http://bugzilla.spamassassin.org/show_bug.cgi?id=3407
------- Additional Comments From [EMAIL PROTECTED] 2004-05-19 08:54 -------
Subject: Re: abandon rule processing when score gets high enough
> A point that Loren made was to run negative scores before positive scores,
and
> that made me think of another problem, how to deal with negative meta
tests
> that are really just a combination of positive scoring tests? I have such
meta
> to counter SUSPICIOUS_RECIPS && SORTED_RECIPS. Those are two positive
scoring
> tests of 2.63, so this would need to be considered.
>
> Then another idea, what about negative meta tests that are combined
positive
> meta tests? For example, if SUSPICIOUS_RECIPS && SORTED_RECIPS were both
meta
> rules themselves, would my negative scoring rule be taken into
consideration
> and how?
Both of these seem the same to me. As I pointed out in my original post
(not realizing it had to be posted to bugzilla to be seen), you have to do a
tree analysis at some point to determine the roots of any meta rule, and you
have to continue the walk until all nodes are non-meta. This tells you the
rules you must run to be able to run all of the negative scoring meta rules.
For that matter it tells you someting about overall rule ordering.
This also gives you a chance to detect circular meta rule chains. I am
presuming that detection is already done during lint if no other time.
Since that much of a tree analysis must already be being done, the full
treewalk shouldn't be that much more expensive.
I would hope that the tree analysis could be done at some reasonably static
time such as when rules are loaded so that it didn't have to be repeated for
each message processed. Others that know far more than I about the innards
of SA can determine if that is actually feasible.
Loren
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.
