I've decided to cool down today's development for my current project to make some careful design decisions. I'm developing a tree generation library that will probably be released under the LGPL.

I already have a feature where you may specify the UV-mapping basis (i.e. virtual origin, x-axis and y-axis vectors) for the stem geometry and leafy geometry both discretely within individual property blocks, which I simply call the "virtual texturing" options i.e. entire trees can use single textures. Do you think this will be adequate, or should I implement the ability to seperate indexing onto multiple index buffers to support having multiple materials? The biggest problem with this feature is how it may increase the complexity/effectiveness of instancing for forest rendering... but nevertheless my users may still choose to go with a single material, one index buffer and a unified texture without any additional complexity.

How important do you think this feature might be? I'm also uncomfortable with the complexity it adds to the internal libray, but it's not that bad. Actually it's just a matter of adding a few more attributes to the property system and handling an array of index buffers for every LOD (just one more ' * '), rather than just a single buffer.

_________________
What most people don't understand is what I like to explain as a thing that I understand.

Tree generation sounds like a lot of fun, and i gave it a lot of thought a couple of years ago aswell.
I think you'd need several textures, or maybe a texture atlas, to render nice trees, different bark, moss etc. You'd also need normal or parallax maps (or go for tesselation).

_________________
Long pork is people!

wzl's burrow

Nice ideas. I can go for texture blending and normal mapping, but I don't want to use parallax mapping for bark. I also want to avoid the need for changing the texturing state more than once; instancing needs to work well. Texture atlasing is already implemented on the software side, but I just didn't know the formal term for it.

I plan on experimenting with special shaders for highly-realistic tree rendering, involving both texture mapping and lighting. I think if I develop a custom solution for bark-texture rendering, it will blow a lot of hats off. It's already setup for skinning and I have a paper to read about practical/low-cost tree targeted wind simulation.

Well, for multiple materials I guess I'd need another vertex attribute... that sounds like no fun. If I went for multiple index buffers I could save a lot of data bloat, prevent a lot of shader complexity, and it can be optional, but instancing will be half as effective. I guess when I have more time, I can add this feature as another option.

_________________
What most people don't understand is what I like to explain as a thing that I understand.