Penetrating terrain

[Leadwerks]

I don't believe you would need to change the serialization format between Newton 2.20 and 2.21.

[macklebee]

no, I had changed the origin of the gmf model but was still using the pre-existing PHY file. So when looking at the model/mesh with DebugPhysics ON, the physics body did not line up with the gmf model.

[Julio Jerez]

With the mesh model and the collision mesh properlly aligned, did you try if the collison to see if still misses?

[macklebee]

yes, that makes no difference. I had made that error when I was trying to clean up the second demo for you. All that did was make the resulting bounces look strange. If you run the demo from the second link i sent you, you will see that you can penetrate the terrain at around 52-53 degrees consistently.

[Julio Jerez]

I just download the new demo, and I see the bug at 52 degree. is happen consistety.

Thank you fo the demo that can reproduce so nicelly, I will fix it.
I am in the middle of another problem right now, it will take me a day to fix and I do no wnat to lose my train of thogh now.
I will fix this BUG tomorrow, or this weekend.

I notice that the cans granade works in most case, so teh fix I mane made it better, this most by some other special condition that at that angle exposes some othe bug,
or shorcoming in the algorith, but thsi si no aceprtable like that, I will fix it of course.

[macklebee]

thanks. understand... one thing at a time!

It appears that its not just an issue with cylinders either... if you load a more complex model, instead of the cylindrical model, it happens more often. I will send you a link in a PM with a different model/PHY file being used. Granted i could easily use a cylindrical physics mesh to help with this, but i thought you might wish to see this...

[Julio Jerez]

I believe I found teh problem, but no the solution.

I did not find it before because it look like to guys are also calling convex cast in the same shape,
so I was aslway getting double calls, and some time evnt triple and quadruple calls for the same shape.
I was untill I realsie that tah teh call start to make sence.

Anyway I thongg teh proble is beacsue tshi shape have and orgin tah is oggset form eth gemerical center, and teh collsion calculation do not compensate fo rthat.
This is a bug bug in the engine, if it is what I think, because when eth engine sping a body in the collision it does it relatvbe to the matrix center and no relative to the center of mass.

It does it right for all other joint and contacts, but I forget to take that into account with collision.
Anyway I have to debug this more, but I thing I know now what it is.

I have not tested wit the other sample you send me I will try that after I get this bug fixed.

[Julio Jerez]

Ok good news, the bug is fixed now, it was relativally simple to fix once i new what was causing it,
I tested with the last demo you send me, the one with a very small capsule and I did not see one single collisiontuneeling when using this new dll.
when using the current DLL, it tuneled in abput 95% of the shots I would say.

Please download SDK 2.22 and test it, let me know if you find any other problem,
also if you find any bug you can post also here and let me know.
I am always in the look for problem and recomendation on how to make the engine better,
the more people report the better ther engien gets because I can put the tiem I spend working on it to were it counts the most.

[macklebee]

Julio - it works great! I even doubled the speed and I still cannot get it to fall through the terrain! nice work!

thanks for tracking this down; its much appreciated!

[Julio Jerez]

Be gentile; do not increase the speed just for the sake of having huge speed.

[macklebee]

Be gentile; do not increase the speed just for the sake of having huge speed.

will do.