Okay so I have an idea down to how I will be getting dynamically loaded classes. Effectively I am going to need a cache class entry in the constant pool. This entry will contain the class pointer. The code generator will check if it is zero, if it is then it will perform a class load. If that load is a success then it will write the true class pointer over the spot in the pool. It will call a JVM function or system call to load the class. Probably a system call to the supervisor. This code will never be hit for the supervisor since all classes are alreafy loaded. Load class will not be a true system call, it will be purely handled in software. But I feel it is the only way to get it actually working.
The system call will take a byte array and also one with a UTF pointer (with both the same data) and it will copy and use the class name there wheb it does the loading. There will also have to be invoking of static initializers but that has to wait until I can actually call into a task (it would be a recursive operation pretty much). Ironically tasks will be able to get the stack trace information from the kernel so I should actually have a flag to indicate the process something came from. So this means I will need a "context switch" API call that will basically set the context ID and then set a bunch of restored state. Effectively a setjmp sort of. But it would actually be a context enter as it will make a new stack frame with the new info. I think this will work out.
Okay, so I am refactoring the system call documentation so that way it is more well known how they work rather than just being basic fields.