What makes the difference between griddata and griddatan?

Since griddatan is the n-d generalization of griddata, I assume that for 2-d scenarios, they should have identical behavior. This is true for most cases. For example:

uv = [-6,-13;-3,-9;-9,-2;-12,7;25,-14;16,-24;3,15];
xy = [8,6;15,1;4,-11;-3,9];
z1 = griddata(uv(:,1),uv(:,2),uv(:,2),xy(:,1),xy(:,2))
z2 = griddatan(uv,uv(:,2),xy)

The output is:

z1 =
  6.0000
     NaN
-11.0000
  9.0000
z2 =
  6.0000
     NaN
-11.0000
  9.0000

But in the following example, griddata gives valid output while griddatan fails.

uv = [0,0;0,2;2,2;2,0];
xy = [1,1;3,1];
z1 = griddata(uv(:,1),uv(:,2),uv(:,2),xy(:,1),xy(:,2))
z2 = griddatan(uv,uv(:,2),xy)

The error message indicates that some internal check of the convex hull in griddatan fails. But why is that? Why griddata run without a similar convex check? Thanks.

[Honglei Chen]

I'd suggest you report this to MathWorks. Regardless of the reason, you certainly should get a better error message.