Check class type parameter variance violations in inheritance

[JukkaL]

In generic inheritance, we can't change the variance of a type from invariant to covariant/contravariant. This should probably be rejected:

IV = TypeVar('IV')
CV = TypeVar('CV', covariant=True)

class A(Generic[IV]): ... # Bunch of methods
class B(A[CV], Generic[CV]): pass  # Error, type variable should be invariant

Going from covariant to invariant is fine (for example, List has to do this since it subclasses Sequence, which is covariant).

[jhance]

Seems like we break this in our own tests but fixing the lib-stub to make Sequence covariant on T should fix that.

[jhance]

I have this done on a branch, will submit PR later when my other stuff related to variance is in (since fixing this depends on fixing variance in general)

[jhance]

It just struck me that we should reject things like List[Sequence[T]] where T is covariant. This is going to be fun...

[JukkaL]

Yeah, List[Sequence[T]] would also be invalid.

I'm sure there's a research paper that goes through all the special cases. Some of these might be relevant, in case you want to dig deeper (warning greek alert):

http://research.microsoft.com/en-us/um/people/akenn/generics/