Then I'm not sure why that would refute the intermediate goal to "enable JIT codegen without sacrificing too much performance" as stated in my initial comment, since the proposed copy-and-patch JIT compiler can't make the impact by itself.