How to create an array with a dimension that is generic?

[AdityaGudimella]

Hello,
I'm trying to create a circular buffer that is backed by ndarray::Array. I want the circular buffer to be implemented for different dimensions and dtypes of arrays. A fixed dimension version looks like so:

pub struct CircularBuffer1D<S>
    where
        S: Clone + Zero,
{
    capacity: usize,
    head: usize,
    data: Array1<S>,
    total_num_samples_seen: usize,
}


impl<S> CircularBuffer1D<S>
    where
        S: Clone + Zero,
{
    pub fn new(capacity: usize) -> Self {
        CircularBuffer1D {
            capacity: capacity,
            head: 0,
            data: Array1::<S>::zeros(capacity),
            total_num_samples_seen: 0,
        }
    }
}

I'm now trying to convert it into a generic on the dimension too. This is what I have so far:

struct CircularBufferND<S, D>
    where
        S: Clone + Zero,
        D: Dimension + RemoveAxis
{
    capacity: usize,
    head: usize,
    data: Array<S, D>,
    total_num_samples_seen: usize,
}

impl<S, D> CircularBufferND<S, D>
    where
        S: Clone + Zero,
        D: Dimension + RemoveAxis
{
    fn new(capacity: usize, element_shape: Vec<usize>) -> Self {
        let mut final_shape_vec: Vec<Ix> = vec![capacity];
        final_shape_vec.extend(element_shape);
        CircularBufferND {
            capacity: capacity,
            head: 0,
            data: Array::<S, D>::zeros(final_shape_vec),
            total_num_samples_seen: 0,
        }
    }
}

The compiler throws an error on the line data:Array::<S, D>::zeros(final_shape_vec), saying:

type mismatch resolving `<std::vec::Vec<usize> as ndarray::ShapeBuilder>::Dim == D`

expected type parameter `D`, found struct `ndarray::Dim`

I tried building a Shape from the Vec<Ix> but wasn't successful either. Could you please help me?

[jturner314]

The issue is that the Array::zeros(shape) method requires the shape argument to be compatible with the dimension type, but Vec<usize> as a shape is compatible only with IxDyn.

A good alternative is to create an empty shape using D::zeros(ndim), and then assign the axis lengths to the shape. Here's an example:

impl<S, D> CircularBufferND<S, D>
    where
        S: Clone + Zero,
        D: Dimension + RemoveAxis
{
    fn new(capacity: usize, element_shape: Vec<usize>) -> Self {
        let ndim = 1 + element_shape.len();
        let mut final_shape: D = D::zeros(ndim);
        final_shape[0] = capacity;
        final_shape.as_array_view_mut().slice_mut(s![1..]).assign(&ArrayView::from(&element_shape));
        CircularBufferND {
            capacity: capacity,
            head: 0,
            data: Array::<S, D>::zeros(final_shape),
            total_num_samples_seen: 0,
        }
    }
}

Note that this will panic if ndim doesn't match D.

[AdityaGudimella]

That worked. Thank you very much! And thank you for all the hard work on the crate!