Crashes when porting a newton 2.x program to 3.00

[Julio Jerez]

I do not remember is code has has evolve a lot form core 200, it is better in every way but I do not have a ldetail ist of the changes.

like I say before try these totorial that are all in the download
http://newtondynamics.com/wiki/index.ph ... gid_Bodies

[carli2]

Maybe you can look over this patch in my game which basically introduces newton 3
http://goldenwipf.de/hg/index.cgi/rev/42d273945e12

Maybe you find what I have missed.

[carli2]

I do not remember is code has has evolve a lot form core 200, it is better in every way but I do not have a ldetail ist of the changes.

like I say before try these totorial that are all in the download
http://newtondynamics.com/wiki/index.ph ... gid_Bodies

NewtonSetWorldSize (world, &minBox[0], &maxBox[0]); is a Newton 2.xx function. But Newton 2.xx is working for me.
I'm looking for Newton 3 Tutorial.

[Julio Jerez]

I said the tutorial are up to day in the svn archive.
Delfi who was helping me with the the forum and the wiki put the tutorial text when I wrote them,
but he has not being aroun in a long time and I forget how to log int.

In the SDK dowload for SVN, those differences are corrected and they work with SDK core 300.

[carli2]

I looked at the tutorials in the SVN multiple times. But i don't see anything I'm missing though my code worked for newton 2.xx

Did you look at the patch? http://goldenwipf.de/hg/index.cgi/rev/42d273945e12

And did you change the matrix row/col order or anything like that?

[Julio Jerez]

It is difficult to see what is different in a patch. all I can see i sthat yuo can NewtonUpdate from with in a critical section, I do not think that is right.
no the matrix order has not changed.

also I do not know why you have so many probem moving to newton 300, they chnage requore no more thjan a n hour of two of editing and recompiling.
Not one else has has so many problems.

[carli2]

It is difficult to see what is different in a patch. all I can see i sthat yuo can NewtonUpdate from with in a critical section, I do not think that is right.
no the matrix order has not changed.

Removing the critical section code is not working, too.

also I do not know why you have so many probem moving to newton 300, they chnage requore no more thjan a n hour of two of editing and recompiling.
Not one else has has so many problems.

Yes, that's exactly what I've done. But after that, I saw a world where most objects had some nasty matrices and some objects fell through the ground.

I removed the call to NewtonUpdate() and now I saw that some objects had strange matrices. I should see if the binary representation of matrices changed.

[carli2]

I just logged all matrices of the objects and I observed the following:
- On creation, correct matrices were set
- After simulation, the matrices hat 0 in the 3x3 submatrix
- This explains why my objects are showing up that weird
- ConvexCast returns "no collision" even if there is the ground plate

Here's the log:
http://paste.ubuntu.com/1089780/

A line "create" means a object is created
A line "set new" means a setMatrix call
A line "MoveTill" means a setmatrix after the path was checked with convexCast
A line "Matrix" means the updated matrix of the object after a simulation step

[Julio Jerez]

can you buidl the library in debug mode? the engine has many assert that will show where and went the matrix when wrong.

[carli2]

I built Newton with "-DNDEBUG -g", but no errors on stdout.
I assume it's sth with the data representation that changed. Matrices are dFloat *. I assume they are single float aligned to 4 and there are 16 floats behind such a pointer.
But the matrix corruption seems to happen in newton, not in the application.


btw the compilation showed the following:

Code: Select all

../../source/physics/dgNarrowPhaseCollision.cpp: In member function ‘void dgWorld::SceneContactsSimd(dgCollidingPairCollector::dgPair*, dgCollisionParamProxy&) const’:
../../source/physics/dgNarrowPhaseCollision.cpp:1315:101: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]

[Julio Jerez]

you need to open file ../coreLibrary_300\source\core\dgTypes.h
and define _ASSERTE to the proper Linux assert macro and compile the library

Code: Select all

#if !(defined (_WIN_32_VER) || defined (_WIN_64_VER))
   #ifdef _DEBUG
      #define _ASSERTE(x) assert(x)
   #else
// define the standrd OS assert here.
      #define _ASSERTE(x)
   #endif
#endif

and I repeat the matrices had not changed, all matrices in the low level library are 32 byte aligned in core 300 and 16 bytes aligned in core 200.
matrices has copy contructor to do teh proper aligmnet.

[carli2]

After some changes at the trace code and other things, I got the assertions working.
Here's the first assertion fail:

Code: Select all

gwX.engine: ../../source/core/dgThreadHive.h:122: void dgThreadHive::GetIndirectLock(dgInt32*, dgUnsigned32) const: Assertion `dgInt32 (sizeof (dgInt32)) == dgInt32 (sizeof (long))' failed.

Next assertion:

Code: Select all

gwX.engine: ../../source/physics/dgBody.cpp:495: virtual void dgBody::UpdateCollisionMatrix(dgFloat32, dgInt32): Assertion `0' failed.

[Julio Jerez]

Ha, I see you are in linux 64. This:

Code: Select all

gwX.engine: ../../source/core/dgThreadHive.h:122: void dgThreadHive::GetIndirectLock(dgInt32*, dgUnsigned32) const: Assertion `dgInt32 (sizeof (dgInt32)) == dgInt32 (sizeof (long))' failed.

Is a legacy from when I was not using ptread, but now I am. all the lock are using int not long.
In 64 bit linux long are 64 bit ints, in 64 VS long are 32 bit int, xcode has an option.
You can comment that line out for now, I fix it in next check in.
//_ASSERTE (dgInt32 (sizeof (dgInt32)) == dgInt32 (sizeof (long)));

The secund you has Continue collision enable, do no use for now until it is fully ready.

[carli2]

Yep. I commented out the assertions for the first case.
After deactivating continuous collision, i get the following:

Code: Select all

gwX.engine: ../../source/physics/dgBody.h:766: virtual void dgBody::SetMatrixOriginAndRotation(const dgMatrix&): Assertion `dgAbsf (val - 1.0f) < 1.0e-5f' failed.

[Julio Jerez]

where it is called from?