New btInternalEdgeUtility has been added

[edl]

I use the default vehicleDemo to modify as this program,after initialization, drive the car inside the tunnel , press "w" to show the wireframe, move the car slowly along the wall. Crash the triangular dense point, sometimes, it suddenly mad.

There is also a problem, when I put this InternalEdge patch used in a large scene,the program just crash after it starts, and I am still checking what 's the problem.

InternalEdge patch is very important, when the scene goes big,when the cars across the shape edge, 50% chance would make car go mad or get a big impulse.

[edl]

and more,in debug mode, the quaternion value can be zero value.

it cause some trouble

this is the call stack

[Dirk Gregorius]

Maybe the quaternion is not a bug. It depends where the axis comes from. You could construct an identity quaternion in this case well, but it depends on the context.

[Erwin Coumans]

Thanks a lot for the reproduction case.

The current btInternalEdgeUtility tries to deal with double sided triangles, but that can cause troubles with the heuristic that is used. I'm going to simplify the utility to only work with single sided triangles, that should make it more robust.

The btQuaternion assert is likely a zero-area / degenerate triangle (two vertices are very close or identical). Can you check this?
Thanks,
Erwin

[edl]

Maybe the quaternion is not a bug. It depends where the axis comes from. You could construct an identity quaternion in this case well, but it depends on the context.

i think you are right, i have two scene,a big one,and another simple small one,the small one didn't encounter this problem.

Thanks a lot for the reproduction case.

The current btInternalEdgeUtility tries to deal with double sided triangles, but that can cause troubles with the heuristic that is used. I'm going to simplify the utility to only work with single sided triangles, that should make it more robust.

The btQuaternion assert is likely a zero-area / degenerate triangle (two vertices are very close or identical). Can you check this?
Thanks,
Erwin

waitting for your single sided mode .

I have a temporary ignore this problem,the artists making the scene(a very big scene,contain a lot of elements),i have tried ,and found it seems not very easy to find the problem, and i have a lots of work to do , so i abandoned....at least, the release mode is ok....(do not laugh at me lazy )

[sparkprime]

It'd be good if there was an option to keep the 2-sided behaviour. Has it been established that there is a serious problem in the absence of degenerate triangles? If you have to penetrate the mesh then it could be the same bug as I was reporting with the cylinder spawning slightly within the mesh. So if you do add a single-sided mode I'll test it out too.

[Erwin Coumans]

It'd be good if there was an option to keep the 2-sided behaviour. Has it been established that there is a serious problem in the absence of degenerate triangles? If you have to penetrate the mesh then it could be the same bug as I was reporting with the cylinder spawning slightly within the mesh. So if you do add a single-sided mode I'll test it out too.

Can you please provide instructions how to reproduce this cylinder issue? A video doesn't help fixing the issue.

Thanks a lot,
Erwin

[sparkprime]

I will try but i'm buried at the moment

[Erwin Coumans]

I improved the btInternalEdgeUtility, hopefully it works better now, in particular for single sided triangle meshes.

edl, can you please test latest trunk with your race track? By default the btAdjustInternalEdgeContacts favors front facing triangles normals. It has some optional settings for BT_TRIANGLE_CONVEX_BACKFACE_MODE.

@sparkprime: it is best to use single sided triangles, consistently front facing or backfacing. There is some experimental support for double sided triangles, enable it using BT_TRIANGLE_CONCAVE_DOUBLE_SIDED+BT_TRIANGLE_CONVEX_DOUBLE_SIDED.

Thanks for the help,
Erwin

[JanK]

How does one use the utility in Bullet 2.76 ? For now i have to create the data-structure, put the pointer somewhere and do a lookup manually in a collision-callback. Is this supposed to stay that way?

I think either it should work automatically out of the box, or it should simply be a switch on triangle-meshes to enable or disable it (at creation time).

Jan.

[Erwin Coumans]

Once the utility has been tested and developers are happy with it, we will do a more tight integration (after 2.76 release later this week)

Have you tried the latest version?

Thanks,
Erwin

[edl]

I improved the btInternalEdgeUtility, hopefully it works better now, in particular for single sided triangle meshes.

edl, can you please test latest trunk with your race track? By default the btAdjustInternalEdgeContacts favors front facing triangles normals. It has some optional settings for BT_TRIANGLE_CONVEX_BACKFACE_MODE.

@sparkprime: it is best to use single sided triangles, consistently front facing or backfacing. There is some experimental support for double sided triangles, enable it using BT_TRIANGLE_CONCAVE_DOUBLE_SIDED+BT_TRIANGLE_CONVEX_DOUBLE_SIDED.

Thanks for the help,
Erwin

i have test the latest googleCode svn ver 1945 with my race track ,it seems the new code work well,
thanks for the work! ,if i find problems,i will post it here.

[Stillmoon]

I found a very minor issue with btInternalEdgeUtility.h. Line 5 reads

Code: Select all

#include <LinearMath/btHashMap.h>

But since I set the Bullet source directory as a user path, not a system path, I had to change it to

Code: Select all

#include "LinearMath/btHashMap.h"

The quotes appear to be consistent with all other Bullet header #includes.

I'm using SVN revision 1936 of the file.

[scarrow]

I found a small issue in the case of degenerate triangles. In this loop numshared could go to 3 and then try to write a value beyond the end of the array:

Code: Select all

		for (int i=0;i<3;i++)
		{
			for (int j=0;j<3;j++)
			{
				if ( (m_triangleVerticesA[i]-triangle[j]).length2() < m_triangleInfoMap->m_equalVertexThreshold)
				{
					sharedVertsA[numshared] = i;
					sharedVertsB[numshared] = j;
					numshared++;
				}
			}
		}

My modified version looks like this:

Code: Select all

		for (int i=0;i<3;i++)
		{
			for (int j=0;j<3;j++)
			{
				if ( (m_triangleVerticesA[i]-triangle[j]).length2() < m_triangleInfoMap->m_equalVertexThreshold)
				{
					sharedVertsA[numshared] = i;
					sharedVertsB[numshared] = j;
					numshared++;
					if(numshared >= 3)
						break;
				}
			}
			if(numshared >= 3)
				break;
		}

As yet untested, but presumably fine.

[Stillmoon]

I found a very minor issue with btInternalEdgeUtility.h. Line 5 reads

Code: Select all

#include <LinearMath/btHashMap.h>

But since I set the Bullet source directory as a user path, not a system path, I had to change it to

Code: Select all

#include "LinearMath/btHashMap.h"

The quotes appear to be consistent with all other Bullet header #includes.

I'm using SVN revision 1936 of the file.

This is now fixed in the svn trunk, thanks Erwin!