Regarding extraction from &PyAny (or PyObject) to Rust object

[minsungkim-qraft]

Hello. I wanted to extract the Rust object from Python function call.

Impl

/// 'source is added because of below paragraph
impl<'source, I, O> SomeHiddenTrait<I, O> for SomeHiddenClass
where
    I: ToPyObject,
    O: PyClass + Clone,
{
    fn update(&mut self, data: &I) -> anyhow::Result<O> {
        Python::with_gil(|py| {
            let args = PyTuple::new(py, &[data]);
            let called = self.func.as_ref(py).call1(args)?;
            let extracted = called.extract::<O>()?;
            Ok(extracted)
        })
    }
}

If I change O: PyClass + Clone into O: FromPyObject<'source> + Clone (this was my first trial), then this does not work anymore. Rust compiler keeps complaining that the borrowed data must not move out of the closure, which confused me because O satisfies Clone anyway, so I imagined the cloned value is moving out of closure instead of borrowed value.

What is the reason why O: FromPyObject<'source> + Clone failed, and O: PyClass + Clone works?

[davidhewitt]

The problem is that in the example you give above, 'source is a lifetime limited to inside the with_gil closure.

To avoid this, either pass py: Python<'source> as an argument, or use O: for<'a> FromPyObject<'a> + Clone as the bound, which will limit the function to types which don't capture the FromPyObject lifetime in their output.

[minsungkim-qraft]

The first option, py: Python<'source> made an error - compiler complains that 'source must outlive 'static.

The second option works good, I didn't know that generic for prefix was available on trait constraints. Guess I have to find some source from internet.

Thank you.