np.unique raises a deprecation warning when all entries are nan.

[thomasjpfan]

np.unique raises a deprecation warning when all entries are nan.

Reproducing code example:

import numpy as np
import warnings

warnings.filterwarnings("error", category=DeprecationWarning)

a = np.array([np.nan] * 10)
result = np.unique(a)
print(result)

Error message:

ValueError: operands could not be broadcast together with shapes (0,) (9,) 

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/Users/thomasfan/Repos/numpy/wow.py", line 7, in <module>
    result = np.unique(a)
  File "<__array_function__ internals>", line 5, in unique
  File "/Users/thomasfan/Repos/numpy/numpy/lib/arraysetops.py", line 272, in unique
    ret = _unique1d(ar, return_index, return_inverse, return_counts)
  File "/Users/thomasfan/Repos/numpy/numpy/lib/arraysetops.py", line 343, in _unique1d
    mask[1:aux_firstnan] = (aux[1:aux_firstnan] != aux[:aux_firstnan - 1])
DeprecationWarning: elementwise comparison failed; this will raise an error in the future.

NumPy/Python version information:

1.21.0.dev0+1989.g80b387c66 3.9.5 | packaged by conda-forge | (default, Jun 19 2021, 00:27:35) 
[Clang 11.1.0 ]