What happens if you pass an array that is fully masked?
I think an easier fix may be to remove the check for Sized-ness in errorbar (which is broken in this case because np.ma.masked inherits from np.ndarray (...) and thus appears as Sized (the ABC can't tell an exception is about to be thrown at C-level). Instead, use np.size(), which is much more lenient with its input (np.size(fe) at the call site).
(Strictly speaking, np.size casts its input to an array first (and then returns the product of dimensions) which may be inefficient if your input is a big non-array iterable, but on the other hand if that's the case you're about to error out -- so don't pass invalid input :-))

@anntzer Thanks for the comment - I've updated the PR based on your suggestions.