Comment by haberman
5 years ago
I haven't implemented the fallback yet, but my thought was to keep most of the structure the same (dispatch function tail calls to the function for an individual operation) but then have the individual operation just return instead of tail calling back to dispatch.
I think this would be portable while still avoiding some of the performance problems associated with a traditional switch (like the bounds check).
I'm curious how this approach would fair for microcontrollers, well cortex m0-m4's actually. They're arm32 bit and may not have enough useful call registers. Still parsing protobuf's or interpreting WASM faster would result directly in better battery performance, etc.