Yes that would he great! For me a diff function would be useful to calculate the DNL which requires the previous sample regardless of timestep. I can also imagine some correlation calculations where t is diff'ed with t-m and if no sample exist graphs no datapoint.