`TypeError` when accessing `.real` and `.imag` attribute on sparse array of string type

[Anderson Banihirwe]

Describe the bug

attempting to access the .real attribute on a sparse.COO object with string data type results in a TypeError: string indices must be integers.

To Reproduce

In [18]: import numpy as np

In [19]: import sparse

In [20]: a = np.array(['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j'])

In [21]: a.real
Out[21]: array(['a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j'], dtype='<U1')

In [22]: data = sparse.COO(a)

In [23]: data
Out[23]: <COO: shape=(10,), dtype=<U1, nnz=10, fill_value=>

In [24]: data.real
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[24], line 1
----> 1 data.real

File ~/mambaforge/envs/playground/lib/python3.10/site-packages/sparse/_sparse_array.py:884, in SparseArray.real(self)
    859 @property
    860 def real(self):
    861     """The real part of the array.
    862 
    863     Examples
   (...)
    882     numpy.real : NumPy equivalent function.
    883     """
--> 884     return self.__array_ufunc__(np.real, "__call__", self)

File ~/mambaforge/envs/playground/lib/python3.10/site-packages/sparse/_sparse_array.py:305, in SparseArray.__array_ufunc__(self, ufunc, method, *inputs, **kwargs)
    302     inputs = tuple(reversed(inputs_transformed))
    304 if method == "__call__":
--> 305     result = elemwise(ufunc, *inputs, **kwargs)
    306 elif method == "reduce":
    307     result = SparseArray._reduce(ufunc, *inputs, **kwargs)

File ~/mambaforge/envs/playground/lib/python3.10/site-packages/sparse/_umath.py:49, in elemwise(func, *args, **kwargs)
     12 def elemwise(func, *args, **kwargs):
     13     """
     14     Apply a function to any number of arguments.
     15 
   (...)
     46     it is necessary to convert Numpy arrays to :obj:`COO` objects.
     47     """
---> 49     return _Elemwise(func, *args, **kwargs).get_result()

File ~/mambaforge/envs/playground/lib/python3.10/site-packages/sparse/_umath.py:466, in _Elemwise.__init__(self, func, *args, **kwargs)
    463 self._dense_result = False
    465 self._check_broadcast()
--> 466 self._get_fill_value()

File ~/mambaforge/envs/playground/lib/python3.10/site-packages/sparse/_umath.py:535, in _Elemwise._get_fill_value(self)
    525 """
    526 A function that finds and returns the fill-value.
    527 
   (...)
    531     If the fill-value is inconsistent.
    532 """
    533 from ._coo import COO
--> 535 zero_args = tuple(
    536     arg.fill_value[...] if isinstance(arg, COO) else arg for arg in self.args
    537 )
    539 # Some elemwise functions require a dtype argument, some abhorr it.
    540 try:

File ~/mambaforge/envs/playground/lib/python3.10/site-packages/sparse/_umath.py:536, in <genexpr>(.0)
    525 """
    526 A function that finds and returns the fill-value.
    527 
   (...)
    531     If the fill-value is inconsistent.
    532 """
    533 from ._coo import COO
    535 zero_args = tuple(
--> 536     arg.fill_value[...] if isinstance(arg, COO) else arg for arg in self.args
    537 )
    539 # Some elemwise functions require a dtype argument, some abhorr it.
    540 try:

TypeError: string indices must be integers

Expected behavior

given that NumPy arrays with string data type return the array itself when accessing .real, i was expecting the same behavior for sparse.COO objects.

System

• OS and version: macOS-14.0-arm64-arm-64bit
• sparse version (0.14.0)
• NumPy version (1.24.4)
• Numba version (numba.__version__)

Additional context
Add any other context about the problem here.