[Rspamd-Users] multimap or composites?

[Vu Ngoc VU]

Hello,

So, multimap works well for me. But as, you said, having actions in this module enables prefilter and bypass some learning (bayes, neural, etc).
I'm trying to keep actions (reject and accept) for exceptions only in multimap and have currently scores.

The next step is to play with composites. I have tried a few rules but that does not work as I expected after reading the documentation.

Example:
friendly_domain {
   expression = "(from_env_friendly_domain | from_header_friendly_domain) and (g:composite | g:url | g:rbl | g:headers)";
   score = -10;
   policy = "remove_weight";
}

When this composite matches, only one of my multimap on left part (from_env_friendly_domain or from_header_friendly_domain) gets its score removed.
By reading the documentation, I thought that all symbols in the expression should have their scores removed when the policy is remove_weight.
Maybe I didn't understand well the +/-/~ in groups related to groups on that page. But I've tried with g+ too.

My goal: for friendly domains (corporate, partners), when some checks are OK (like dkim/dmarc, etc) are verified, then remove postive scores of others checks like RBL or headers groups.

Is there an way to setup a composite (or something else) to satisfy this need?

Thanks,

-- 
Vu~