This would also be great for compiling to cranelift (especially since I modeled the block system off of their ir, a bit 🙃 ), but I'm not sure if there's any way to preserve the block information all the way to the function object that you call .__jit__() on

Best I can tell, there's little performance improvement and this may be very, very slightly slower (I get stuff like norm 1.00 ± 0.01 faster than opt), but I think there's a lot more room for optimization with this compilation model than what we had before.

It'd be nice for a JIT where we just remove the interpreter loop and directly call the functions imo.

it looks like arrayvec exists in crates.io but boxvec doesn't. maybe it would be useful as an independent crate