Decorators are sensitive to import cycle ordering

[ddfisher]

If we have two files:
test1.py:

from typing import Callable, Any
import test2

def deco(f: Callable[[int, int], Any]) -> Callable[[int, int], Any]:
    pass

@deco
def add(self, other: int) -> int:
    pass

test2.py:

import test1
test1.add(1, 1)

Running mypy test1.py test2.py results in:
test2.py:2: error: Cannot determine type of 'add'

Removing the @deco decorator from add or reversing the order of the files on the command line makes the error go away.

[JukkaL]

Yeah, this is a known issue.

[ddfisher]

I didn't see it documented elsewhere, though?

[JukkaL]

See #481 (though it's not very specific).

[JukkaL]

I marked this as a bug as there is no acceptable workaround right now, and this is a common thing.

[JukkaL]

The following was copied from #481 because it's relevant here:

Infer types for (most) decorated functions before the final type checking pass. Either do this during semantic analysis (we already do some, but it's pretty limited) or introduce another type checking pass. Decorated functions often cause trouble as mypy sometimes only infers their types during the final pass, when it can be too late as modules must be type checked in some specific, linear order.

A fairly simple approach would to improve the lightweight type inference engine we have just for function decorators.

Some ideas of additional things to support:

1. Decorator contains a call

@decorator()
def func(...): ...

We could just "unwrap" the decorator type by taking the return type without checking the call otherwise.

2. Decorator that always returns the same type (no type variables)

def decorator(f: Callable[[int], str]) -> Callable[[int], str]:
    ...

@decorator
def func(x: int) -> str:
    ...

We would just have to recognize if the return type has no type variables, which is easy, and then we can trivially infer the type of the decorated function.

The current implementation lives in semanal.py (ThirdPass.visit_decorator). I suggest implementing at least these improvements in 0.3.2, as they should be simple. What do you think?

[gvanrossum]

Let's make this a stretch goal for the 0.3.2. release. It's important, but we feel only Jukka understands it well enough to be able to fix it. If Jukka can't do it, we can release 0.3.2 without this and push it to the next milestone -- rather than hold up the release.

[JukkaL]

I started working on this a while back but haven't had time to finish it. I'll see if the current branch is good enough to merge already, even if it's not complete.

[JukkaL]

The original example now works, so I'm closing this. We can create new issues for remaining cases where decorators don't work.