unittest types not being pulled from typeshed

[jakebailey]

Simplified case:

import unittest

class Simple(unittest.TestCase):
    def test_success(self):
        self.assertEqual(42, 42)

    def test_exception(self):
        with self.assertRaises(Exception) as em:
            raise Exception

These have typeshed types:

def assertEqual(self, first: Any, second: Any, msg: Any = ...) -> None: ...

@overload
    def assertRaises(self,
                     exception: Union[Type[_E], Tuple[Type[_E], ...]],
                     msg: Any = ...) -> _AssertRaisesContext[_E]: ...

But in VSC you see:

The parameter types aren't shown for assertEqual, and there are no type hints for em in test_exception (not shown in the screenshot).

[MikhailArkhipov]

In Python 2 we don't use typeshed since basic analysis gives better results as source is in Python. In 3. we do pull typeshed, see #277, but all we get is exception since

class _AssertRaisesContext(Generic[_E]):
    exception = ...  # type: _E
    def __enter__(self) -> _AssertRaisesContext[_E]: ...
    def __exit__(self, exc_type: Optional[Type[BaseException]], exc_val: Optional[BaseException],
                 exc_tb: Optional[TracebackType]) -> bool: ...

[jakebailey]

Shouldn't there be more info than just "bound built-in method" regardless? I did my testing in 3.6 (also checking tensorflow analysis), and if typeshed info is really being used then I would have thought I'd see more info on both of these functions (and others).

Expanding the exception example out to:

    def test_exception(self):
        with self.assertRaises(Exception) as em:
            raise Exception
        exc = em.exception

em.exception does't show any types, so I don't think we're getting exception at all.

[MikhailArkhipov]

There are probably issues with TypeVar declarations

_T = TypeVar('_T')
_FT = TypeVar('_FT', bound=Callable[..., Any])
_E = TypeVar('_E', bound=Exception)

ex #254

[MikhailArkhipov]

Same as #254

[qubitron]

I tried Jake's example above and the completions returned are very limited:

Compared to the following with Jedi:

Is Jedi over-including?

[MikhailArkhipov]

No. Is it Python 2 or 3 and what LS version is it.

[MikhailArkhipov]

2:

3:

[qubitron]

False alarm, I was on old version of extension, this works for me now.