Why is NaN used as extrapolation value?

[Søren Hauberg]

Hi,
  The current interpolation functions all use NaN as extrapolation 
value. I know that Matlab does this, which (I guess) is why we do it. 
But since we have NA why isn't that used? Using NA as default 
extrapolation value should be Matlab compatible (right?) and would IMHO 
make more sense.

Søren

P.S. Since I brought it up I'll be happy to send patches if requested 
(they should be damn easy to produce...)