A partial clone of Minecraft made from scratch, without frameworks. Uses a concurrent work queue, caching data structures, and better rendering algorithms to achieve high FPS. Contains implementations of procedural generation, basic physics simulation, etc.
At least a basically realistic world generator is needed, so that render times can be evaluated under realistic conditions for example. This might involve some hills, some trees, some caves, some lakes, etc. It should be qualitatively similar to the desired final product, but need not "look real".
Currently, we refer to Chunks by their "chunk positions" and we iterate over Chunks indirectly by iterating over chunk positions. This is verbose and awkward at times. The detail of Chunks being identified by their positions should be hidden from code that deals with Chunks.
Occlusion culling would probably reduce render time by a lot (?) because although it presents a large upfront cost, it would really decrease the vertex count and the upfront cost would be handled by the prerenderer. Should be done after #7. Occlusion culling in the current "every other block is a dirt block" world would not help nearly as much as in a realistic world, where the ground is almost completely solid.
There is a lot of repeated information in VertexBufferObjects.
Positions are repeated 3 times each, once for each face sharing a given vertex.
Normal vectors are repeated 4 * 25 * 25 * 25 = 62,500 times each, once for each vertex of a given face, once for each block.
Colors are repeated many times each, once for each vertex on each face of that color, 4 * 6 * 25 * 25 * 25 = 375,000 times currently.
Indexed VBOs should be used. They would dramatically cut down VBO size, thereby shortening VBO build/send time and increasing frame rate/rendering distance.
This might be done by registering positions, normals, colors, etc. as Vectors in some centralized directory so that they can be reused. The Prerenderer would avoid repeating these Vectors in the vertex data, instead referring to them using indices. This could be implemented in two stages to separate the OpenGL-related bugs from the non-OpenGL related bugs.
The capacity of the FloatBuffers underlying VertexBufferObjects should not be hardcoded. Instead, some sort of quad limits and entity limits should be imposed.