I was sugesting a single method since it could avoid loops going through all the classes, right ?
No, that's the whole point of having this scanning "framework" -- we do it all (all sorts of indexing, ...) in a single pass.
What we need is just to properly prepare the info, so when the query is invoked, it's basically O(1) lookup.
So, split methods make sense to me, as then all we do is Set::intersect.
Or am I missing something?