I'm trying to port the
float1,float2,float3,float4,double1,double2,double3,double4 classes from
Brook for GPU's at
http://cvs.sourceforge.net/viewcvs.py/*checkout*/brook/brook/include/brtvector.hpp?rev=1.43

I've run into a problem with using templates for type promotion:

struct XVector(qfloat){
qfloat x;qfloat y;qfloat z;
static XVector opCall (qfloat x, qfloat y, qfloat z) {
XVector ret;
ret.x=x;ret.y=y;ret.z=z;
return ret;
}

XVector opMul (XVector b) {
return  XVector(this.x*b.x,
this.y*b.y,
this.z*b.z);
}
}

I wish my b vector to be able to be cross-product'd on any template
instantiation of my template class.
in C++ I woudl write
template <class qfloat> class XVector {
template <class K> XVector<BetterType<K,qfloat>::type > operator %(XVector<K> b)
{
..
}
};
with
template <class A, class B> class BetterType{
typedef A type;
}
template <> class BetterType<float,double>{
typedef double type;
}


that way I could do the cross product between a XVector<float> and XVector<real> like happens with float,double C++ types.

PS: I can't get opCall to work with templates
XVector!(real)(1,2,3);
fails, but
XVector!(real).opCall(1,2,3);
succeeds