There are cases where returning (and checking) for nan values would be useful.
Currently, you have to do something awkward like this to get a nan value:

vector v = "nan nan nan";
float v2 = v[0];