Support for memoryview-safe variable length arrays?

[shz9]

Hi there,

I'm using Zarr to store ragged arrays in a fashion that's similar to what's outlined in the documentation:

import numcodecs, zarr, numpy as np
z = zarr.empty(4, dtype=object, object_codec=numcodecs.VLenArray(int))
z[0] = np.array([1, 3, 5])
z[1] = np.array([4])
z[2] = np.array([7, 9, 14])

In my case, I need to retrieve those arrays and then process them in a cython function:

cpdef process_ragged_arrays(int[:] r_arr):
    ...

for i in range(z.shape[0]):
    process_ragged_arrays(z[i])

However, here I get the following error message: ValueError: buffer source array is read-only. This error has already been discussed and tackled elsewhere (e.g. Dask#1978, scikit-allel#208), typically by running the array through a function like this (h/t @alimanfoo):

def memoryview_safe(x):
    """Make array safe to run in a Cython memoryview-based kernel. These
    kernels typically break down with the error ``ValueError: buffer source
    array is read-only`` when running in dask distributed.
    See Also
    --------
    https://github.com/dask/distributed/issues/1978
    https://github.com/cggh/scikit-allel/issues/206
    """
    if not x.flags.writeable:
        if not x.flags.owndata:
            x = x.copy(order='A')
        x.setflags(write=True)
    return x

My question is: Is it possible to make ragged arrays memoryview-safe natively? I can definitely run memoryview_safe on each array I retrieve, but it will incur an overhead that I would like to avoid in my program.