issue_comments: 391481382

If you implement arithmetic special methods like __add__ and __radd__ on your accessor, you could write things like: dt = DataArray(timestep.total_seconds(), attrs={'units': 's'}) for key in tendencies_list[0].keys(): return_state[key] = state[key].sympl + dt.sympl * ( 1.5 * tendencies_list[-1][key].sympl - 0.5 * tendencies_list[-2][key].sympl ) It's awkward but you can still use the infix notation.

Subclassing or writing a DataArray wrapper like what @kmpaul did in physarrays seems like the cleanest way to do this currently. The main challenge is to avoid relying on non-public parts of the DataArray API, but I think overwriting arithmetic would be pretty safe.

In the long term, I think the best solution here is probably an extensible interface for data in xarray objects that would let you put an array with units into an xarray object, similar to the hooks we want for sparse arrays (https://github.com/pydata/xarray/issues/1938).