Bug fix for btAxisSweep3

[ole.k]

btAxisSweep3Internal::quantize does not properly implement the clamping of the input point. The point is clamped at first and then scaled and quantized. Due to numerical imprecision, the quantized result may exceed the bounds (m_handleSentinel) for large objects like a btStaticPlaneShape, and the algorithm gets mixed up. Therefore, clamping should be done after scaling:

Code: Select all

340,351c336,343
< 	btPoint3 clampedPoint(point);
< 	
< 
< 
< 	clampedPoint.setMax(m_worldAabbMin);
< 	clampedPoint.setMin(m_worldAabbMax);
< 
< 	btVector3 v = (clampedPoint - m_worldAabbMin) * m_quantize;
< 	out[0] = (BP_FP_INT_TYPE)(((BP_FP_INT_TYPE)v.getX() & m_bpHandleMask) | isMax);
< 	out[1] = (BP_FP_INT_TYPE)(((BP_FP_INT_TYPE)v.getY() & m_bpHandleMask) | isMax);
< 	out[2] = (BP_FP_INT_TYPE)(((BP_FP_INT_TYPE)v.getZ() & m_bpHandleMask) | isMax);
< 	
---
> 	btVector3 v = (point - m_worldAabbMin) * m_quantize;
> 	for (int i = 0; i < 3; i++)
> 	{
> 		out[i] = (v[i] <= 0) ? (BP_FP_INT_TYPE) isMax
> 			: (v[i] >= m_handleSentinel) ? (BP_FP_INT_TYPE) ((m_handleSentinel & m_bpHandleMask) | isMax)
> 			: (BP_FP_INT_TYPE) (((BP_FP_INT_TYPE) v[i] & m_bpHandleMask) | isMax);
> 	}

[Erwin Coumans]

Would it be possible to generate a full patch using the latest SVN version, and add it to http://code.google.com/p/bullet/issues/list ?

It makes it easier to apply it using TortoiseSVN or the patch command.
Thanks,
Erwin

By the way, we just applied the inertia tensor computation: http://code.google.com/p/bullet/source/detail?r=1253

[ole.k]

Would it be possible to generate a full patch using the latest SVN version, and add it to http://code.google.com/p/bullet/issues/list ?

Yes, I have just done this.

By the way, we just applied the inertia tensor computation: http://code.google.com/p/bullet/source/detail?r=1253

Great, thanks!

[Erwin Coumans]

The bugfix for the clamping/quantization has now be committed,

http://code.google.com/p/bullet/source/detail?r=1345

Thanks a lot for this!
Erwin