Do people really use recursion when traversing trees in production software? Wouldn't it be faster to use your own stack instead of hacking the call stack? Or would it really make any difference?

I have an odd question. I implemented a quadtree that from what I can test, doesnt care about the size being by 1 or 2. When I process my map I have it split the lowest nodes by dividing say 257 by 2. Which is 128.5 on both sides. If A triangle crosses that boundry either split it or render it depending on whatcha like. If your working with ints I can understand why it wouldnt work, but most terrain needs to be more exact than that doesnt it?

Paul7, I break my terrain up at compile time, and then send the triangles to the program with the size of the boundaries. so say the size is - 256 to 256, I simply cut at 0(the middle) at run time and walk the nodes (recursion) and place everything together. With vectors in the optimize compiler settings, it takes me roughly 5 seconds to load a large map. I dont think any1 would need it faster, but if so you can precompile every node then write your file output for that.

I upon compile time walk though the nodes and each triangle is linked to a index number(part of the structure) that is in the most child node. I simply say, if im at the child node, then it must be in view and add the vertex's index numbers to a dynamic index buffer. And continue through the nodes. Once this index is built I render the terrain. It might be slower than others, but again Its so damn fast for me I dont complain.

Try this article.