Threaded execution has been used as a higher performance alternative to a byte-code interpreter, by utilizing hardware dispatch to replace software interpreter dispatch.

Traditional JIT code is compiled from byte-code to native code for the current machine, with an even higher performance result. Unfortunately, when debugging is required for a method, most JIT-based interpreters discard the JIT code and revert to the byte-code interpreter. Additionally, switching between interpreted code and native code requires clever trampolines to bring the models into alignment.

We describe a technique that maintains the threaded code and native code as parallel implementations of the program. This provides seamless transitions between the implementations and supports full debugging, while providing near full native execution performance.