False positive error on return in generator

[JelleZijlstra]

If you use return in a generator, an incorrect TypeError is generated:

$ cat srcdir/genreturn.py 
from typing import Iterable

def gen(x: int) -> Iterable[int]:
    for i in range(x):
        if i > 10:
            return
        else:
            yield i
$ pyre --source-directory srcdir/ check
 ƛ Setting up a .pyre_configuration file may reduce overhead.
 ƛ Found 1 type error!
srcdir/genreturn.py:6:12 Incompatible return type [7]: Expected `Iterable[int]` but got `None`.
$ python --version
Python 3.7.0b4
$ pip freeze | grep pyre  # related minor bug: "pyre --version" throws a TypeError
pyre-check==0.0.5

[dkgi]

Good catch. I remember implementing something to handle returns inside generators at some point. I'll take a look at this.

[JelleZijlstra]

I actually looked into fixing this myself. There is indeed a check for (Type.is_none actual && Type.is_generator expected) in analysisTypeCheck.ml that suppresses the error I reported. However, this check simply looks for typing.(Async)Generator in the return type, which is wrong. Instead, we should check whether the function is a generator by looking for yields inside the function. I started writing an is_generator method in astStatement.ml, but I think we'll need some change in data structures to get access to this information in analysisTypeCheck.ml.

[dkgi]

Oh, in this case what you probably want something like a Define.is_generator function that looks at all the statements in the function body. The one complication I see here is that you probably want to run a visitor to find the yield statement in the body. We can't really use Visitor in astStatement because that would introduce a circular dependency. You might have to do it in annotatedDefine.ml instead.