[Animats]

How are the differences in the execution model handled? The defining quality of Python execution is that anything can monkey-patch anything at any time, using "setattr()". This means the obvious implementation has dictionaries for everything, and optimizers, as in PyPi have to work very, very hard.

So what did they do here? Is every object a dictionary to the Java runtime? That's easy enough to do, but no gain over CPython. Did they weaken the ability to modify objects so they could map to Java objects? That gets a performance win, as with Shed Skin, but a lot of Python code won't work. They're probably not doing all the stuff PyPi does; that's hard and took a decade of work by many people.