Converting nb::ndarray to PyObject *

[ncullen93]

Hey there! I was wondering if anyone knows how to convert a nb::ndarray to a PyObject * so that it can be passed into the PyObject_GetBuffer function, for instance. I've tried calling .data() and casting to a PyObject * but i'm getting lots of segfaults.

In pybind11, you could simply call array.ptr() but that only seems possible in nanobind now for things like nb::tuple that have a nb::handle ?. I guess I would have to get the owner or somehow do the reverse of nb::handle...

I think this might be related to #519 but unsure

To be more specific, I am trying to take in a numpy array and create an ITK image from it. Something like this:

template <typename ImageType>
void fromNumpy( nb::ndarray<nb::numpy> data, nb::tuple datashape )
{
    typedef typename ImageType::Pointer ImagePointerType;
    ImagePointerType myimage = ImageType::New();
    typedef itk::PyBuffer<ImageType> PyBufferType;
   
   // data.ptr() doesnt exist anymore
  PyObject * arrData = .... 

    myimage = PyBufferType::_GetImageViewFromArray( arrData , datashape.ptr());
}

[wjakob]

nb::object o = nb::cast(data); and then use o.ptr()

[ncullen93]

Thanks I had JUST come upon nb::find and nb::cast from https://nanobind.readthedocs.io/en/latest/why.html#minor-additions and was going to try it. Much appreciated.