Use assert_type for type-stub checks intended for exact type match

[leonarduschen]

Checks in tests/typing intended to be exact matches should be done with assert_type instead of simply assigning to variables

Explanation

Consider the following functions

def f_any() -> Any:
    return ...

def f_int() -> int:
    return 1

Assigning to vars to check

# It might look like we're asserting the return type of function is Any
var1: Any = f_any()

# But all it does it asserting that return_any() is assignable to Any 
# which is always true for any functions

var2: Any = f_int() # This is ok with mypy

Using assert_type

assert_type(f_any(), Any)
assert_type(f_int(), Any)  # error: Expression is of type "int", not "Any"  [assert-type]

---

The usecase for simply assigning to variables is to check if a variable is a subclass (or "assignable" to be specific) of a specific type. For example.

var1: Animal = f_cat()  # We don't care if the stub declares f_cat to return Animal or Cat

But in the usecase of stub-testing, we probably almost always want to check exact type insteaad of assignability.

Examples

Example 1:

python-dependency-injector/tests/typing/selector.py

Line 11 in 5a1aef9

var1: Any = provider1()

This will pass regardless of the return type of provider1

Example 2:

python-dependency-injector/tests/typing/factory.py

Line 92 in 5a1aef9

animal10: Animal = factory10()

The example is supposed to check if explicitly typing factory10: Provider.Animal makes factory10() return Animal instead of Cat.

But it doesn't check that, since Cat is assignable to Animal, which defeats the purpose of the test

factory10 = providers.Factory(Cat)
animal10: Animal = factory10()  # Here will still pass even though it returns cat

This, on the other hand, will work as intended

factory10: providers.Provider[Animal] = providers.Factory(Cat)
assert_type(factory10(), Animal)

---

There are probably a lot more examples

[ZipFile]

assert_type is a Python 3.11 feature, those test were written before this function even existed. If you're willing to retrofit them, I'll gladly accept a PR.

[leonarduschen]

Right, I noticed the issue when working on the Selector typevar, I thought I'd open a new issue to discuss first to make sure I understand the problem correctly :)

I'll go ahead with the PR to retrofit first then, I think we'll benefit from a stronger stub test suite before making the Typevar variance change.