Using iris cubes directly in unit tests

[blimlim]

We currently rely on the DummyCube class to imitate iris cube objects in our unit tests. We've tried to imitate the cube methods that are relevant to the tests, however the real iris methods can be much more complicated than our implementations in the DummyCube class, leading to situations where our imitations don't properly match and producing tests that would fail with real cubes.

Additionally, a couple ofiris methods used in process_cubes() are too hard to try and imitate (such as iris.util.reverse). We've ended up mocking a couple of the process_cube() steps as a result and leaving some functionality untested .

It turns out that it's possible to create iris cubes from scratch:
https://github.com/SciTools/iris/blob/d1125eb613e3c49c4128e004284e572ad337cad1/lib/iris/cube.py#L1098-L1103, though with the caveat:

Not typically used - normally cubes are obtained by loading data (e.g. :func:iris.load) or from manipulating existing cubes.

Creating small iris cubes instead of DummyCubes would allow us to test previously untested functions, remove mocks, and also avoid situations where the DummyCubes don't match the real iris cube behaviour, e.g:

> test_data = np.array([1,2,3])
> test_coord = iris.coords.DimCoord(points = np.array([0,0.1,0.2]),
                              standard_name="longitude")
> test_cube = cube.Cube(data = test_data,
                   standard_name="air_temperature",
                   dim_coords_and_dims = [(test_coord,0)]
                  )

> iris.util.reverse(test_cube, 'longitude').data
array([3, 2, 1])

One downside is that the Cube appears like it's generally meant to be created by loading in data, rather than being created from scratch as above. On the other hand, iris appears to use the above approach, creating small cubes from scratch, in their unit tests: https://github.com/SciTools/iris/blob/d1125eb613e3c49c4128e004284e572ad337cad1/lib/iris/tests/unit/cube/test_Cube.py#L88-L93