To do this, I'm working backwards from the compiler back-end first to the lexer/parser last. This allows me to retain all of the language's features at all times, without worrying about anything falling through the cracks along the way - which would be a serious concern if I rewrote everything from the front-end out.
Tonight I finished a nice milestone in the process: bytecode generation is now handled (mostly) by an Epoch program instead of a C++ program. This is probably the simplest piece of code to move over to Epoch because all it really amounts to is a bunch of byte sequences streamed into a buffer and then spat back out.
A handful of instructions aren't supported yet, but I'll be working on that over the weekend, and ideally sometime soon I'll have the entire compiler test suite passing with the new bytecode emitter.
Once that's in place, it's time to work on the code generator itself, which traverses the marked-up semantic IR of the language and generates bytecode streams from that. That's a much bigger piece of work and will take a lot longer, but it should be entertaining either way.
