cftime plotting fails on upstream-dev

[dcherian]

seen in https://dev.azure.com/xarray/xarray/_build/results?buildId=3365&view=logs&jobId=2280efed-fda1-53bd-9213-1fa8ec9b4fa8&j=2280efed-fda1-53bd-9213-1fa8ec9b4fa8&t=175181ee-1928-5a6b-f537-168f7a8b7c2d

=========================== short test summary info ============================
FAILED xarray/tests/test_plot.py::TestCFDatetimePlot::test_cfdatetime_line_plot
FAILED xarray/tests/test_plot.py::TestCFDatetimePlot::test_cfdatetime_pcolormesh_plot
FAILED xarray/tests/test_plot.py::TestCFDatetimePlot::test_cfdatetime_contour_plot

e.g.

=================================== FAILURES ===================================
_________________ TestCFDatetimePlot.test_cfdatetime_line_plot _________________

self = <xarray.tests.test_plot.TestCFDatetimePlot object at 0x7f71d66219d0>

    def test_cfdatetime_line_plot(self):
E       ValueError: setting an array element with a sequence. The requested array would exceed the maximum number of dimension of 1.

/usr/share/miniconda/envs/xarray-tests/lib/python3.8/site-packages/matplotlib/transforms.py:943: ValueError

[spencerkclark]

Thanks for noting this @dcherian; it seems like there may have been a change in NumPy that caused this. E.g. with NumPy master this doesn't work:

>>> import numpy as np
>>> np.__version__
'1.20.0.dev0+2d12d0c'
>>> a = np.array([[1., 2.], [3., 4.]])
>>> a[:, 0] = (np.array([0.]), np.array([3650.]))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: setting an array element with a sequence. The requested array would exceed the maximum number of dimension of 1.

But with earlier versions of NumPy it does:

>>> import numpy as np
>>> np.__version__
'1.19.1'
>>> a = np.array([[1., 2.], [3., 4.]])
>>> a[:, 0] = (np.array([0.]), np.array([3650.]))
>>> a
array([[0.00e+00, 2.00e+00],
       [3.65e+03, 4.00e+00]])

I'll need to investigate more to see whether this is something to bring up in the matplotlib or NumPy issue tracker (it could be nc-time-axis related too). I'm not sure if there's anything we can do in xarray to fix this. It's probably best to xfail these tests for the time being.

[max-sixty]

xfailed with: #4271

[spencerkclark]

I think this NumPy issue is related: numpy/numpy#16939. numpy/numpy#16943 was merged yesterday, which appears to have at least allowed this to continue to work for the time being:

>>> import numpy as np
>>> a = np.array([[1., 2.], [3., 4.]])
>>> a[:, 0] = (np.array([0.]), np.array([3650.]))
<stdin>:1: DeprecationWarning: setting an array element with a sequence. This was supported in some cases where the elements are arrays with a single element. For example `np.array([1, np.array([2])], dtype=int)`. In the future this will raise the same ValueError as `np.array([1, [2]], dtype=int)`.

Given that this will be deprecated, I think it's still worthwhile to investigate what's causing this upstream in matplotlib or nc-time-axis, but it should be safe to un-xfail these tests again.

[spencerkclark]

I think I tracked down the root of this odd behavior in nc-time-axis; see SciTools/nc-time-axis#45 for more details.

[dcherian]

nice work spencer!