reported?
> 
> Thanks
> 
> Anne

Hi Anne,

Assuming you can modify the commands used to run bogofilter (for
"normal" and "Mark as Spam"), you should be able to learn more about
what is happening.

Add the "-l" flag to all invocations of bogofilter so that syslog will
track what's happening.  Also add the following to your bogofilter.cf
file:

  log_header_format = %h: %c, spamicity=%f, version=%v, queueID=%Q

These 2 changes should give you a record of what bogofilter is doing
and should help you track the cause of your problem.

Possibly what's happening is that KMail is looking at the message's
X-Bogosity line and that's not being rewritten.  Also, simply using the
"-Sn" flag combo to unregister as spam and register as ham does not
guarantee that the message will then register as ham.  A wordlist with
many messages in it has a lot of "momentum" can take a while to
"retrain".  Rescoring a single message with "-Ns" has a lot of
effect on a small wordlist but not on a large one.

HTH,

Regards,

David