bounding box center

[pca006132]

If I understand it correctly, the collider expects the leaf nodes to be ordered by the morton code of the box center, which can be different from the center of a triangle.

manifold/src/manifold/src/sort.cpp

Lines 84 to 113 in 1514c15

	struct FaceMortonBox {
	VecView<const Halfedge> halfedge;
	VecView<const glm::vec3> vertPos;
	const Box bBox;
	void operator()(thrust::tuple<uint32_t&, Box&, int> inout) {
	uint32_t& mortonCode = thrust::get<0>(inout);
	Box& faceBox = thrust::get<1>(inout);
	int face = thrust::get<2>(inout);
	// Removed tris are marked by all halfedges having pairedHalfedge = -1, and
	// this will sort them to the end (the Morton code only uses the first 30 of
	// 32 bits).
	if (halfedge[3 * face].pairedHalfedge < 0) {
	mortonCode = kNoCode;
	return;
	}
	glm::vec3 center(0.0f);
	for (const int i : {0, 1, 2}) {
	const glm::vec3 pos = vertPos[halfedge[3 * face + i].startVert];
	center += pos;
	faceBox.Union(pos);
	}
	center /= 3;
	mortonCode = MortonCode(center, bBox);
	}
	};

@elalish am I misunderstanding something here or is this a bug?

[elalish]

It actually doesn't matter - the bounding boxes are what actually get queried against, and so long as they are unioned properly up the tree, the result will be correct. The Morton code is about sorting such that those unioned bounding boxes tend to stay small, thus keeping the BVH efficient. Using centers does a decent job, and there's way more variance introduced by the z-order curve, so in practice it doesn't matter much.

[pca006132]

oh ok, because I am trying to implement something related to finding minimum gap, so I have to use the BVH stuff and was worried that using an incorrect center will cause the BVH to break.