Henry Kwan wrote: > Hi, > > Am running 3.18 on a vanilla CentOS 4 box and tried to upgrade to 3.20. The > make/make test/make install cycle went fine but I noticed that the scoring is > slightly different between the two version. I sent myself an email and here's > how they scored: > > 3.18: > > X-Spam-Status: No, score=-104.4 required=7.0 > tests=AWL,BAYES_00,FH_RELAY_NODNS, > SPF_PASS,USER_IN_SPF_WHITELIST,USER_IN_WHITELIST autolearn=no > version=3.1.8 > > 3.20: > > X-Spam-Status: No, score=-3.1 required=7.0 tests=AWL,BAYES_40,FH_RELAY_NODNS, > RDNS_NONE,USER_IN_WHITELIST autolearn=no version=3.2.0 > > The email was exactly the same. The only difference was one was sent 10 > minutes > later. Did the Bayes scoring change that much between 3.18 and 3.20? No, bayes scoring did not change *THAT* much. The big difference here is no USER_IN_SPF_WHITELIST. That's a -100 point rule, so it accounts for the largest part of the score difference.
That said, in the first case, theoretically that message had a -200 point advantage, due to both whitelists firing, but only wound up with -104. Sounds like your AWL for that user is raising the scores quite a lot. > Also, it > looks like 3.20 SPF checking is broken. How can I debug where it broke? Start with spamassassin -D <somemessage.txt , see if the debug output explains it.
