Decorated __init__ confuses callback type if decorator is defined with type alias or if decorator is imported using relative import

[posita]

This one's just weird. I'm not sure if this is a dup of #1927, but I figured I'd raise it anyway.

# test_case.py
from typing import Callable, Iterable, TypeVar, Union

_T = TypeVar("_T")

def _identity(__: _T) -> _T:
    return __

# decr: Callable[[_T], _T] = _identity  # <-- this approach works
DecoratorT = Callable[[_T], _T]
decr: DecoratorT = _identity  # <-- this causes the false positives below

class Result:
    @decr
    def __init__(self):
        pass

res = Result()  # ; reveal_type(res)  # -> Any (should be test_case.Result)

def some_results(op: Callable[[], Union[Result, Iterable[Result]]]) -> Iterable[Result]:
    result_or_results = op()
    results: Iterable[Result]
    if isinstance(result_or_results, Result):
        results = (result_or_results,)  # ; reveal_type(result_or_results)  # -> (should be test_case.Result)
    else:
        results = result_or_results  # ; reveal_type(result_or_results)  # -> Union[test_case.Result, typing.Iterable[test_case.Result]] (should be typing.Iterable[test_case.Result])
    return results

% python --version ; mypy --version
Python 3.9.7
mypy 0.910
% mypy --config-file=pyproject.toml test_case.py
test_case.py:26: error: Incompatible types in assignment (expression has type "Union[Result, Iterable[Result]]", variable has type "Iterable[Result]")
Found 1 error in 1 file (checked 1 source file)

Note that if DecoratorT is defined as follows, things work again:

# …
class DecoratorT(Protocol):
    def __call__(self, __: _T) -> _T:
        ...
decr: DecoratorT = _identity  # <-- all good
# …

But if we move the decorator into its own module and use a relative import, it breaks no matter how we define the decorator type:

# test_case.py
from typing import Callable, Iterable, TypeVar, Union
# from test_case_import import decr  # <-- this works
from .test_case_import import decr  # <-- this causes the false positives below 

class Result:
    @decr
    def __init__(self):
        pass

res = Result()  # ; reveal_type(res)  # -> Any (should be test_case.Result)

def some_results(op: Callable[[], Union[Result, Iterable[Result]]]) -> Iterable[Result]:
    result_or_results = op()
    results: Iterable[Result]
    if isinstance(result_or_results, Result):
        results = (result_or_results,)  # ; reveal_type(result_or_results)  # -> (should be test_case.Result)
    else:
        results = result_or_results  # ; reveal_type(result_or_results)  # -> Union[test_case.Result, typing.Iterable[test_case.Result]] (should be typing.Iterable[test_case.Result])
    return results

# test_case_import.py
from typing import Callable, TypeVar
_T = TypeVar("_T")

def _identity(__: _T) -> _T:
    return __

decr: Callable[[_T], _T] = _identity  # <-- doesn't matter how this is defined

% mypy --config-file=pyproject.toml test_case.py test_case_import.py
test_case.py:19: error: Incompatible types in assignment (expression has type "Union[Result, Iterable[Result]]", variable has type "Iterable[Result]")
Found 1 error in 1 file (checked 2 source files)

Using the Protocol approach does not salvage the second (relative import) scenario.

[posita]

What does note: Revealed type is "def [_T] (_T`-1) -> _T`-1" mean? (I found that when looking into the real world case that inspired the above distillations.)

[JelleZijlstra]

It's a function with the type signature (T) -> T where T is a typevar, i.e. an identity function.

[posita]

It's a function with the type signature (T) -> T where T is a typevar, i.e. an identity function.

Good to know it's legit. I've never seen that before. Is it documented anywhere? I was unable to find it in the official docs. (Apologies if I missed it.)