Consider the following cases:

# A list of mixed cats and dogs is a list of animals.
class Animal: pass
class Cat(Animal): pass
class Dog(Animal): pass
pets = [Cat(), Dog()]  # note: Revealed type is 'builtins.list[Animal*]'

# A list of mixed ints and strings is a list of objects.
shallow = [1, "2"]  # note: Revealed type is 'builtins.list[builtins.object*]'

# A list of mixed lists-of-integers and lists-of-strings is also a list of objects!
nested = [[1], ["2"]]  # note: Revealed type is 'builtins.list[builtins.object*]'

Inferring the most precise type which is a parent of all the element types makes sense, but IMO the element type of nested should be inferred to be List[object] instead of object.

I'm using List here as a placeholder for any generic type - this is really more of an issue for things like Hypothesis strategies such as fixed_dictionaries(), with signature (Mapping[K, SearchStrategy[V]],) -> SearchStrategy[Mapping[K, V]].

This was reported downstream in HypothesisWorks/hypothesis#2349, because fixed_dictionaries({"num": integers(), "string": text()}) works but fails to type-check unless all values are strategies for the same type or the mapping is assigned to a type-annotated variable first.

Python 3.8 and Mypy 0.761, but it's not dependent on Python version.