Package: spamassassin Version: 3.1.0a-1 Severity: normal Hello Duncan, Since I have upgraded to 3.1.0a-1, bayesian filtering is far less efficient than with the sarge version or 3.0.4.
Symptoms are a ton of spam with BAYES_50, sa-learn not increasing the bayesian probability most of the time (but some time it does) seen by spamassassin -t, and an overall increase of false-negative. I tried sa-learn --import, but it was apparently a noop. sa-learn --lint did not complain either. I installed a sarge chroot to test: Sarge sa-learn and spamassassin seems to perform much better with the same spams and the same bayesian database and the same configuration. In case it can be useful: %file .spamassassin/bayes_* .spamassassin/bayes_journal: ASCII text .spamassassin/bayes_seen: Berkeley DB (Hash, version 8, native byte-order) .spamassassin/bayes_toks: Berkeley DB (Hash, version 8, native byte-order) This spamassassin is used to take care of my own (non-Debian) debbugs server and increase of false-negative is a immediate problem. Cheers, Bill. -- System Information: Debian Release: testing/unstable APT prefers unstable APT policy: (500, 'unstable') Architecture: i386 (i686) Shell: /bin/sh linked to /bin/bash Kernel: Linux 2.4.27-skas-5 Locale: LANG=fr_FR, LC_CTYPE=fr_FR (charmap=ISO-8859-1) Versions of packages spamassassin depends on: ii libdigest-sha1-perl 2.10-1 NIST SHA-1 message digest algorith ii libhtml-parser-perl 3.45-3 A collection of modules that parse ii perl 5.8.7-5 Larry Wall's Practical Extraction Versions of packages spamassassin recommends: pn libnet-dns-perl <none> (no description available) ii perl [libmime-base64-perl] 5.8.7-5 Larry Wall's Practical Extraction pn spamc <none> (no description available) -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]