Extract cubewrite fill value modifications

[blimlim]

This pr closes #99.

It extract's the cubewrite fill value modifications into a seperate function, adds a custom_fill_val optional argument, and adds unit tests.

Any suggestions or ideas are welcome!

[blimlim]

There's a slight amount of messiness with the added custom_fill_val argument, and so any ideas or suggestions would be great! Specifically, the final line of fix_fill_value forces the fill value to have the same type as the cube.data, which is required by the netCDF conventions

To prevent e.g. a float custom_fill_val from silently being rounded to an integer, I thought it would be good to raise an error when the types don't match:

um2nc-standalone/umpost/um2netcdf.py

Lines 334 to 340 in f3fbd44

	if custom_fill_val is not None:
	if type(custom_fill_val) == cube.data.dtype:
	fill_value = custom_fill_val
	else:
	msg = (f"custom_fill_val type {type(custom_fill_val)} does not "
	f"match cube {cube.name()} data type {cube.data.dtype}.")
	raise TypeError(msg)

This will also complain e.g. if a float was going to be rounded down to a float32, however this is stricter than what's applied to the default fill value for floats, which is would get rounded down to a float32 if the the cube's data is float32:

um2nc-standalone/umpost/um2netcdf.py

Lines 342 to 350 in f3fbd44

	elif cube.data.dtype.kind == 'f':
	fill_value = DEFAULT_FILL_VAL_FLOAT
	else:
	fill_value = default_fillvals[
	f"{cube.data.dtype.kind:s}{cube.data.dtype.itemsize:1d}"
	]
	# Use an array to force the type to match the data type
	cube.attributes['missing_value'] = np.array([fill_value], cube.data.dtype)

I don't think it's too important, but mainly wanted to check whether we're happy for the type requirements for the custom_fill_val to be stricter than what's applied to the default values?

[blimlim]

Ah oops, didn't notice the additional places the fill_value is required. Will fix this now

[truth-quark]

It might possible to drop attributes={} when this is de-conflicted with the newer code in main.

[blimlim]

Good catch. Fixed in the new PR #147

[truth-quark]

Can the type check be skipped now?

[blimlim]

Will keep these ones in, as it would be good to check that the get_default_fill_value() function correctly converts the default fill values to match the cube types.

[blimlim]

Oh no... I broke the branch. I've set up a clean one in a new PR #147