@implements(Protocol) decorator possible without Higher Kinded Types?

[randolf-scholz]

Sometimes I wish I there was a decorator @implements(<Protocol>) so that I could ensure a class satisfies some Protocol (or a function some Callback-Protocol) without having to subclass the Protocol explicitly, at the point of definition.

class MyProtocol(Protocol): ...

@implements(MyProtocol)
class MyImplementation: ...

Currently, the workaround I am using is writing static tests like _: type[MyProtocol] = MyImplementation, however there are multiple reasons to prefer having this check right at the definition of the implementation.

Question: Am I overlooking something, or is it currently impossible to implement @implements due to the lack of HKTs (#548)?

Here's what I think it should look like:

def implements[T](proto: type[T]) -> ???:
     # here, we want the identity function on the set of subtypes of T
     def decorator[C: type[T]](cls: C) -> C:
        return cls
    return decorator

I believe we need the dependent type C: type[T], because def decorator(cls: type[T]) -> type[T] would result in the type-checker inferring the type of MyImplementation being type[MyProtocol] afterwards.

[Daverball]

Indeed, this does appear impossible without HKTs or a dedicated mypy plugin, like mypy-zope which does provide support for zope.interface's Interface objects as types and its @implementer decorator.

But it seems to me like first-class support for a decorator like this seems preferable over a generic approach using HKTs, purely due to the quality of the error messages you're likely to get out of it. Like a non-generic version of this currently does:

Value of type variable "T" of "implements" cannot be "type[Bar]"  [type-var]

It doesn't tell you what attributes/methods are missing or incorrect, so not very helpful at all. Although to be fair, it also doesn't tell you in an x: type[Proto] = ProtoImpl assignment, you specifically need something like x: Proto = ProtoImpl() in order to get helpful errors. So this is definitely an area where mypy can and should improve its diagnostics.