I've written a fairly robust rendering engine using OpenGL ES 1.1 (fixed-function.) I've been looking into revamping the engine to use OpenGL ES 2.0, which necessitates that I use shaders. I've been absorbing information all day long and still have some questions.

Firstly, lighting. The fixed-function pipeline is guaranteed to have at least 8 lights available. My current engine finds lights that are "close" to the primitives being drawn and enables them; I don't know how many lights are going to be enabled until I draw a given model. Nothing is dynamically allocated in GLSL, so I have to define in a shader some number of lights to be used, right? So if I want to stick with 8, should I write my general purpose shader to have 8 lights and then use uniforms to tell it how many / which lights to use?

Which brings me to another question: should I be concerned with the amount of data I'm allocating in a shader? Recent video cards have hundreds of "stream processors." If I've got a fragment shader being used on some number of fragments in a given triangle, I assume they must each have their own stack to work on. Are read-only variables copied here, or read when needed?

My initial goal is to rework my code so that it is virtually identical to the current implementation. What I have in mind is to create my own matrix stack so that I can implement something along the lines of push/popMatrix and apply all my translations, rotations, and scales to this matrix, then provide the matrix to the vertex shader so that it can make very quick vertex translations. Is this approach sound?

Edit: My original intention was to ask if there was a tutorial that would explain the bare minimum necessary to jump from fixed-function to using shaders.

Thanks!