Question about Infs and Nans

This is a pure question, although it could also be a potential very minor improvement suggestion (at a very low priority). I notice that when my input data is a (numpy) array of Nans, then sometimes I would get Infs as distance values, which is easy to see and demonstrate in simple ways such as :

>>> dtw([2, 4], [np.nan, np.nan])
inf

Whereas sometimes I get Nans as the distance. I wonder if the Nans come from some numeric stability aspects or whether there is some other known edge case that is known to yield Nan from this function.

Obviously feeding Nans for a distance measure might be an abuse of a feature, as one might want to decide for themselves what are the semantics of comparing to Nan in their own particular case and it is probably not always a wise idea to let a package decide this without giving it thought as per one's mathematical ideas.

That said, it seems that all-Nan vectors are yield zero distance:

>>> dtw([[np.nan, np.nan]], [[np.nan, np.nan]])
0.0

And also have zero distance from an empty vector:

>>> dtw([[np.nan, np.nan], [np.nan, np.nan]] , [])
0.0

So just curious about the Nans as the output of the dtw function, and also asking this more as a way of getting slightly more acquainted with the library.

By the way, and not that the semantics need to be the same as the underlying numpy, when it is dealing with equality not distance:

>>> np.array_equal([np.nan], [np.nan])
False 
>>> np.array_equal([0], [np.nan])
False

Thanks to anyone who happen to care posting a quick comment!

Begin with the dtw function and reproduce the finite, NaN, and empty-vector examples from the issue. Trace how the distance values are produced; done should be a documented explanation of the differing outputs, or a clearly specified behavior and regression test if maintainers decide an improvement is warranted.