Breaking up is hard to do: an investigation of decomposition for assume-guarantee reasoning