Newton Crash

[SpiderPig]

I've not been able to reproduce this problem in a smaller program yet.  I thinks it's happening because I am using CreateMeshCollider().  Sometimes I'm creating single triangles and others are collection of 2 or 3, all with non-shared vertices.  I'm trying to figure out how to narrow down the problem and thought there might be a direct answer to a few questions in the process -

• Does newton not like really small triangles?
• OR vertices that are at the same position or really close?
• Is it meant to work with flat planes (or single triangles) or does it like volume collisions like cube colliders?

[SpiderPig]

Well I think it's understandable why this code crashes.  The second vertex is too close to the first or it makes the triangle to small for newton to handle or something.  @Josh should CreateMeshCollider() this have a some sort of safety check maybe?

#include "UltraEngine.h"
#include "ComponentSystem.h"

using namespace UltraEngine;

int main(int argc, const char* argv[])
{
    auto displays = GetDisplays();
    auto window = CreateWindow("Ultra Engine", 0, 0, 1280, 720, displays[0], WINDOW_CENTER | WINDOW_TITLEBAR);
    auto world = CreateWorld();
    auto framebuffer = CreateFramebuffer(window);

    auto camera = CreateCamera(world);
    camera->SetClearColor(0.125);
    camera->SetFov(70);
    camera->SetPosition(0, 0, -10);

    auto light = CreateDirectionalLight(world);
    light->SetColor(5.0f);
    light->SetRotation(35, 45, 0);

    vector<shared_ptr<Entity>> entities;
    vector<shared_ptr<Entity>> pivots;
    Vec3 emitter_position = Vec3(0.0f, 10.0f, 0.0f);
    float emitter_size = 5.0f;
    uint64_t start = Millisecs(), interaval = 250;



    while (window->Closed() == false and window->KeyDown(KEY_ESCAPE) == false)
    {
        auto time = Millisecs();
        if (time > start + interaval) {
            start = time;

            float x, y, z;
            x = emitter_position.x + Random(-emitter_size, emitter_size);
            y = emitter_position.y + emitter_size;
            z = emitter_position.z + Random(-emitter_size, emitter_size);

            auto e = CreateBox(world);
            e->SetPosition(x, y, z);
            e->SetColor(Random(), Random(), Random());
            e->SetMass(1.0f);

            entities.push_back(e);

            //Create collision object
            auto mesh = CreateMesh();
            mesh->AddVertex(29.0, 3.97082719, 25.0);
            mesh->AddVertex(29.0, 4.0, 25.0178719);
            mesh->AddVertex(29.0397224, 4.0, 25.0);
            mesh->AddPrimitive(0, 1, 2);

            auto pivot = CreatePivot(world);
            auto collider = CreateMeshCollider(mesh);

            pivot->SetCollider(collider);
            pivot->SetCollisionType(COLLISION_SCENE);
            pivots.push_back(pivot);
        }

        world->Update();
        world->Render(framebuffer);
    }
    return 0;
}

 

[Josh]

I added a check for minimum triangle area and edge lengths. I dislike the fact that this is not exact. Please let me know if you encounter this again after today's update.

[SpiderPig]

I will, thanks.  What did you set the minimum too?  Does newton define a minimum or is it a bug on their end you think?

[Josh]

I set both to 0.01. I've brought this up with Julio before and his response was basically "don't do that".

[SpiderPig]

I'll have to re-think a few things then.  Some of the triangles with marching cubes can be a quite small.

[Josh]

If they are that small they will not contribute to collision anyways, so don't worry about it. I'll have an update later today.

[SpiderPig]

My example above still crashes.  I have also seen it crash on something a little larger too.  When I find it again I'll test it in the above program.  Seeing as it's not an accurate test maybe there should just be a warning in the documentation and make users aware of it?  If anything I think Newton should have a check for it on their end.

[Josh]

If you can reliably produce the error please give me some code to run.

[SpiderPig]

I updated the example in my second post.  I'm using a triangle that was crashing in my other project.  It's still pretty small but it is a visible triangle.  I think the edge lengths of this triangle is 0.03 and 0.017.

[SpiderPig]

I've been trying to solve this.  While this code does seem stop small triangles being created I still will get a crash sooner rather than later.  So who knows.  Maybe my code isn't flawless.

void CollisionPatch::AddTriangle(int a, int b, int c) {
	Vec3 min = { FLT_MAX, FLT_MAX, FLT_MAX }, max = { FLT_MIN, FLT_MIN, FLT_MIN };
	int i[3] = { a,b,c };
	for (auto id = 0; id < 3; id++) {
		min.x = Min(min.x, vertices[i[id]].position.x);
		min.y = Min(min.y, vertices[i[id]].position.y);
		min.z = Min(min.z, vertices[i[id]].position.z);

		max.x = Max(max.x, vertices[i[id]].position.x);
		max.y = Max(max.y, vertices[i[id]].position.y);
		max.z = Max(max.z, vertices[i[id]].position.z);
	}

	auto threshold = 0.5f;
	auto dx = max.x - min.x;
	auto dy = max.y - min.y;
	auto dz = max.z - min.z;

	auto count = 0;
	if (dx < threshold) { count++; }
	if (dy < threshold) { count++; }
	if (dz < threshold) { count++; }

	if (count > 1) {
		return;
	}
	
	indices.emplace_back(a);
	indices.emplace_back(b);
	indices.emplace_back(c);
}

 

[SpiderPig]

As well as the crash in newton I get a crash here sometimes too.  Don't know what's causing it, it happens in conjunction with the newton crash.

[Josh]

An update is available now which fixes this. Let me know if you see any other problems.

[SpiderPig]

I'm downloading it now.  Do you mean you perfected the triangle size test for CreateCollsionMesh()? 

[Josh]

I don't know if it's perfect, but your example no longer causes any problem.

[SpiderPig]

My example is working now but I still get crashes in my voxel terrain project.  There must be something else going on.

[SpiderPig]

Are we able to debug newton ourselves?  Is that possible?

[Josh]

It is. If you download the last versiion of Newton 3.14 on Github you can compile a DLL and run your app to debug the DLL.

[SpiderPig]

I think I've compile the right one "newton-dynamics-master\newton-3.14\sdk\projects\visualStudio_2015_dll", I copied the "newton_d.dll" that was generated into my project and it crashes.  Nothing is ever easy.  I also tried replacing "dCustomJoints_d.dll" but it gave a similar error.  Perhaps some things have changed?

The DLL is also about 2mb bigger than the one shipped with Ultra.  Anyway it's here if you like to try it for yourself.  I don't know how I'm going to pinpoint theses errors without debugging newton but we will see.

newton_d.zip

[Josh]

I have uploaded my build of Newton 3 here:
https://github.com/UltraEngine/NewtonDynamics3

It would probably be better for me to debug your project. I don't think you will get any useful information out of the Newton debug output.

[SpiderPig]

Thanks that worked.  You are correct.  Although it gives me an idea on what's going on it doesn't help me much.  But it is cool to see it break to the physics code and at least tell me something.  Your welcome to remote debug if you wish.

[Josh]

I need the code anyways, so might as well send me the project.

[SpiderPig]

Okay, I will collect everything into a smaller project for you tomorrow evening.

[Josh]

I can't say what the problem was but I clean and rebuild of the engine fixed it. The update is available now on 1.0.1.

BTW, it seems a lot of your colliders consist of 1-2 triangles. This is probably very inefficient and it would be better if they were in bigger pieces, but that should still not cause a crash like you were experiencing.

[SpiderPig]

Thanks its all working now.

26 minutes ago, Josh said:

BTW, it seems a lot of your colliders consist of 1-2 triangles. This is probably very inefficient and it would be better if they were in bigger pieces, but that should still not cause a crash like you were experiencing.

They do and I was actually wondering if it would be better to make less colliders with more triangles each so I will do this.