Fix normalize_fill_value for structured arrays

[alanhdu]

The FutureWarning identified in #552 is now a TypeError in the latest numpy release. We can fix this bu checking whether isinstance(fill_value, np.void) before checking whether fill_value == 0 (that way, if fill_value is already a structured dtype, then we do not compare it with a scalar).

One symptom of this is that

import numpy as np
import zarr

dtype = np.dtype([("a", np.float64), ("b", np.int64)])
zarr.open_array([], shape=1000, dtype=dtype

will crash with

TypeError                                 Traceback (most recent call last)
Cell In[4], line 1
----> 1 zarr.open_array({}, shape=1000, dtype=dtype)

File ~/workspace/zarr-python/zarr/creation.py:586, in open_array(store, mode, shape, chunks, dtype, compressor, fill_value, order, synchronizer, filters, cache_metadata, cache_attrs, path, object_codec, chunk_store, storage_options, partial_decompress, write_empty_chunks, zarr_version, dimension_separator, **kwargs)
    584         if contains_group(store, path=path):
    585             raise ContainsGroupError(path)
--> 586         init_array(store, shape=shape, chunks=chunks, dtype=dtype,
    587                    compressor=compressor, fill_value=fill_value,
    588                    order=order, filters=filters, path=path,
    589                    object_codec=object_codec, chunk_store=chunk_store,
    590                    dimension_separator=dimension_separator)
    592 elif mode in ['w-', 'x']:
    593     if contains_group(store, path=path):

File ~/workspace/zarr-python/zarr/storage.py:438, in init_array(store, shape, chunks, dtype, compressor, fill_value, order, overwrite, path, chunk_store, filters, object_codec, dimension_separator, storage_transformers)
    435 if not compressor:
    436     # compatibility with legacy tests using compressor=[]
    437     compressor = None
--> 438 _init_array_metadata(store, shape=shape, chunks=chunks, dtype=dtype,
    439                      compressor=compressor, fill_value=fill_value,
    440                      order=order, overwrite=overwrite, path=path,
    441                      chunk_store=chunk_store, filters=filters,
    442                      object_codec=object_codec,
    443                      dimension_separator=dimension_separator,
    444                      storage_transformers=storage_transformers)

File ~/workspace/zarr-python/zarr/storage.py:515, in _init_array_metadata(store, shape, chunks, dtype, compressor, fill_value, order, overwrite, path, chunk_store, filters, object_codec, dimension_separator, storage_transformers)
    513 chunks = normalize_chunks(chunks, shape, dtype.itemsize)
    514 order = normalize_order(order)
--> 515 fill_value = normalize_fill_value(fill_value, dtype)
    517 # optional array metadata
    518 if dimension_separator is None and store_version == 2:

File ~/workspace/zarr-python/zarr/util.py:298, in normalize_fill_value(fill_value, dtype)
    295 if fill_value is None or dtype.hasobject:
    296     # no fill value
    297     pass
--> 298 elif fill_value == 0:
    299     # this should be compatible across numpy versions for any array type, including
    300     # structured arrays
    301     fill_value = np.zeros((), dtype=dtype)[()]
    303 elif dtype.kind == 'U':
    304     # special case unicode because of encoding issues on Windows if passed through numpy
    305     # https://github.com/alimanfoo/zarr/pull/172#issuecomment-343782713

TypeError: Cannot compare structured or void to non-void arrays.

Closes #552.

[joshmoore]

Thanks for taking a look at this, @alanhdu! I've triggered the workflows.

[codecov]

Codecov Report

Merging #1397 (2ebbf19) into main (d54f25c) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##              main     #1397   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           37        37           
  Lines        14852     14857    +5     
=========================================
+ Hits         14852     14857    +5     

Impacted Files	Coverage Δ
zarr/tests/test_util.py	100.00% <100.00%> (ø)
zarr/util.py	100.00% <100.00%> (ø)

... and 2 files with indirect coverage changes

[joshmoore]

Thanks again, @alanhdu. I've added a changelog entry if you'd like to review. Workflows are running again with an updated main branch. Unless anyone has an objection, I'd say let's get this into 2.15.x.