Are you saying there is a bug? The concept of CompoundMesh has always existed just handled differently now. Previously every sub mesh was converted directly to a Geometry object but now all sub meshes are stored in the new CompoundMesh. Then the collision checker handle this new type and processes all of the sub meshes. The only difference now is the CompoundMesh is captured at the highest level.

Here is the code which process the CompoundMesh.

I checked the code indeed. There is a bug, I tested adding some meshes separately (within a link) or as a CompoundMesh, and for Bullet this works for both, but for FCL the CompoundMesh version crashes. And with my HPP-FCL version (mostly similar to the FCL version) it is the same. I did not check in detail what went wrong, but CompoundMesh is the cause.

I thought this was covered by a unit test but I will look into it.

There is a unit test which does confirm this is working but may not be testing everything. Do you have a backtrace or could you add a unit test which triggers the failure?

I'll take a look tomorrow. I'm not really using FCL, so I just found this by accident.

@rjoomen Did you get a chance to look into this?

I just did. It was quite hard to find, but the problem is this:

In the COW constructor, a number of shapes and shape_poses is added. Before, each one shape was added as one COW, and thus had a matching shape_pose (and shapes_.size() == shape_poses_.size() == collision_objects_.size()). With a CompoundMesh, one shape is added as multiple COWs, and shape_poses_.size() != collision_objects_.size(). So when for example setCollisionObjectsTransform is executed for a CompoundMesh COW, it runs through, say, 5 objects, but only 1 object gets a valid shape_pose_.

I have submitted a PR to fix this, but please take a good look if this is the best solution.

The tests didn't pick this up as no setCollisionObjectsTransform is performed on a CompoundMesh.