I think that would be what I would try to do. If ...
# random
d
I think that would be what I would try to do. If you had lots of time to polish (or needed to optimize) if you can figure out when a sub-graph is complete, you can resolve the references for chunks at a time.