Irrlicht Engine

Seems like commenting just the ray.end manipulation in CSceneCollisionManager fixes the regression tests. I'm still not convinced that it really fixes the problem.
How to handle this bug is up to you. As it is not yet clear what the actual cause is, it's also not clear how it influences your apps. So just check your app and see if there's an error.
BTW: Using only a few nodes for collision, and giving all others a 0 as ID, should also reduce the number of errors in these routines.

Reflecting to the original question: I've bumped into a similar problem when making FC. I had to check if the result matches the original ray direction. If it did not, it must be "behind", so the result is ignored, and then another should be found. Can this idea fix the problem?

@Others: You could try out FunCollision, it has ray-casting too (for static objects).

This problem is fixed in Irrlicht 1.7. The problem was, that inside box tests used a different distance to the intersection point. And this could lead to trouble when having huge bboxes (as it's the case for lights). Now we're calculating the distance always the same, should be correct for all cases now.

After upgrading irrlicht engine from version 1.5.2 to 1.7.1 still mouse node picking doesn't work as well as in 1.5.2. In my orbit camera mode nodes couldn't be picked in some views angles or some distances.

I've also noticed that in some angles of view added nodes aren't rendered.

So 1.7.1 in this case is a little better than in 1.6 but not as good as 1.5.x

We would need reproducing examples for such behavior. All test cases now run through correctly.

This issue is fixed in 1.7