Verify if there's a more intuitive or memory-efficient implementation for refinement types

[espetro]

I'm not sure if the current implementation for refinement types is intuitive, or if even it's memory efficient. Currently, a refinement predicate is a class that inherits from RefinementPredicate, and has a type_guard method:

class EmptyPredicate(Generic[_C], RefinementPredicate):

    @staticmethod
    def type_guard(value: _C, *args: Tuple[Any, ...], **kwargs: Dict[str, Any]) -> TypeGuard[_C]:
        return len(value) == 0

from typing import List, Annotated

EmptyIntList = Annotated[List[int], EmptyPredicate[List[int]]]

Alternatives I've thought about:

• RefinementPredicate classes should be instantiated with a self.value param. To improve memory management, they could use __slots__.