CopyNode or similar

[arcondello]

When reshaping, it's not always possible to return a view. In that case, NumPy makes a copy of the array to be reshaped, and then reshapes the copy.

In [1]: arr = np.arange(40).reshape(10, 4)

In [2]: arr
Out[2]: 
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11],
       [12, 13, 14, 15],
       [16, 17, 18, 19],
       [20, 21, 22, 23],
       [24, 25, 26, 27],
       [28, 29, 30, 31],
       [32, 33, 34, 35],
       [36, 37, 38, 39]])

In [3]: arr.base
Out[3]: 
array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
       17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
       34, 35, 36, 37, 38, 39])

In [4]: arr[::3, ::2]
Out[4]: 
array([[ 0,  2],
       [12, 14],
       [24, 26],
       [36, 38]])

In [5]: arr[::3, ::2].base
Out[5]: 
array([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16,
       17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33,
       34, 35, 36, 37, 38, 39])

In [6]: arr[::3, ::2].reshape(-1)
Out[6]: array([ 0,  2, 12, 14, 24, 26, 36, 38])

In [7]: arr[::3, ::2].reshape(-1).base
Out[7]: 
array([[ 0,  2],
       [12, 14],
       [24, 26],
       [36, 38]])

In [8]: arr[::3, ::2].reshape(-1).base.base

The CopyNode would take a non-contiguous node and make a contiguous copy.