Clarify some behavior around user-defined generic classes

[rchen152]

• Protocol[T] behaves like Generic[T] when defining a generic class.
• It's an error to include both Generic[T] and Protocol[T].
• When Generic[T] is omitted, type variables are taken in order of first appearance in generic base classes.

Discussion: https://discuss.python.org/t/clarifying-the-rules-for-subclassing-generic-classes/69698.

[JelleZijlstra]

Should we mandate an error here instead? This code doesn't make sense, because a Child[int, str] is both a Parent[int, str] and a Parent[str, int] due to the double inheritance. So a type checker should error here, because it can't build up this class in a correct way.

[rchen152]

The reason I didn't mandate an error is that mypy currently doesn't error on this. But I'm happy to change it if we think that would be better.

[carljm]

I don't know if there is a rationale behind mypy's behavior, or how hard it would be to change, but it seems to me that this should be mandated as an error; I don't see any way to make sense of this definition.

This code doesn't make sense, because a Child[int, str] is both a Parent[int, str] and a Parent[str, int] due to the double inheritance.

To make sure I'm understanding: you meant Grandparent here? There is no double inheritance from Parent.

[rchen152]

Ok, I updated this section to mandate an error.