Bug in spamassassin 3.3.0

[aus-city]

Scott,

Just so your aware spamassassin have done something in the latest spamassassin.

It works fine except with psa in spam filtering / training.

The messages in training are all gone, despite the mailbox still has messages in it (yes I know if the mailbox is empty the bayes count is shown 0), but this behaves like the mailbox is empty when it's not.

It shows 0 learn. However bayes works as new emails trigger size and timestamp changes in the users bayes files.

So it's something psa-spamassassin is doing in relation to spamassassin.

I tried not only your package, but F11 one from updates-testing both the same.

Roll back to last version and immediately I see messages in training.

Any ideas?

[aus-city]

Update

Scott I tracked down this bug

I found its two perl files in the package:

/usr/lib/perl5/vendor_perl/5.10.0/Mail/SpamAssassin/Bayes.pm

and

/usr/lib/perl5/vendor_perl/5.10.0/Mail/SpamAssassin.pm

I rolled back these two perl files from the last working 3.2.5

So seems there are perl changes here that break psa-spamassassin.

Its still not right as if I try to train any message it then gives me a 0 messages screen.

Could you run the new spamassassin and use the old perl files?

[scott]

You can, but the odds are really high that a lot of tests are no longer working. SA is really good about failing "safe" when something internally breaks (a la CPAN). Honestly you'd have to compare the same spammy message against it in both conditions to really know whats working and what isn't.

[aus-city]

I thought as much and stuck with 3.2.5

Its bad that 3.3.0 won't play with plesk spam training. The ease of in plesk tweaking bayes is so nice.

I am not holding my breath that what ever is broken in the new perl libraries will be rectified as it's already rc1.

Will take parallels until plesk 10 to get that working

Anyway just be aware that 3.3.0 does break psa spam training, so 3.3.0 should not make stable until parallels fix plesk.