BUG: distribute_jobs no longer works on Win/MacOS

[carterbox]

    def dump(obj, file, protocol=None):
        '''Replacement for pickle.dump() using ForkingPickler.'''
>       ForkingPickler(file, protocol).dump(obj)
E       TypeError: cannot pickle 'memoryview' object

This error is occuring on the conda-forge feedstock OSX builds. Research tells me that it is caused by trying to pass memory views instead of numpy arrays to the multiprocessing distrbute_jobs function.

[carterbox]

A recent patch to the numpy library (1.22.4), changed the behavior of np.from_buffer(), but only on non-linux platforms? On these platforms, the "base" attribute of a shared array is now memory view instead of the same as thee array before from_buffer is applied.

The affected code is here:

tomopy/source/tomopy/util/dtype.py

Lines 149 to 153 in 51b58c8

	shared_obj = mp.RawArray(ctype, arr.size)
	# create numpy array from shared object
	# shared_arr = np.ctypeslib.as_array(shared_obj)
	shared_arr = np.frombuffer(shared_obj, dtype=arr.dtype)
	shared_arr = np.reshape(shared_arr, arr.shape)

where shared_arr.base is now a memoryview instead of a mp.RawArray.

[carterbox]

@michael-sutherland, do you have any opinions about whether we should refactor the TomoPy implementation of distribute_jobs or pursue the patch in NumPy? I don't really understand the reasons/motivations behind either case.

[carterbox]

The API change which causes this issue was patched upstream.