Just make them call dataset_differences and variable_differences. Obviously this has potential performance problems, but so be it. E.G. cube comparison in Iris does it anyway.