Maybe redesigning everything wasn't such a good idea...
Back from my trip to Thredbo (the snowies in Australia for you non-aussies). Obviously there is no snow (or very little) at this time of the year. However, we did get in some good walks. Did a 13km one on Saturday and a 10km one on Sunday. The scenery was gorgeous. Makes me want to spend a week camping up there and exploring the terrain.
So okay, been sitting down and trying to nut my way through this surface subdivision algo change. To be honest, I really should just leave it the way it is right now and move onto something more interesting... like collision detection and game objects and such. So I've basically decided to "can" the coons patches idea for the moment so as to keep the project flowing. It means I won't be quite as flexible with the size of meshes I can make, but I should be able to make something decent with what I have anyway. The upside is that should I decide to include it at a later point in the engine or for a sequel to my first title, the new algo should plug right into my engine with a minimum of fuss or modification.
So as for what I have done, today I modified the loose octtree code to use half-widths and centre-points for the nodes instead of max and min points. This allows me to do some quick comparisons and cut down the number of recursive calls in an insert/remove/update/etc. Should be a bit faster. Next I'm gonna redo the vertex buffer stuff so that I'm using large vb's for better batching instead of one vertex buffer per object. Again, should yield some decent performance boosts for larger (more populated) levels.