Fix proc macro for init fn.

[sebpuetz]

The if statement evaluated to true if the return type was not "()" or "PyResult<()>" which was always the case since "()" != "PyResult<()>".

[sebpuetz]

test_datetime_check failed on an earlier build for windows on python 3.6 with an unwrap on Err which must have originated this method:

pyo3/tests/test_datetime.rs

Lines 8 to 35 in da1ab1b

	fn _get_subclasses<'p>(
	py: &'p Python,
	py_type: &str,
	args: &str,
	) -> PyResult<(&'p PyAny, &'p PyAny, &'p PyAny)> {
	// Import the class from Python and create some subclasses
	let datetime = py.import("datetime")?;
	let locals = [(py_type, datetime.get(py_type)?)].into_py_dict(*py);
	let make_subclass_py = format!("class Subklass({}):\n pass", py_type);
	let make_sub_subclass_py = "class SubSubklass(Subklass):\n pass";
	py.run(&make_subclass_py, None, Some(&locals))?;
	py.run(&make_sub_subclass_py, None, Some(&locals))?;
	// Construct an instance of the base class
	let obj = py.eval(&format!("{}({})", py_type, args), None, Some(&locals))?;
	// Construct an instance of the subclass
	let sub_obj = py.eval(&format!("Subklass({})", args), None, Some(&locals))?;
	// Construct an instance of the sub-subclass
	let sub_sub_obj = py.eval(&format!("SubSubklass({})", args), None, Some(&locals))?;
	Ok((obj, sub_obj, sub_sub_obj))
	}

The tests passed after pushing the same code again.

[kngwyu]

Thank you, but... is this feature actually used?
This is called for

impl Hoge {
    #[init] 
    fn __init__(...) -> {}
}

, but I can't find any document about this.
If this feature is not used, we should remove it instead of this fix.

@konstin
Do you know what's this?

[sebpuetz]

It does get called, although I'm not sure about the use of it:

#[pyclass]
pub struct Foo {
    bar: usize,
}

#[pymethods]
impl Foo {
    #[init]
    fn init(&mut self) -> () {
        println!("{}", self.bar);
        self.bar = 10;
        println!("{}", self.bar);
    }

    #[new]
    fn new(obj: &PyRawObject) -> PyResult<()> {
        obj.init(Self { bar: 0 });
        Ok(())
    }
}

>>> Foo()
0
10

[konstin]

Do you know what's this?

Unfortunately not

[programmerjake]

Do you know what's this?

Unfortunately not

I would guess #[init] is the equivalent of Python's __init__ method, whereas #[new] is the equivalent of __new__.

[Alexander-N]

Relevant: dgrunwald/rust-cpython#11 (comment)

[kngwyu]

Thank you for comments.
I opened #658 which removes #[init] feature.
Please feel free to write your opinion on the thread.

[kngwyu]

I'm sorry for closing this PR via #658.
Thank you for your contribution, though.