bmf not unlearning

Hi,

I used BMF almost 10 years ago and I am getting back to it very 
recently.  I now need a bayesian filter to enpower my procmail MTA again 
and I though, why the hell wouldn't I use just the same one I was using 
before?  In fact I am quite happy with it, for two or three weeks now, 
but I encounter an issue about some messages that just won't be 
unlearned.  They keep being recognized as spam and I find it quite strange.

I have three mboxes being used by BMF:
   _bmf.learn (crontab processes it nightly with bmf -s
   _bmf.spam (procmail puts the spam in here depending on what bmf decides)
   _bmf.unlearn (crontab processes it nightly with -n or -N, both fail 
in some cases)

It seemed to me that -n and -N did work at the beginning, and maybe 
still work partially, but some emails just keep being identified as spam 
falsly.

Note that the bmf -n and -N commands don't even exit properly (I can't 
get the same exit code I get with bmf -s to let my nighly shell script 
tell me everything went through).

Maybe I should have a look into ~/.bmf/*db* to try to analyse what's 
working or not but I don't know how to read that content, probably a 
command line associated with db4.  The permissions are fine for that dot 
folder.

I am using CentOS6 and db 4.7.25 on my mail server.

Thanks,
don't hesitate to ask if I missed some required information
Pierre-philipp

------------------------------------------------------------------------------
Everyone hates slow websites. So do we.
Make your web apps faster with AppDynamics
Download AppDynamics Lite for free today:
http://p.sf.net/sfu/appdyn_d2d_feb