Voxel Engine?

[Carli]

I have an other problem:
NewtonCreateCompoundCollision crashes when i want to put some ConvexHulls into it.
I'm using Linux64 and Freepascal.

Code: Select all

type
     TCollisioninfo = array of array of TVec3;
     TCollisioninfos = array of record col: TCollisioninfo; pos: TVec3; end;


procedure TLogicEngine.CreatePhysicVoxelMap;
var col: TCollisioninfos;
    col2: array of PNewtonCollision;
    col3: PNewtonCollision;
    bod: PNewtonBody;
    mat: TMatrix4;
    i, j: integer;
begin
 mat:=IdentityMatrix;
 col:=graphic.map.exportcollision;
 for i:=0 to high(col) do begin
  if length(col[i].col)<>0 then begin
   mat[3][0]:=col[i].pos[0];
   mat[3][1]:=col[i].pos[1];
   mat[3][2]:=col[i].pos[2];
   setlength(col2,length(col[i].col));
   for j:=0 to high(col2) do begin
    col2[j]:=NewtonCreateConvexHull(world, length(col[i].col[j]), @col[i].col[j][0], sizeof(TVec3), 0.1, 0, @IdentityMatrix);
   end;
   col3:=NewtonCreateCompoundCollision(world, length(col2), col2, 0);
   bod:=NewtonCreateBody(world, col3);
   NewtonBodySetMatrix(bod, @mat[0][0]);
  end;
 end;
end;


Without binding the Convex Hulls to Compounds, it works fine. But with creating a compound collision, I get a SIGSEGV error.

[Julio Jerez]

it should not crash.
The only thing I can see tah coudl be a problem is the you do no check if teh conve genereate a shape

Code: Select all

col2[j]:=NewtonCreateConvexHull(world, length(col[i].col[j]), @col[i].col[j][0], sizeof(TVec3), 0.1, 0, @IdentityMatrix);
   end;

if should do:

Code: Select all

shape = NewtonCreateConvexHull(world, length(col[i].col[j]), @col[i].col[j][0], sizeof(TVec3), 0.1, 0, @IdentityMatrix);
   end;
if (shape)
     col2[j]:=NewtonCreateConvexHull(world, length(col[i].col[j]), @col[i].col[j][0], sizeof(TVec3), 0.1, 0, @IdentityMatrix);

NewtonCreateConvexHull is the only shape that can return NULL, and this is because if the points do not form a convex hull then it is not possible to make a convex collssion shape.

It is hard to see the code you posted.
if you have a exe that I can test, I can check it out.

[Carli]

The Shape does never return null because I have 8 points each (-->naive voxels)

i will upload a .exe for you.

http://www.file-upload.net/download-230 ... X.exe.html

You will need some Libs: http://wwwpub.zih.tu-dresden.de/~s3734770/RawLibs.rar
and the data:
http://goldenwipf.de/hg/index.cgi/archive/tip.zip

The debugging symbols should be left in the program.

I hope you will find the bug.
It's independend from the OS i noticed.
It is also not relevant if I have just one object or more than one.
If i use other collision types instead, i get the same result.

[Carli]

AAh solved!!

the pascal headers had a bug:

@Pascal Header: There should be some changes:

Code: Select all

type
  PPNewtonCollision = ^PNewtonCollision;


function NewtonCreateCompoundCollision( const newtonWorld : PNewtonWorld; count : int;
                                        const collisionPrimitiveArray : PPNewtonCollision; shapeID : Int ) : PNewtonCollision; cdecl; external{$IFDEF __GPC__}name 'NewtonCreateCompoundCollision'{$ELSE}NewtonDLL{$ENDIF __GPC__};



Freepascal has an other internal representation of a dynamic array.
PPNewtonCollision should be the right datatype for the collisionPrimitiveArray.

thx for your effort

[Julio Jerez]

Ha, glad you figured it out.

[Carli]

yes, it would have been an odyssee when you would have started debugging

[JernejL]

I don't understand.. i use this:

Code: Select all

type
PNewtonCollision = ^Pointer;

function  NewtonCreateCompoundCollision (const newtonWorld: PNewtonWorld; count: integer; const collisionPrimitiveArray: array of PNewtonCollision; shapeid: integer): PNewtonCollision; cdecl; external newtondll;

And this works fine with freepascal (in delphi compatibility mode ofcourse)

Whose header did you use? there's.. i think 3 pascal headers for 2.0 floating around.

[Stucuk]

@Carli: Why would that make any difference? ^Pointer is still a Pointer. Also the "Official" headers(Ones found in the 2.0 Header thread) have that in them already (With NICE_CODE_PARAMS defined, tho its defined by default).

Code: Select all

{$IFDEF NICE_CODE_PARAMS}
  PNewtonMesh = ^Pointer;
  PNewtonBody = ^Pointer;
  PNewtonWorld = ^Pointer;
  PNewtonJoint = ^Pointer;
  PNewtonContact = ^Pointer;
  PNewtonMaterial = ^Pointer;
  PNewtonCollision = ^Pointer;
  PNewtonSceneProxy = ^Pointer;
  PNewtonBreakableComponentMesh = ^Pointer;
  // JointLibrary
  PNewtonUserJoint = ^Pointer;

  //PNewtonRagDoll = ^Pointer;
  //PNewtonRagDollBone = ^Pointer;
{$ELSE}

They also have it as an array:

Code: Select all

type
  TCollisionPrimitiveArray = array of PNewtonCollision;

function NewtonCreateCompoundCollision( const newtonWorld : PNewtonWorld; count : int;
                                        const collisionPrimitiveArray : TcollisionPrimitiveArray; shapeID : Int ) : PNewtonCollision; cdecl; external{$IFDEF __GPC__}name 'NewtonCreateCompoundCollision'{$ELSE}NewtonDLL{$ENDIF __GPC__};

i think 3 pascal headers for 2.0 floating around.

@Delfi: I thought there was only yours and the one that is a descendent of the 1.53 one (One i did).

[JernejL]

There's my header (which was built based on ancient sury's header and even predates newton 1.53 )
And then there is yours, and i think sascha also made one on his own, or am i mistaken?

[Dave Gravel]

I have one too, but where is the importance to know about it ?
All is the same if they is up-to-date with last newton version.

[Carli]

I'm using mode objfpc, maybe this has an other internal structure of a dynamic array.

The lines are my working lines.

The lines in the official header crashed on Win/linux with FPC

[Dave Gravel]

Maybe you need to disable FPU EX with FPC, no idea it's just a suggestion.

Code: Select all

initialization
  {$IFDEF CPU386}
    Set8087CW($133F);
  {$ENDIF}


[Stucuk]

And then there is yours, and i think sascha also made one on his own, or am i mistaken?

Sascha's are a continuation of mine. Tho he merged the 2 separate units into 1. His is the latest version of it.

All is the same if they is up-to-date with last newton version.

Thats why it would be great if there was just one, it would have more users checking for changes/mistakes.

[Carli]

Maybe we could set the current header into a repository for that everyone can add fixes or updates and I even can have the latest version of it as a subrepos of my project. (i'm using mercurial hg version control)

[JernejL]

Maybe you need to disable FPU EX with FPC, no idea it's just a suggestion.

Code: Select all

initialization
  {$IFDEF CPU386}
    Set8087CW($133F);
  {$ENDIF}


Actually, that won't work you need to use Math.SetExceptionMask in fpc! use this instead:

Code: Select all

{$ifdef fpc}
  Math.SetExceptionMask([exInvalidOp,exDenormalized,exZeroDivide,exOverflow,exUnderflow,exPrecision]);
{$else}
  Set8087CW($133F); // opengl delphi default!
{$endif}