Crashes when porting a newton 2.x program to 3.00

[carli2]

Hi,

I translated the newton 3 header to pascal now and corrected all calls where the interface had changed.
But there are still problems:
The objects fall through the ground and through the landscape (the landscape is a compound collision of convex collisions).
After I add my player (which is a object that does lots of convex casts), I get the following crash:

Code: Select all

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fffd704e700 (LWP 4235)]
0x00007ffff73a9be8 in dgCollisionCompound::CalculateContactsToSingle (this=0xef7b80,
    pair=<optimized out>, proxy=..., useSimd=<optimized out>)
    at ../../source/physics/dgCollisionCompound.cpp:1512
1512                     dgCollisionInstance* const childInstance = me->GetShape();
(gdb) backtrace
#0  0x00007ffff73a9be8 in dgCollisionCompound::CalculateContactsToSingle (this=0xef7b80,
    pair=<optimized out>, proxy=..., useSimd=<optimized out>)
    at ../../source/physics/dgCollisionCompound.cpp:1512
#1  0x00007ffff73acbf1 in dgCollisionCompound::CalculateContacts (this=0xef7b80,
    pair=0x7fffd7032f60, proxy=..., useSimd=0)
    at ../../source/physics/dgCollisionCompound.cpp:1324
#2  0x00007ffff742db8c in dgWorld::CompoundContacts (this=0xde07c0, pair=0x7fffd7032f60,
    proxy=...) at ../../source/physics/dgNarrowPhaseCollision.cpp:1023
#3  0x00007ffff742de98 in dgWorld::CalculateContacts (this=0xde07c0, pair=0x7fffd7032f60,
    timestep=<optimized out>, threadIndex=<optimized out>)
    at ../../source/physics/dgNarrowPhaseCollision.cpp:1410
#4  0x00007ffff742f787 in dgWorld::CollideContinue (this=0xde07c0, collisionSrcA=0x1523f00,
    matrixA=..., velocA=..., omegaA=..., collisionSrcB=<optimized out>, matrixB=...,
    velocB=..., omegaB=..., retTimeStep=@0x7fffd7033da0: 1e+10, points=0x7fffd70338e0,
    normals=0x7fffd7033a60, penetration=0x7fffd7033cb0, maxSize=32, threadIndex=0)
    at ../../source/physics/dgNarrowPhaseCollision.cpp:1617
#5  0x00007ffff738f8a0 in dgBroadPhaseCollision::ConvexCast (this=0xde07f0, shape=0x1523f00,
    matrix0=..., target=..., timeToImpact=@0x7fffd70340ec: 0,
    prefilter=0x5d1350 <MYPREFILTER>, userData=0x7fffd502b840, info=0x7fffd542d060,
    maxContacts=128, threadIndex=0) at ../../source/physics/dgBroadPhaseCollision.cpp:561
#6  0x00007ffff7478a1f in NewtonWorldConvexCast (newtonWorld=<optimized out>,
    matrix=<optimized out>, target=<optimized out>, shape=<optimized out>,
    hitParam=<optimized out>, userData=<optimized out>, prefilter=0x5d1350 <MYPREFILTER>,
    info=0x7fffd542d060, maxContactsCount=128, threadIndex=0)
    at ../../source/newton/Newton.cpp:1360
#7  0x00000000005d1668 in MOVETILL (HOME=0x7fffd70340fc, PARS=0x1523f00)

[Julio Jerez]

do yo uhave a test demo for debugging?

[carli2]

do yo uhave a test demo for debugging?

Win32 or Linux64? (for win32, i need a dll, is "applications/newtonModelEditor/Win32/newton.dll" a valid newton3 dll?)

[Julio Jerez]

win32

[carli2]

Here's the working one using Newton 2.x: http://botwars.tk/winrelease.zip

And this is the broken one with ported newton 3. It should behave nearly equally: http://botwars.tk/winrelease2.zip

Just ignore the JSON parser errors and go to "New game -> Techtest" (Techtest is bottom right)
I use objects with mass 0 to give them a static position.

[Julio Jerez]

somethong is wrong with yor program, when I ruun it Norton says the is a security risk gin wl.exe and shot it down and delete the program.

[carli2]

somethong is wrong with yor program, when I ruun it Norton says the is a security risk gin wl.exe and shot it down and delete the program.

The correct executables filename is gwx.exe, not wl.exe.
In winrelease2 (the newton3 version), it is gwx.exe. In winrelease (newton2) it is gwX.engine (which is basically a renamed .exe, and gwx.exe runs it. you can rename gwX.engine to engine.exe)

[Julio Jerez]

yes that file Norton says that has securtity provblem.

[carli2]

rename it to sth with .exe extension and it will work.
anti viruses are snake oil. when you want to be sure to not get infected by programs sent to you, use a windows live cd.

Edit: try the second zip which is the one using newton 3

[Julio Jerez]

I will again try tonight.

[Julio Jerez]

Ok I download again and rename it, but norton say that there si a dangerus risk on that file and remove it from the machine.

[carli2]

Did you really try this file? http://botwars.tk/winrelease2.zip

And what exactly did you change from newton 2.xx to 3.00 that all objects fall through each other? Did I forget to set a collide bit for the material combination 0-0?

Can you please come to freenode #newtondynamics? it's hard to solve this problem with the forum. (and your commit messages are pointless, i cannot read when exactly you fixed a bug or introduced some incompatibility)

[Julio Jerez]

yes I download thsi file http://botwars.tk/winrelease2.zip
and run with and with out renaming the .exe in both cases norton is very unhappy about and delete the exe


as to what I change form 2.00 to 300, well ther are lot of changes, bu teh main one are thet teh collsion system is now instance base as oppose to reference count base.
the collision system now is a lot more flexible and supports general scaling in all shapes.
the secund major change is that now the engine run on its own thread, this can make run conrurrent or sequencial with the main thread.
The thread system is also exposed so that the application can do stuff by hooking pre or post listeners.
there are many other changes but I do not remember now.

[carli2]

I get the following backtrace:

Code: Select all

0x00007ffff73a5658 in dgCollisionCompound::CalculateContactsToSingle (this=0xeee580, pair=<optimized out>, proxy=..., useSimd=<optimized out>)
    at ../../source/physics/dgCollisionCompound.cpp:1508
1508                     processContacts = constraint->GetMaterial()->m_compoundAABBOverlap (*constraint->GetMaterial(), *compoundBody, *otherBody, proxy.m_threadIndex);
(gdb) backtrace
#0  0x00007ffff73a5658 in dgCollisionCompound::CalculateContactsToSingle (this=0xeee580, pair=<optimized out>, proxy=..., useSimd=<optimized out>)
    at ../../source/physics/dgCollisionCompound.cpp:1508
#1  0x00007ffff73a8661 in dgCollisionCompound::CalculateContacts (this=0xeee580, pair=0x7fffd6dd2f60, proxy=..., useSimd=0)
    at ../../source/physics/dgCollisionCompound.cpp:1320
#2  0x00007ffff7429f0c in dgWorld::CompoundContacts (this=0xddf680, pair=0x7fffd6dd2f60, proxy=...)
    at ../../source/physics/dgNarrowPhaseCollision.cpp:1023
#3  0x00007ffff742a218 in dgWorld::CalculateContacts (this=0xddf680, pair=0x7fffd6dd2f60, timestep=<optimized out>, threadIndex=<optimized out>)
    at ../../source/physics/dgNarrowPhaseCollision.cpp:1410
#4  0x00007ffff742bb07 in dgWorld::CollideContinue (this=0xddf680, collisionSrcA=0xeb79c0, matrixA=..., velocA=..., omegaA=...,
    collisionSrcB=<optimized out>, matrixB=..., velocB=..., omegaB=..., retTimeStep=@0x7fffd6dd3da0: 1e+10, points=0x7fffd6dd38e0,
    normals=0x7fffd6dd3a60, penetration=0x7fffd6dd3cb0, maxSize=32, threadIndex=0) at ../../source/physics/dgNarrowPhaseCollision.cpp:1617
#5  0x00007ffff738ab30 in dgBroadPhaseCollision::ConvexCast (this=0xddf6b0, shape=0xeb79c0, matrix0=..., target=...,
    timeToImpact=@0x7fffd6dd40ec: 0, prefilter=0x5d15c8 <MYPREFILTER>, userData=0x7fffd5d59840, info=0x7fffe1d94060, maxContacts=128,
    threadIndex=0) at ../../source/physics/dgBroadPhaseCollision.cpp:561
#6  0x00007ffff7477d4f in NewtonWorldConvexCast (newtonWorld=<optimized out>, matrix=<optimized out>, target=<optimized out>,
    shape=<optimized out>, hitParam=<optimized out>, userData=<optimized out>, prefilter=0x5d15c8 <MYPREFILTER>, info=0x7fffe1d94060,
    maxContactsCount=128, threadIndex=0) at ../../source/newton/Newton.cpp:1360
#7  0x00000000005d18e0 in MOVETILL (HOME=0x7fffd6dd40fc, PARS=0xeb79c0) at physicscript.pas:620

For threadIndex, I always tell 0 to the engine. Is that wrong?
Also, can two callbacks be called at the same time?

[carli2]

Btw what exactly did you change in the contacts code?
Afaik the contacts are responsible for objects not falling through everything and a call to ConvexCast also crashes at some contacts code.