xarray 2022.6.0 doesn't work well with numpy 1.20

[seisman]

What happened?

Fail to import xarray when xarray 2022.6.0 and numpy 1.20 are installed. It seems xarray 2022.6.0 is not compatible with numpy 1.20. I also tried numpy 1.19 and 1.21, they all work well with xarray 2022.6.0.

What did you expect to happen?

xarray 2022.6.0 should work well with numpy 1.20.

Minimal Complete Verifiable Example

$ conda create --name xarraytest xarray=2022.6.0 numpy=1.20
$ conda activate xarraytest
$ conda list
# Name                    Version                   Build  Channel
_libgcc_mutex             0.1                 conda_forge    conda-forge
_openmp_mutex             4.5                       2_gnu    conda-forge
bzip2                     1.0.8                h7f98852_4    conda-forge
ca-certificates           2022.6.15            ha878542_0    conda-forge
importlib-metadata        4.11.4           py39hf3d152e_0    conda-forge
importlib_metadata        4.11.4               hd8ed1ab_0    conda-forge
ld_impl_linux-64          2.36.1               hea4e1c9_2    conda-forge
libblas                   3.9.0           15_linux64_openblas    conda-forge
libcblas                  3.9.0           15_linux64_openblas    conda-forge
libffi                    3.4.2                h7f98852_5    conda-forge
libgcc-ng                 12.1.0              h8d9b700_16    conda-forge
libgfortran-ng            12.1.0              h69a702a_16    conda-forge
libgfortran5              12.1.0              hdcd56e2_16    conda-forge
libgomp                   12.1.0              h8d9b700_16    conda-forge
liblapack                 3.9.0           15_linux64_openblas    conda-forge
libnsl                    2.0.0                h7f98852_0    conda-forge
libopenblas               0.3.20          pthreads_h78a6416_0    conda-forge
libstdcxx-ng              12.1.0              ha89aaad_16    conda-forge
libuuid                   2.32.1            h7f98852_1000    conda-forge
libzlib                   1.2.12               h166bdaf_2    conda-forge
ncurses                   6.3                  h27087fc_1    conda-forge
numpy                     1.20.3           py39hd249d9e_2    conda-forge
openssl                   3.0.5                h166bdaf_0    conda-forge
packaging                 21.3               pyhd8ed1ab_0    conda-forge
pandas                    1.4.3            py39h1832856_0    conda-forge
pip                       22.2               pyhd8ed1ab_0    conda-forge
pyparsing                 3.0.9              pyhd8ed1ab_0    conda-forge
python                    3.9.13          h2660328_0_cpython    conda-forge
python-dateutil           2.8.2              pyhd8ed1ab_0    conda-forge
python_abi                3.9                      2_cp39    conda-forge
pytz                      2022.1             pyhd8ed1ab_0    conda-forge
readline                  8.1.2                h0f457ee_0    conda-forge
setuptools                63.2.0           py39hf3d152e_0    conda-forge
six                       1.16.0             pyh6c4a22f_0    conda-forge
sqlite                    3.39.2               h4ff8645_0    conda-forge
tk                        8.6.12               h27826a3_0    conda-forge
typing_extensions         4.3.0              pyha770c72_0    conda-forge
tzdata                    2022a                h191b570_0    conda-forge
wheel                     0.37.1             pyhd8ed1ab_0    conda-forge
xarray                    2022.6.0           pyhd8ed1ab_0    conda-forge
xz                        5.2.5                h516909a_1    conda-forge
zipp                      3.8.0              pyhd8ed1ab_0    conda-forge
zlib                      1.2.12               h166bdaf_2    conda-forge

$ python
Python 3.9.13 | packaged by conda-forge | (main, May 27 2022, 16:58:50) 
[GCC 10.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import xarray as xr
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/user/miniconda/envs/xarraytest/lib/python3.9/site-packages/xarray/__init__.py", line 1, in <module>
    from . import testing, tutorial
  File "/home/user/miniconda/envs/xarraytest/lib/python3.9/site-packages/xarray/testing.py", line 9, in <module>
    from xarray.core import duck_array_ops, formatting, utils
  File "/home/user/miniconda/envs/xarraytest/lib/python3.9/site-packages/xarray/core/duck_array_ops.py", line 26, in <module>
    from . import dask_array_compat, dask_array_ops, dtypes, npcompat, nputils
  File "/home/user/miniconda/envs/xarraytest/lib/python3.9/site-packages/xarray/core/npcompat.py", line 72, in <module>
    _SupportsDType[np.dtype],
  File "/home/user/miniconda/envs/xarraytest/lib/python3.9/typing.py", line 277, in inner
    return func(*args, **kwds)
  File "/home/user/miniconda/envs/xarraytest/lib/python3.9/typing.py", line 1004, in __class_getitem__
    _check_generic(cls, params, len(cls.__parameters__))
  File "/home/user/miniconda/envs/xarraytest/lib/python3.9/site-packages/typing_extensions.py", line 92, in _check_generic
    raise TypeError(f"{cls} is not a generic class")
TypeError: <class 'numpy.typing._dtype_like._SupportsDType'> is not a generic class
>>> 

MVCE confirmation

• Minimal example — the example is as focused as reasonably possible to demonstrate the underlying issue in xarray.
• Complete example — the example is self-contained, including all data and the text of any traceback.
• Verifiable example — the example copy & pastes into an IPython prompt or Binder notebook, returning the result.
• New issue — a search of GitHub Issues suggests this is not a duplicate.

Relevant log output

No response

Anything else we need to know?

No response

Environment

[headtr1ck]

Thanks for the report.
I have to admit, that when I implemented the problematic part I also only tested numpy 1.19 and >1.21... (same with CI)

It seems that in 1.20 they used a different approach.

[TomNicholas]

What changed in numpy such that only one version is incompatible with our new release? Seems quite odd to have something break for one version, then work again for the next version.

[headtr1ck]

It seems that in 1.20 they started introducing typing but _SupportsDType[np.dtype] was not a generic yet.

Not really the fault of numpy, we are using a not-public interface.

So either we do a version check for 1.20, or we create our own SupportsDType (I assume a Protocol?)

[TomNicholas]

Right okay, thanks for the explanation @headtr1ck .

So either we do a version check for 1.20, or we create our own SupportsDType (I assume a Protocol?)

If we are still using a non-public interface, then we can't guarantee that it won't break again in numpy version 1.22 etc. So if that's the case perhaps we should try to make a sustainable solution. Otherwise I think special-casing one version is okay.