http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5185
Summary: Bayesian learning uses different message checksums
during exiscan_acl and later sa_learn
Product: Spamassassin
Version: 3.1.7
Platform: Other
OS/Version: other
Status: NEW
Severity: normal
Priority: P5
Component: Learner
AssignedTo: [email protected]
ReportedBy: [EMAIL PROTECTED]
This is a bit of a crossover bug between spamassassin and exim.
I'm using the exiscan support to do my spam scanning at SMTP time with exim.
This works well, however it seems that message checksums used to establish
whether or not a message has already been learnt is different between scanning
at SMTP time and scanning after delivery.
The implication is that should a SMTP-time scan result in an autolearn which I
later wish to revert (or, more often, run a learning process over a batch of
messages without relearning those which have already been autolearnt), the
message is always just relearnt.
After a considerable amount of digging, it appears this is due to a difference
in the top Received header. From
http://www.exim.org/exim-html-4.63/doc/html/spec_html/ch43.html#id2689094:
Once a message is accepted, the timestamp in the Received: header line is
changed to the time of acceptance ...
Thoughts from the spamassassin community would be appreciated on whether:
(a) Bayes.pm should ignore the timestamp on the Received: header when
calculating checksums
(b) exim shouldn't be rewriting the timestamp between the first pass and
delivery
(c) what I'm doing is silly/it doesn't matter anyway.
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.
