torch.Module typing __init__ no return value

[Vozf]

init return value should be specified as None. Now it doesn't have any
https://github.com/pytorch/pytorch/blob/master/torch/nn/modules/module.py#L223

https://www.python.org/dev/peps/pep-0484/#the-meaning-of-annotations

cc @albanD @mruberry @ezyang @malfet @rgommers @xuzhao9 @gramster

[mruberry]

Thanks for filing this issue, @Vozf. Your reference does make this seem straightforward. I'll let our typing experts weigh-in.

[ezyang]

Yes. We're not on strict type checking yet so mypy isn't reporting failures on errors like this, but we are happy to take patches that fix it.

[Vozf]

Mypy reports this error with disallow_untyped_calls = True in mypy.ini
This forces me to do type: ignore for any subclass of nn.Module

class MultiBCELoss(nn.Module):
    def __init__(self) -> None:
        super().__init__()  # type: ignore

Is something else expected from PR beyond adding -> None? If it's the only thing, I can create a PR

[ezyang]

Nope, that's good enough. Of course if you want to fix this entirely, or get our mypy to globally enforce disallow_untyped_calls, that would be even better, but those patches will be A LOT more work, so feel free to focus on just fixing cases that are affecting you personally.

[Vozf]

Also I got a question about typing in nn.functional.
Type hints exits in nn.functional.py but they also exist in stub nn.functional.pyi. And in pyi file they contain errors. Let's take interpolate
In nn.functional.py
https://github.com/pytorch/pytorch/blob/master/torch/nn/functional.py#L2975-L2976
Python2 style but correct in general.
In pyi
https://github.com/pytorch/pytorch/blob/master/torch/nn/functional.pyi.in#L294-L295
Types have any, and no recompute_scale_factor parameter at all.
That seems quite strange. It messes with mypy saying that there is no recompute_scale_factor parameter

[ezyang]

Our general goal is to eliminate all of the pyi files. So it would also be super helpful if you can merge the pyi comments in to the py using Python 3 style.