Fix test_incompatible_dimensions test

[jamesp]

🚀 Pull Request

Description

test_incompatible_dimensions used a ragged array for the test, which has been deprecated in numpy, and now fails if dtype is anything other than object. This test appears to be checking that the addition of a [2x4] masked array to a [2x3] masked cube, iris should raise a ValueError. This commit fixes the creation of data3 object to be a [2x4] non-ragged array, preventing the different ValueError raised by numpy!

---

Consult Iris pull request check list

[bjlittle]

@jamesp Ohhh your first PR! Thank you for contributing! 🎉 🍻

i.e., we're hitting https://numpy.org/devdocs/release/1.19.0-notes.html#deprecate-automatic-dtype-object-for-ragged-input for numpy 1.2.0 ?

Going from this:

>>> np.__version__
'1.19.5'

>>> ma.MaskedArray([[3, 3, 3, 4], [2, 2, 2]], mask=[[0, 1, 0, 0], [0, 1, 1]])
<snip>/lib/python3.8/site-packages/numpy/ma/core.py:2831: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray
  _data = np.array(data, dtype=dtype, copy=copy,
masked_array(data=[--, --],
             mask=[ True,  True],
       fill_value='?',
            dtype=object)

to this:

>>> np.__version__
'1.20.0'
>>> import numpy.ma as ma
>>> ma.MaskedArray([[3, 3, 3, 4], [2, 2, 2]], mask=[[0, 1, 0, 0], [0, 1, 1]])
<snip>/lib/python3.9/site-packages/numpy/ma/core.py:2825: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray.
  _data = np.array(data, dtype=dtype, copy=copy,
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/tmp/persistent/itwl/miniconda3/envs/np-1p20/lib/python3.9/site-packages/numpy/ma/core.py", line 2888, in __new__
    mask = np.array(mask, copy=copy, dtype=mdtype)
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (2,) + inhomogeneous part.

[bjlittle]

@jamesp I think this very much deserves an entry in the whatsnew, simply because it's your very first contribution 😉

How's about putting a suitable entry in the " 💼 Internal" section of docs/iris/src/whatsnew/latest.rst as part of this PR? 👍

[bjlittle]

@jamesp Awesome!

Thanks for your first contribution to iris. It's very much appreciated!

🍻 💯 😀

[rcomer]

Hi @jamesp. Welcome! 😃