On Sat, 6 Nov 2010 16:17:43 +0100 Oliver Block <oliver_blo...@web.de> wrote:
> Hello, > > I run into a problem calling sa-learn from mutt. This problem was > reported at least two times in the past, but no solution to it > was posted: > > http://mail-archives.apache.org/mod_mbox/spamassassin-users/200406.mbox/%3c20040611085128.gc14...@kassi.oddi.is%3e > http://mail-archives.apache.org/mod_mbox/spamassassin-users/200404.mbox/%3cs090db0a....@dsbfs2.denmarkstate.com%3e > The common symptom here is that it's failing to learn correctly because all the message ids are the same: 25587ed3dfaa6532fc9e20177019ceec3ba0b...@sa_generated Modern versions of SA identify emails by a hash of certain headers and the first n bytes of the body. From a bit of experimentation it appears that 255...0a7 is what you get if there are neither any relevant headers nor a body. e.g. echo | sa-learn -D bayes --ham 2>&1 |grep "sa_generated" Nov 6 18:07:08.289 [61897] dbg: bayes: 25587ed3dfaa6532fc9e20177019ceec3ba0b...@sa_generated already learnt correctly, not learning twice
