I'm simulating a box falling onto a conveyor from above, and its possible to reach a point where the produced boxes will just fall straight through the conveyor.
If the boxes are just added into a scene with a conveyor, every works fine, but if the conveyor belt is rotated, then boxes will start to fall through. This only happens if the box falls off center though (not sure why).
This would lead me to think that it's a bounding issue, however with physics bounding debugging turned on, its possible to see that the physics geometry is in the right place. Moreover, if there is a box on the conveyor when it is rotated, that box continues to collide with the correct physics geometry, whereas newly produced boxes will fall through the exact same spot the already moving box was interacting with.
Could this be a known ODE bug, or is there something I'm possibly missing?
Oh yea, also if a newly produced box falls onto the one that's already moving, then falls off and lands on the conveyor, it interacts normally, whereas it would have fallen through if it hadn't hit the other box first. Maybe there's something being set/updated in the first collision action that needs to be done on box init, rather than on its first collision? :?
Sorry, I didn't get it. Can you post a demo or video? Or a shorter explanation 
Oh, and make sure you are really using ODE not JOODE (it's logged with info level).
i put together a video demo thingy, but it looks like the hosting site is a bit slow, so its a little laggy
i tried to show how this only happens when the box lands offcenter of the conveyorbelt, and only when its rotated to down
it’s tough to get a box to hit another one before landing on the conveyorbelt, but i think i kinda showed at spots that if the falling one hits something before falling through the conveyor, it doesn’t fall through, but collides normally
anyhow, the rollers themselves just have surface motion applied, and that’s the only thing thats a physicsnode
this is such a strange bug…
http://tinypic.com/player.php?v=71ojzmo&s=1
That video is good. Now even I got it 
I noticed this bug in a game of mine as well! It seems the box-box collider in ODE has a bug with two boxes which have exactly the same rotation, one of them is small and at the edge of the other large one - I solved it by rotating my small boxes a tiny little bit. But I did not take the time to report it on the ODE mailing list or even provide an ODE-only (C++) testcase…
it's not related to surface motion, btw
aha! that's strange.
Thanks a lot!
