On Wed, 6 Mar 2002, Matt Sergeant wrote:
> On 5 Mar 2002, Craig Hughes wrote:
[...]
>> Matt, take a look at bugzilla #62 -- there is more discussion of
>> exactly this there. If you re-order the rules, then the only problem
>> with short-circuit scoring is razor submission. If "-L" is used
>> though, this is irrelevant, you can just exit when the threshold is
>> exceeded. It would probably be good to indicate that score evaluation
>> was short-circuited in some header or other (probably just tack on
>> the X-Spam-Status) so people don't get confused. Also, you'll want to
>> make sure none of the "make test" stuff needs adjusting for the new
>> scores some of the mails will get.
>
> OK, all implemented and checked in (along with the other stuff (conf
> ||=, and subject in body). Please test CVS anyone who's interested in
> this.
I am somewhat dubious about the effects of the subject-in-body change.
It would be really good if someone (Craig ;) were to run SpamAssassin
over the corpus with and without it being make, then compare the
results.
I have a strong suspicion that it will result in odd effects and,
overall, reduce the effectiveness of the scoring. Rules that were based
on guesswork as to the header content or body content will now flap
wildly.
Worse, some rules will now score on more things and others less,
resulting (I suspect) in changes to the false positive/negative rates
or, at least, a difficulty in being able to apply them.
I think that a better solution would be to tag certain rules as 'header
or body' or 'subject or body' rules, then allow them to score each half
of that 'or' as a separate thing...
Daniel
--
The problem with defending the purity of the English language is that English
is about as pure as a cribhouse whore. We don't just borrow words; on
occasion, English has pursued other languages down alleyways to beat them
unconscious and rifle their pockets for new vocabulary.
-- James D. Nicoll
_______________________________________________
Spamassassin-talk mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
