Return mutable array from rust to python

[Rikorose]

It would be nice to have a function like into_pyarray_mut to be able to do the following:

fn some_fn<'py>(py: Python<'py>) -> PyResult<&'py mut PyArray1<f32>> {
    let mut zeros: Array1<f32> = ndarray::Array1::zeros((10));
    Ok(zeros.into_pyarray_mut(py))
}

Is this possible in a safe way?

[kngwyu]

&'py PyArray is actually mutable in Python. We also provide PyReadonlyArray for immutable arrays

[Rikorose]

Thanks for your answer. Unfortunately, I still struggle to write to an array, which was allocated via rust-numpy. Maybe you can spot a mistake?
Here is my code:

  fn ret_some_array<'py>(&self, py: Python<'py>) -> PyResult<&'py PyArray1<f32>> {
        let v = vec![1f32; 10];
        Ok(v.into_pyarray(py))
    }

On the python side I get the following error:

a = ret_some_array()
print(a.flags)
# a.setflags(write=True)  # -> ValueError: cannot set WRITEABLE flag to True of this array
# a[0] = 0  # -> ValueError: assignment destination is read-only

flags:

a.flags:   C_CONTIGUOUS : True
               F_CONTIGUOUS : True
               OWNDATA : False
               WRITEABLE : False
               ALIGNED : True
               WRITEBACKIFCOPY : False
               UPDATEIFCOPY : False

[asmh1989]

@Rikorose

#[macro_export]
macro_rules! nparray_return {
    ($s:expr) => {{
        let dot = $s;
        unsafe {
            (*dot.as_array_ptr()).flags |= NPY_ARRAY_WRITEABLE;
        }
        dot
    }};
}

[kngwyu]

@Rikorose
Ah, I understand. The array created by into_pyarray is actually not writable since it can be returned to the Rust world eventually when destructing. I think we need to change the API a bit.

[kngwyu]

I keep this issue opening until I fix this inconsistent API issue.

[adamreichold]

The array created by into_pyarray is actually not writable since it can be returned to the Rust world eventually when destructing. I think we need to change the API a bit.

I don't really follow the reasoning here. The underlying SliceBox is dropped only when there are no more NumPy arrays referencing the data owned by it hence there cannot be any aliasing and we should be safe to set the WRITEABLE flag in Array::from_boxed_slice as we do have ownership at that point.