python
spec: Definition of materialization is insufficient
The spec currently defines materialization as follows:
Given a gradual type A, if we replace zero or more occurrences of Any in A with some type (which can be different for each occurrence of Any), the resulting gradual type B is a materialization of A.
A type A is assignable to B if there exists a pair of materializations A' and B' of A and B such that A' is a subtype of B'.
Now let's consider whether list[int] | list[str] is assignable to list[Any]. To do so, we have to consider materializations of list[Any], created by substituting Any with some other type. But there's nothing we can substitute for Any that would make for a fully static type that is a supertype of list[int] | list[str], since list is invariant in its type parameter. Therefore, list[int] | list[str] is not assignable to list[Any]. Relatedly, this means list[Any] | list[Any] is not equivalent to list[Any]. This is obviously not a desirable conclusion; current type checkers treat list[int] | list[str] as assignable to list[Any], and they should.
I think to fix this, we could say that if A' and A'' are materializations of A, then the union A' | A'' must also be a materialization of A.
I've been thinking about this more and what I said above isn't quite enough. We also need to allow unions in "intermediate" positions: list[list[Any]] should be able to materialize list[list[int] | list[str]]. We possibly also should allow intersections and differences of other materializations.