Object not falling down when changing UserMeshCollision

[Marc]

Hi !

I have a UserMeshCollision. There is a large horizontal triangle. I put a sphere on top and add gravity to it. it falls on the triangle and gets to rest.

Now I remove that triangle, so the next time the collide callback gets called, that triangle won't be returned. But the sphere doens't fall down, it just rests in midair until I put a small force on it.

I guessed it has to do with auto sleep. So I tried

NewtonBodySetAutoSleep(body, 0);
NewtonBodySetFreezeState(body, 0);

(even in front of every NewtonUpdate) on the sphere,

but it's not helping. What could be wrong?

[Julio Jerez]

my gueee is that the collision is no where you think it is.

enable the debug display.

[Marc]

Hmm debug display. I tried to capture the geometry I create dynamically when my NewtonUserMeshCollisionCollideCallback collideCallback gets called. But surprisingly, in this moment, it doesn't get called. When I'm on the flat ground moving around, it gets called exactly once per NewtonUpdate. As soon as I stop moving and velocity gets equal 0, the callback doesn't get called anymore. How is this possible? I always apply a force in -Z direction, and the flat ground is the XY plane. So my sphere gets pushed inside the plane every frame, but without a collidecallback, how come it doesn't fall through the ground?

Just an idea: maybe Newton caches the dynamically created meshes by the callback for the next frame? If the position doesn't change, it uses the same again?

[JernejL]

For debug display julio meant rendering collision shapes with NewtonCollisionForEachPolygonDo

see here: http://www.newtondynamics.com/forum/vie ... 41&start=0

[Marc]

I see.

But that's probably not giving much more insight. My point in my last post was: Newton doesn't ask me for my updated geometry. It can't know that I changed the Geometry, because it doesn't call my callback in this case. But why doesn't it do it?

[Julio Jerez]

Just an idea: maybe Newton caches the dynamically created meshes by the callback for the next frame? If the position doesn't change, it uses the same again?

the is not internal mesk caching, the reason for the call to stop are:
1- the body is sleeping (make sur eteh body is active)
2- the AABB of you mesh is no what is should be, an they body do no overlap (see if teh on AABB overlap is call)
3- the body is smalled then the wold, (see fi teh body is ouzide teh work by settin teh on exit worl call back)

[Marc]

Thank you Julio! I examined those points

1.
I put

Code: Select all

   NewtonBodySetAutoSleep(phys, 0);
   MNewtonBodyUnfreeze(phys);

in my code. Afterwards, I do

Code: Select all

   int freezestate = NewtonBodyGetFreezeState(phys);
   int autosleep = NewtonBodyGetAutoSleep(phys);
   int sleepstate = NewtonBodyGetSleepState(phys);

before every NewtonUpdate() and they all return 0. I'm not 100% sure which values should there be, but I guess all 0 means everything gets simulated. The issue persists with this setup.

2.
My user collision is spans from (0, 0, 0) - (1024, 1024, 1024), placed at (0, 0, 0). I call NewtonSetWorldSize() with an additonal border of 100. So it's
(-100, -100, -100, 1124, 1124, 1124) in size. My sphere is at (512, 512, 512), z radius is 0.9, x and y are 0.5, and falls down 10 in z direction before it hits the triangle.
I added the OnAABBOverlap callback. It gets called every frame, the whole time from the beginning, hitting the triangle, standing still and floating in the air
after removing the triangle.

3.
From the values from 2., I don't think there is a world leaving issue. I added the LeaveWorldCallback anyways just to make sure: It never gets called.


I'm not sure if that helps, but when the sphere is hovering and I call NewtonInvalidateCache() once, the hovering stops and the sphere falls down.

[Julio Jerez]

Is there a test demo?

[Marc]

Sort of. It's similar to my other game you debugged a few times. You can download it here http://dl.dropbox.com/u/103286/filehost/walkover2/walkover2_setup.exe.
You can easily reproduce the issue if you look downwards and click left. That'll remove a part of the terrain like in some popular mining game. When you did this once, you'll stand on a flat quad. Qait until you stand still completly, usually only takes a few seconds, then click again to remove more. => You'll hover in the air until you move around a bit.

I added some debug output. If you have a 2nd monitor, you can move the console window there and have a look while moving around:

If a callback doesn't get called, that line won't appear.

As a note: The onfloor value gets set in NewtonContactProcessCallback, but it's also checked if the collision normal is upwards and the contact is somewhere near the feet and so on, so it's not an indicator to check if NewtonContactProcessCallback processed a collision.