On Thu, 2004-12-09 at 12:17 +1300, Tony Meyer wrote:
> I would (personally) have preferred an optional change that let sb_filter
> add appropriate entries to the messageinfo database, but yes I think we'd be
> interested.  Please make sure that the diff is a context diff against
> current CVS (1.1's sb_imapfilter is very different to 1.0.x's).

Now that you brought it up, updating sb_filter to use the messaginfo db
looks like the better solution.  Unconditionally using the Spambayes-
Trained header as a fallback will prevent training a new database on
messages trained on an old database.  To handle this case, my original
changes added the -f (force) flag to sb_imapfilter, which worked as it
does in sb_mboxtrain, ignoring any previous training results.  If useful
I can submit that as well.

I'm submitted the patch for the header fallback with your issues
addressed.  It also fixes a small bug where the spambayes headers would
be lost if a message was untrained.  I'll look at implementing use of
the messageinfo db in sb_filter, which would solve a number of issues.
If I go forward with it, I think it will require an option(or at least a
cmdline arg) to choose the header or messageinfo db method, because at
least sb_mboxtrain uses the header method for retraining.  Any thoughts?


Patch URL(wordwrap will probably mess with it):
https://sourceforge.net/tracker/index.php?
func=detail&aid=1082344&group_id=61702&atid=498105


-- 
Todd Mokros


_______________________________________________
spambayes-dev mailing list
[EMAIL PROTECTED]
http://mail.python.org/mailman/listinfo/spambayes-dev

Reply via email to