Newton 2.0x Archemedia Open Beta

[JernejL]

EDIT: where is the documentation?

EDIT2: Where is the NewtonContact structure?

It is all here: http://newtondynamics.com/wiki/index.ph ... =Main_Page
newtoncontact is a contact joint (pointer) - NOT a structure.

[rvangaal]

I've updated the SVN version just now (May 10th); when compiling I get this error:

3>..\..\..\source\newton\Newton.cpp(8153) : error C2660: 'dgMeshEffect::RepairTJoints' : function does not take 0 arguments

Indeed it is defined as:

void dgMeshEffect::RepairTJoints (bool triangulate)

and dgMeshEffect.h also has no default argument which could make it work:

void RepairTJoints (bool triangulate);

My environment is Visual Studio 2008, configuration 'releaseDll|Win32'. When trying 'release|Win32' I get the same error.
Any ideas?

[Julio Jerez]

I just commit all my files and it compile fine here,
just to make sure I did no have anythong checked ot Sync Again and see if it works now.

we have teh same vesion of VS so it soudl work now.

[rvangaal]

Ok that fixed part of it; now I get (when doing a full Rebuild):

1>..\..\..\source\physics\dgMeshEffect2.cpp(30) : fatal error C1083: Cannot open include file: 'hacdCircularList.h': No such file or directory

Did you SVN add that?

[Julio Jerez]

Yes I was integration that Library with the engine, thosw filw should be in folder
C:\Users\Julio\Desktop\newton-dynamics\coreLibrary_200\contributions\hacd\src\HACD_Lib\inc

when I check for differences in SVN they aren't any.
you should have it and you should also have the right VS project file,

[rvangaal]

It may be because I'm compiling v3.0; I opened newton\coreLibrary_300\projets\windows\project_vs2008\newton.sln . I just tried opening newton.sln from coreLibrary_200 and that indeed works! Seems the contribs are not in v3.0 (yet?).

Thanks for the help.

[Julio Jerez]

do not run 3.0 yet until I announce it it is ok to use.

[KingSnail]

May I suggest something for Newton 3.0?

Ability to make rigid bodies from many threads without critical sections. It seems nvidia physx 3.0 has this feature.

Ability to move bodies between worlds, preserving state.

Proxy based character controller.

Ability to raycast from any thread without worries of crash.

[Julio Jerez]

Ability to make rigid bodies from many threads without critical sections. It seems nvidia physx 3.0 has this feature.
I belive you can do that now.

Ability to move bodies between worlds, preserving state.
that is a problem in newton, I said I was going to expose the function pair remove/add bodyies from work
The engine do have tha function internally, but it was never exposed.
what the engine can not do and will never do is to have bodies belong to mutiples world at the same time.

Proxy based character controller.
Yes,

Ability to raycast from any thread without worries of crash.
to my knwoledge this is hwo it works now. Reay cast, collision, and convex cats is re-entrance in all versions of newton, since version 1.0

[KingSnail]

Well no matter what I tried, raycast crashes for me when its on main thread
and NewtonUpdate is on dPhysicsThread.

This is with single thread newton architecture mode.

When I made new character controllers for each new connection when newton was updaing, I got many crashes,
I could only solve it by having 1000 controllers pre made before main server loop started.

[Julio Jerez]

let us take one problem at a time. the ray cast should not crash onder any circuntances.

you a test but I could not play it no matter what I did, and I gave up. how can I test it?

[KingSnail]

Well the original issue was fixed (tunnel bug), you were right the scene collision fixed the problem.
I will have to make a new demo for the raycast bug.

I will PM you it when its made.
The only way to run it, is to have the server running same time as client.
And firewall has to allow the connection. Because its online game.

[KingSnail]

One more thing I forgot to ask, are ragdolls being implemeted in newton 3?
Would be cool to be able to decompose a mesh into newton primitives and somehow
link the head, body and arms so when a player dies, they can use physics simulation instead of animation.

[JernejL]

You can already do ragdolls with ball & socket joints.

[rvangaal]

I noticed crashes during NewtonTreeCollisionEndBuild() with large trees. This was in Newton 2.29; I just updated SVN, compiled the 2.34 DLL, copied the new .h/.dll/.lib files and it still crashes.
At the limit I an generate a tree file of around 159Mb. A stacktrace with the release DLL shows this:

0x0033667E d:\source\trunk\external\newton\corelibrary_200\source\core\dgstack.h (line 40): dgStackBase::dgStackBase()
0x003B7359 d:\source\trunk\external\newton\corelibrary_200\source\core\dgtypes.cpp (line 575): dgVertexListToIndexList()
0x003C0B3D d:\source\trunk\external\newton\corelibrary_200\source\core\dgaabbpolygonsoup.cpp (line 1655): dgAABBPolygonSoup::Create()
0x0036A2F0 d:\source\trunk\external\newton\corelibrary_200\source\physics\dgcollisionbvh.cpp (line 97): dgCollisionBVH::EndBuild()
0x0033339F d:\source\trunk\external\newton\corelibrary_200\source\newton\newton.cpp (line 3679): NewtonTreeCollisionEndBuild()
0x005DBE0F [tracked]: (filename not available): (function-name not available)

To debug this, should I:
- use the regular newton.h, link with newton_d.lib, and copy newton_d.dll? In other words, should I rename newton_d.dll to newton.dll in my app directory, or will it load newton_d.dll instead of newton.dll?
- will I probably get more information than the above?

I could generate a software zip that exhibits the problem (I use the newton.dll) but the mesh is around 240Mb unzipped, probably the software package would be around 150Mb zipped.
The problem is consistent; at some point, if I add too much, it starts crashing (I already tried NOT optimized the tree soup).

*EDIT* : I tried the newton_d.dll and it crashes after a failing m_malloc() in MallocLow(), trying to get 224Mb of memory. At that time my process is around 800Mb in memory size. It seems a simple out of memory failure, hm. Thought that didn't happen anymore in these virtual memory days. Hm.