I did write a wonderfully detailed post about my problem, then accidentally closed the tab. So this'll be a little more brief, but hopefully still detailed enough to get some help!

Anyway, I'm trying to convert mouse co-ordinates to world space. At the moment, I'm passing in two normalized device co-ordinates, (x,y,-1.0f) and (x,y,1.0f), then transforming them by the inverse of (proj_matrix*view_matrix). I'm expecting to get two points - one on the near clipping plane and one on the far clipping plane, but I'm not. The near plane is 30, and the far plane is 5000, but I'm getting z values of 0.7 and 7 respectively.

I'm not doing any multiplication by any w value to get to clipping space - could that be the problem? If so, how should I get the w value to multiply all the elements by?

Here's the bits of my code that are relevant:

Ray newray(0.2f,-0.2f,-1.0f,0.2f,-0.2f,1.0f);
newray.SetMatrices(cam_->GetProjectionMatrix(),cam_->GetViewMatrix());
newray.Calculate();

class Ray
{
cml::matrix44f_c inv_mat_;
vector3f start_, end_;

vector3f transformed_start_, transformed_end_;
public:
Ray(float sx, float sy, float sz, float dx, float dy, float dz);
void SetRayEnds(float sx, float sy, float sz, float dx, float dy, float dz);

void SetMatrices(const cml::matrix44f_c & proj, const cml::matrix44f_c & view);
void Calculate();

vector3f GetYIntersection(float y);
};

Ray::Ray(float sx, float sy, float sz, float dx, float dy, float dz) :
inv_mat_(cml::identity_4x4()),
start_(sx,sy,sz),
end_(dx,dy,dz)
{

}
void Ray::SetRayEnds(float sx, float sy, float sz, float dx, float dy, float dz)
{
start_.set(sx,sy,sz);
end_.set(dx,dy,dz);
}

void Ray::SetMatrices(const cml::matrix44f_c & proj, const cml::matrix44f_c & view)
{
inv_mat_ = cml::inverse(proj*view);
}

void Ray::Calculate()
{
transformed_start_ = cml::transform_point(inv_mat_, start_);
transformed_end_ = cml::transform_point(inv_mat_, end_);
}

To all the matrix and graphics wizards, what am I doing wrong? Is this the way that you would approach the problem?

Thanks for your help!
Ben

EDIT: World space, not eye space.

Very sorry, I did mean world space - it was late and I was sleepy!

The only thing I can think it could be is the cml::transform_point function I'm using - I've assumed that this multiplies the input point by the matrix as gluUnProject does, but that might not be right. I'll try making a 4x1 matrix containing the values plus the 1 at the bottom, and use straightforward matrix multiplication.