"Loren Wilton" writes:
>> let me know what you all think.
>
>I had a little trouble following this message.  Part of the problem is I'm
>not positive I know what you mean by "compiled rules".  (Or I'm positive I
>don't know, your choice.  :-)

OK, so here's what I've done so far.

Rules files are now "compiled".  This allows us to

- put possibly-broken rules-in-development in the "rulesrc" source
  directory, without causing damage to the main ruleset if they are broken

- split rules across multiple directories, to allow the "sandbox" idea

- keep rule source in the same place, even as the rule is promoted,
  demoted, etc.

In actual fact, "compilation" is really just a script that copies the
lines from the source file into a destination file in the "rules" output
directory, so it's pretty simple.

>I also have a question on the "code tied" rules.  Which part of these rules
>are these?  You mention evals.  I would assume that the actual eval code for
>release N.M is still in the SA code base, so these would just be the the
>rules that call the eval functions?

Yep, that's the basic issue.  

http://wiki.apache.org/spamassassin/RulesProjPromotion lists the rulesets
that were split, and to where -- pretty much all the current set of eval
rules are "code-tied" and have to stay with the code.

>It seems to me there is the interesting point that some of these rules might
>be common across a few releases and others might be specific to a release.
>I'm not quite sure how you handle that case.  Maybe rules/evals/3.0,
>rules/evals/3.10, etc?

we can't really do that, since they might change *between* releases as
well -- and if we went down that route, we'd have so many
rules/evals/3.0_20050918 etc. directories that we'd eventually wind up
reinventing SVN ;)   The easier option is to keep those "code-tied"
rules as part of the source distribution.

>It definitely seems to me that you don't want the test rules to be landing
>in the same directory with existing rules.  But I didn't follow quite well
>enough to suggest something, other than I'd rather use more directories and
>keep the CF suffix.

I actually don't mind the test rules landing in the same dir as the
compiled existing rules, but just not in the same dir as the *source*
for the existing rules.  Thanks for the vote on the "use more directories"
idea though ;)

--j.

Reply via email to