Unexpected result of .into_shape for f-order array

[raskr]

I applied reversed_axes to get f-order array and then called into_shape:

let a1 = ndarray::arr3(&[
    [
        [0., 1.],
        [2., 3.],
    ],
    [
        [4., 5.],
        [6., 7.],
    ]
]);
println!("a1:\n{:?}", a1);
let a2 = a1.view().reversed_axes();
println!("a2:\n{:?}", a2);
let a3 = a2.into_shape(8);
println!("a3:\n{:?}", a3);

I expected a3 to be [0.0, 4.0, 2.0, 6.0, 1.0, 5.0, 3.0, 7.0], but the result was:

a1:
[[[0.0, 1.0],
  [2.0, 3.0]],
 [[4.0, 5.0],
  [6.0, 7.0]]] shape=[2, 2, 2], strides=[4, 2, 1], layout=C (0x1), const ndim=3
a2:
[[[0.0, 4.0],
  [2.0, 6.0]],
 [[1.0, 5.0],
  [3.0, 7.0]]] shape=[2, 2, 2], strides=[1, 2, 4], layout=F (0x2), const ndim=3
a3:
Ok([0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0] shape=[8], strides=[1], layout=C | F (0x3), const ndim=1)

Is this an expected behavior?

[jturner314]

The reason why this is happening is that .into_shape() uses the memory order of the elements, which is not necessarily standard order. This behavior is quite bad because it's not always easy to determine what memory order an array will have. The behavior of .into_shape() needs to be changed. Please feel free to add comments to #390.

Closing as a duplicate of #390.