Re: Why is NaN used as extrapolation value?

[John W. Eaton]

On 18-Jun-2007, Søren Hauberg wrote:

| Søren Hauberg skrev:
| > Anyway, I'll cook up a patch tonight unless somebody 
| > gives me reason not to. Then we'll see what happens...
| Okay, so I'm attaching a simple patch that replaces NaN with NA in the 
| interpolation functions. I can't see how this could cause 
| difficulties/incompatibilities, but as David points out, people do weird 
| stuff. Apply or discard the patch as you will. The patch does fix one 
| bug in 'interpn' where NaN's are being located by 'vi == NaN' (this 
| should be 'isnan(vi)').

I don't have a strong opinion about this eitehr way, so I applied the
patch.  Since NA is a special NaN value, I don't see how it could
cause trouble, other than Matlab users who expect to see "NaN" might
be confused when Octave prints "NA".

jwe