Hi Alex, thank you for reaching out here on the mailing list. Your paper demonstrated a good understanding of the TinkerPop code base. The development of a Gremlin VM in another language may help us prove out what certain aspects of TinkerPop4 might look like. It would be interesting to learn from the problems you face as you continue to implement it. Of critical point in my mind is that a Gremlin VM in any language should be able to accept Gremlin bytecode. In that way, it immediately makes it possible for the dozen or so existing Gremlin Language Variants compatible with it. Then, if that was all working, my first fun test would be to subgraph() from one VM to another where the client-side subgraph would end up being completely traversible in the host language VM (i.e. in C++ subgraph() a JVM-based graph then traverse that subgraph locally in C++ with Gremlin).
Please keep us informed of the work you're doing as it progresses.