[Dovecot] SpamAssassin and Maildir with Dovecot
A patch has been proposed for SpamAssassin to process Maildir folders of spam:
https://issues.apache.org/SpamAssassin/show_bug.cgi?id=3003
Is this patch compatible with Dovecot's implementation of Maildir? For example, is anything needed to avoid stepping on Dovecot's metadata?
Kenneth Porter wrote:
A patch has been proposed for SpamAssassin to process Maildir folders of spam:
https://issues.apache.org/SpamAssassin/show_bug.cgi?id=3003
Is this patch compatible with Dovecot's implementation of Maildir? For example, is anything needed to avoid stepping on Dovecot's metadata?
I always used sa-learn with maildir (without the patch) with no problems. the only thing the patch adds is to visit the new/ and cur/ subdirs. thus no problems.
On Fri, 2008-08-15 at 09:57 -0700, Kenneth Porter wrote:
A patch has been proposed for SpamAssassin to process Maildir folders of spam:
https://issues.apache.org/SpamAssassin/show_bug.cgi?id=3003
Is this patch compatible with Dovecot's implementation of Maildir? For example, is anything needed to avoid stepping on Dovecot's metadata?
Does it modify any messages? If not, there shouldn't be any problems.
--On Tuesday, August 19, 2008 6:08 AM +0300 Timo Sirainen tss@iki.fi wrote:
Does it modify any messages? If not, there shouldn't be any problems.
No, it just identifies which files are messages and reads them to train SA's Bayes database. The patch just does better identification.
I typically run sa-learn from a cron job a few times a day. I have "Uncaught" and "FalsePositives" folders and run it against those to correct the automatic training that happens on delivery.
Timo Sirainen wrote:
On Fri, 2008-08-15 at 09:57 -0700, Kenneth Porter wrote:
A patch has been proposed for SpamAssassin to process Maildir folders of spam:
https://issues.apache.org/SpamAssassin/show_bug.cgi?id=3003
Is this patch compatible with Dovecot's implementation of Maildir? For example, is anything needed to avoid stepping on Dovecot's metadata?
Does it modify any messages?
no, it's a read-only operation. it does not re-evaluate the message nor mark it as trained (it stores the message-id in the bayes db for this).
If not, there shouldn't be any problems.
participants (3)
-
Kenneth Porter
-
mouss
-
Timo Sirainen