Wrap read_interpolated_variable's data size determination in a try block

[ursg]

This catches the special case in which a derived variable evaluates to
NaN (such as TPerpOverPar), gets masked away, creates an array with zero
effective size, and then all hell breaks loose.

[markusbattarbee]

This works for the read interpolated variable indeed. Perhaps a better fix would be to have read_variable return a NaN instead, if the return object was about to be a fully masked array?

[JonasSuni]

The problem in question is due to a masked element of a masked array having the data type np.ma.core.MaskedConstant, which is a constant, but for some reason inherits np.ma.core.MaskedArray, np.ndarray and Iterable, meaning that it passes the test for isinstance(test_val, Iterable), but cannot be iterated and its size is an empty tuple.

My suggestion is to change

if isinstance(test_val, Iterable):
    value_length=len(test_val)

to

if isinstance(test_val, np.ma.core.MaskedConstant):
    value_length=1
elif isinstance(test_val, Iterable):
    value_length=len(test_val)