It shouldn’t be too much harder than calling lua (hopefully, haven’t tried this yet!!).
The difficultly is to sort the graph depth first (reducing wirebufs) while keeping the ordering. I believe (but could be wrong as I’ve only glanced at the code) scalacollider also gets this wrong.
Just to clarify, I’m not knocking this idea, just trying to point out the moment you start applying optimisations, things get very complex, to the point, where a standard implementation should be provided.