Quote from @jbrockmendel

This matches the numpy behavior:

arr = np.array([]).reshape(0, 0)
nd3 = np.random.randint(5, size=(2, 2, 2))
arr[nd3] = 0

This may be a numpy bug as well cc @seberg ?

The above should validate the integer index and raise an error due to out of bound values. I am aware that there are issues with this when there are multiple array-indices which have an empty broadcast shape. But, I am not (at least anymore) aware of it when the slicing "subspace" is empty as opposed to the "advanced indexing" part being empty, and we should probably fix that.

(had to put the subspace word, it used to be a running gag with Jaime :))