Warn about NaNs when interpolating?

[johnomotani]

When using DataArray.interp() to interpolate a 3d array along one dimension, I was surprised to end up with an output full of NaN. There were NaNs in a few elements in one corner of the original array, but the result has NaN even at points where the interpolation does not depend on the NaN in the input. I think the interpolation is done by scipy.interpolate.interp1d, whose documentation says:

Calling interp1d with NaNs present in input values results in undefined behaviour.

Probably we can't do anything to actually fix this, but I think it'd be nice to at least raise a warning or an error if the input has any NaNs, so that the user can do something about it (possibly suggest using DataArray/Dataset.fillna?).

[TomNicholas]

That definitely seems like something we should warn about! Perhaps even raise an error. Could you raise an issue with a minimal example? Also do you know what happens if you feed scipy.interpolate.interpn any NaNs?