Assertion failed in btSequentialImpulseConstraintSolver

[wittmeis]

Hi,

I wanted to update our robotics simulation software today from a custom (slightly modified) Bullet-2.78 version to the newest Bullet-2.80 (r2531).

The application compiles but as soon as I want to step the btDiscreteDynamicsWorld I get an assertion failed error in btSequentialImpulseConstraintSolver. I checked, the identical assertion also exists in 2.78. What changed from 2.78 to 2.80?
Here is the error message:

bullet-2.80-rev2531/src/BulletDynamics/ConstraintSolver/btSequentialImpulseConstraintSolver.cpp:872: virtual btScalar btSequentialImpulseConstraintSolver::solveGroupCacheFriendlySetup(btCollisionObject**, int, btPersistentManifold**, int, btTypedConstraint**, int, const btContactSolverInfo&, btIDebugDraw*, btStackAlloc*): Assertion `info2.rowskip*sizeof(btScalar)== sizeof(btSolverConstraint)' failed.

And here is the stack backtrace:

#0 0x00132416 in __kernel_vsyscall ()
#1 0x010511ef in raise () from /lib/i386-linux-gnu/libc.so.6
#2 0x01054835 in abort () from /lib/i386-linux-gnu/libc.so.6
#3 0x0104a095 in ?? () from /lib/i386-linux-gnu/libc.so.6
#4 0x0104a147 in __assert_fail () from /lib/i386-linux-gnu/libc.so.6
#5 0x04b36542 in btSequentialImpulseConstraintSolver::solveGroupCacheFriendlySetup (this=0x8996ea0, bodies=0xb110a710, numBodies=2, manifoldPtr=0x0, numManifolds=0, constraints=0xb110a730, numConstraints=2, infoGlobal=...,
debugDrawer=0x0, stackAlloc=0x893e020) at /home/wittmeis/downloads/bullet-2.80-rev2531/src/BulletDynamics/ConstraintSolver/btSequentialImpulseConstraintSolver.cpp:872
#6 0x04b37acb in btSequentialImpulseConstraintSolver::solveGroup (this=0x8996ea0, bodies=0xb110a710, numBodies=2, manifoldPtr=0x0, numManifolds=0, constraints=0xb110a730, numConstraints=2, infoGlobal=..., debugDrawer=0x0,
stackAlloc=0x893e020) at /home/wittmeis/downloads/bullet-2.80-rev2531/src/BulletDynamics/ConstraintSolver/btSequentialImpulseConstraintSolver.cpp:1249
#7 0x04b426af in InplaceSolverIslandCallback::processConstraints (this=0x8581c90) at /home/wittmeis/downloads/bullet-2.80-rev2531/src/BulletDynamics/Dynamics/btDiscreteDynamicsWorld.cpp:202
#8 0x04b3ec31 in btDiscreteDynamicsWorld::solveConstraints (this=0x8598f00, solverInfo=...) at /home/wittmeis/downloads/bullet-2.80-rev2531/src/BulletDynamics/Dynamics/btDiscreteDynamicsWorld.cpp:731
#9 0x04b3e17a in btDiscreteDynamicsWorld::internalSingleStepSimulation (this=0x8598f00, timeStep=0.001) at /home/wittmeis/downloads/bullet-2.80-rev2531/src/BulletDynamics/Dynamics/btDiscreteDynamicsWorld.cpp:505
#10 0x04b3e00a in btDiscreteDynamicsWorld::stepSimulation (this=0x8598f00, timeStep=0.0017210905207321048, maxSubSteps=7, fixedTimeStep=0.001)

Any help is appreciated!

Thanks!!

[wittmeis]

Hi,

I digged deeper and I found that the reason seems to be that we use double precision. When double precision is turned on, the size of btConstraintSolver is 268 bytes and of btScalar 8 bytes. Hence, the info2.rowSkip becomes 33.5 (line 870 of btSequentialImpulseConstraintSolver.cpp) and the assertion in line 872 fails.

If double precision is off, btConstraintSolver is 140 bytes and btScalar 4 bytes. Therefore, rowSkip is 35 and the assertion does not fail.

So maybe there is something wrong with the 64bit alignment of the members of btConstraintSolver????


Any help appreciated!

[morrishazy]

I had the same issue on Linux, but not on Windows and after reading your post I noticed some unions of the btSolverConstraint class had paddings and some didn't. The assertion disappeared after adding union padding for all members, which type was not btScalar.

[Erwin Coumans]

Latest trunk should fix this assert, can you check it?
https://code.google.com/p/bullet/source/detail?r=2553

If not, please file a new issue in the issue tracker at http://bullet.googlecode.com
Thanks!

[morrishazy]

Seems to work fine now