Bug: zarr arrays cannot be pickled

[ieivanov]

zarr arrays created with tifffile cannot be pickled. This prevents using zarr arrays in multiprocessing applications.

Running:

zarr_store = tifffile.imread('path_to_data', aszarr=True)
zarrArray = zarr.open(zarr_store, mode='r')
pickle.dumps(zarrArray)

throws the error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-4-343d376ee953> in <module>
----> 1 pickle.dumps(zarrArray)

AttributeError: Can't pickle local object 'TIFF.CHUNKMODE.<locals>.CHUNKMODE'

I believe this is caused by this part of the code:

tifffile/tifffile/tifffile.py

Lines 12454 to 12461 in d2fe871

	def CHUNKMODE():
	class CHUNKMODE(enum.IntEnum):
	NONE = 0
	PLANE = 1
	PAGE = 2
	FILE = 3
	return CHUNKMODE

and can be solved using functools.wraps as described here: http://gael-varoquaux.info/programming/decoration-in-python-done-right-decorating-and-pickling.html

[cgohlke]

This is a known limitation (see discussion at AllenCellModeling/aicsimageio#178 (comment)) but I don't think it's a bug since zarr does not strictly require stores to be picklable (https://github.com/zarr-developers/zarr-python/blob/master/docs/tutorial.rst#pickle-support). I think aicsimageio works around this limitation by opening a new ZarrTiffStore in each process.
Fundamentally, tifffile.ZarrTiffStore is not picklable because it uses a TiffFile instance, which uses system handles (file handles and threads). To make tifffile.ZarrTiffStore picklable would need a different approach outlined in AllenCellModeling/aicsimageio#178 (comment). However I expect that to be significantly slower than the current implementation, both in initialization and chunk access.

[joshmoore]

@ieivanov, the zarr.convenience.copy_store helper method may do what you want. It won't produce pickled data, but it should convert the data from TIFF to Zarr on disk for later re-use.