Call Stack problems

[Yenza]

Using three.min.js rev 73

Method used:
var roomBSP = new ThreeBSP(_RoomShape); // HUGE SHAPE

Info gathered from a TryCatch:
"number : -2146828260"
"Error: Out of stack space
at ThreeBSP.Vertex.prototype.dot (/ThreeCSG.js)
at ThreeBSP.Polygon.prototype.classifyVertex (/ThreeCSG.js)
at ThreeBSP.Polygon.prototype.classifySide (/ThreeCSG.js)
at ThreeBSP.Polygon.prototype.splitPolygon (/ThreeCSG.js)
at ThreeBSP.Node (/ThreeCSG.js)
at ThreeBSP.Node (/ThreeCSG.js)
at ThreeBSP.Node (/ThreeCSG.js)
at ThreeBSP.Node (/ThreeCSG.js)
at ThreeBSP.Node (/ThreeCSG.js)
at ThreeBSP.Node (/ThreeCSG.js)"

Using the OUTDATED csg-wrapper from kraag22/csg-wrapper it doesn't happen but his wrapper is way outdated and needs an update.

Might this occur when using high values/numbers?

Another example:
var intersect = shape1.intersect(shape2); // Small shapes

Info gathered from a TryCatch:
"number : -2146828260"
-:-

[Wilt]

I have an updated fork here

[Yenza]

Your fork is exactly the same except missing the face.vertexNormal fix

[Wilt]

Owh, sorry. I thought that there were more fixes in my fork...

[Wilt]

I provided the vertex normal fix...
But I accidently pointed to the master, and that fix was in this patch1 branch.

[Yenza]

Your branch is fully up-to-date with Chandlers and you both provide exact same ThreeCSG.js files.
I've just compared with a SVN program. No changes.

[Wilt]

@Yenza Yes, you are right. He recently merged my patch1...

[chandlerprall]

Maximum call stack happens with complex geometry for two reasons:

1. ThreeCSG doesn't use any heuristic for picking a good divider
2. Even if a great divider is selected it is still possible to reach maximum recursion. The proper fix would be to change the recursion into one function that keeps a local stack queue and while() loops over it.

Solving either of these would gain large speed improvements, especially for complex geometries. There are many white papers / thesis on CSG which provide ideas for selecting a good divider. Unfortunately I won't have time for a while to actively look into implementing.

[Yenza]

@Wilt Do you think you could try to fix this issue?

@chandlerprall But it isn't that very complex! It's two round shapes! It's two identical "almost" circle-wise shapes on the same height. The thing is that if I move any of them like 1 unit in any direction.. then it works! I have an image of the red area when using the intersect method that causes an error:
http://i.imgur.com/NW0AOUU.png

Also this problem doesn't seem to be occurring when you use the old csg wrapper but that might just be because it's heavily outdated.

[Yenza]

So this is what I did to fix this for now... I dunno why it works but it does. The "var test = new ThreeBSP(shape)" doesn't though x.x

var intersection;
try {
    // If it doesn't work the first time...
    intersection= shapeBSP.intersect(shapeBSP_2);
}
catch (e) {
    // Then try again..
    var newShape = shapeBSP_2.toMesh(_Material);
    newShape.position.x += 1;
    newShape = new ThreeBSP(newShape );
    intersection= shapeBSP.intersect(newShape);
    // console.warn("Works!");
}

[jgphilpott]

FYI this new CSG repo using Octree data structure overcomes a lot of the performance issues with ThreeCSG and other CSG tools. See this SO thread for up to date information.