Funny business with compound collisions

[hpesoj]

Using Newton 2.09, the code below produces some odd results. Firstly, the compound collision shape is destroyed (or at least the destruction callback is called) before the body is destroyed. Secondly, when the body is destroyed the collision destruction callback is called with an erroneous collision pointer. I suspect something is amiss .

Code: Select all

#include <iostream>
#include <Newton.h>

void collisionDestructorCallback(const NewtonCollision* collision)
{
    std::cout << "destroy: " << collision << std::endl;
}

int main(void)
{
    NewtonWorld* world = NewtonCreate(0, 0);
    NewtonSetCollisionDestructor(world, collisionDestructorCallback);

    NewtonCollision* box = NewtonCreateBox(world, 1.0f, 1.0f, 1.0f, 0, 0);
    std::cout << "box: " << box << std::endl;
    NewtonCollision* compound = NewtonCreateCompoundCollision(world, 1, &box, 0);
    std::cout << "compound: " << compound << std::endl;

    NewtonBody* body = NewtonCreateBody(world, compound);

    std::cout << "Releasing compound..." << std::endl;
    NewtonReleaseCollision(world, compound);

    std::cout << "Releasing box..." << std::endl;
    NewtonReleaseCollision(world, box);

    std::cout << "Destroying body..." << std::endl;
    NewtonDestroyBody(world, body);

    NewtonDestroy(world);
}


[Julio Jerez]

I run it and it is correct, this is the output

box: 028FEEE0
compound: 02AD00A0
Releasing compound...
destroy: 02AD00A0

Releasing box...
Destroying body...
destroy: 02AD0160
destroy: 028FEEE0

it destroy the Cpunpount because Compound are not shared, when you attack a compound collison to a Body, teh Boidy Take a Copy of of that collision, It does tah for all no share shapes.
if it was allow to shared it teh it will malfuntion if you assign a compound to more than one body.

when you destroy the Body, the body release its compound and then compound release all the children and teh get destroyed.
The erroneos collision pointer is the Collision Shape Newton Generate from the Compound you passed in creations of the body.

I think is alright, does it cause preoblems for you?

[hpesoj]

I didn't realise this was the way compound collisions worked. The wiki would lead you to believe otherwise:

Note for Newton 1.53:

* compound collisions are instance objects that can not be shared among other rigid bodies

(In Newton 2.0, this is no longer an issue - compound collisions may be reused/shared by multiple rigid bodies just like any other collision object)

It isn't a huge problem, my program keeps track of collision shapes I have created and when the destructor callback is called it indexes into a lookup table to find and delete the associated user data. I was unaware that Newton was creating a collision shape internally, and so my assert was triggered when the unknown collision pointer was passed to the callback.

[Julio Jerez]

that was in 1.53
But lot of people were having problem with compound collision trying to re-use them.
most people found hard to make muticompound collision so I chnage teh for 2.0 by making a copy.


Keeping Track of the collision is good for leack traking but in production code you do not really need it, It do e no harm.

Youc could trackl teh collision that in in eh Body, not the one you pass in,

soemthong like
if (collision == NewtonBodyGetCollsion()){
track (collision)
} else {
track (NewtonBodyGetCollsion())
}

or you could use the GetInfo and check the Ref Counters.

[JernejL]

that was in 1.53
But lot of people were having problem with compound collision trying to re-use them.
most people found hard to make muticompound collision so I chnage teh for 2.0 by making a copy.

Interesting, i didn't know that each time i do this i get a new copy of compound, i use this a lot in my spaceships game, will be good to know when i start working on cleanup routines.