Fix world to collider local in shape containsPoint

[jmiskovic]

I noticed that shape attached to collider does incorrect query on :containsPoint().

I think the order of translation/rotation was wrong, but since inverseTransformPoint is used twice it was easier to switch to lovrColliderGetLocalPoint. The added benefit is that it will work better with motion interpolation.

Minimal reproduction:

local world = lovr.physics.newWorld()

local c = world:newBoxCollider(1, 0, 0, 1, 4, 1)
c:setOrientation(math.pi/2, 0,0,1)

print(c:getShape():containsPoint(2, 0, 0)) -- should be true

I also checked Shape:raycast(), that one gives correct results.

[bjornbytes]

I think inverseTransformPoint was just broken, it should be this to match inverseTransformRay, right? (translate, then rotate)

diff --git src/modules/physics/physics.c src/modules/physics/physics.c
index b17b08a4..74a46368 100644
--- src/modules/physics/physics.c
+++ src/modules/physics/physics.c
@@ -2090,8 +2090,8 @@ void lovrShapeGetAABB(Shape* shape, float aabb[6]) {
 static void inverseTransformPoint(float* point, float* position, float* orientation) {
   float inverse[4];
   quat_conjugate(quat_init(inverse, orientation));
-  quat_rotate(inverse, point);
   vec3_sub(point, position);
+  quat_rotate(inverse, point);
 }
 
 static void inverseTransformRay(float* origin, float* direction, float* position, float* orientation) {

Also, it looks like all the getLocalPoint/WorldPoint/LocalVector/WorldVector don't work with interpolation, whereas getPose does.

[jmiskovic]

You're right. I've added to my test the shape offset with rotation and that broke down my fix.

Your version works as expected, aside from inertia wonkiness when shape is off center.

Interpolation is a separate issue I guess.

[bjornbytes]

Ok, I applied the diff from the comment. Thanks!