We ended up sticking with AOT (for now, anyway), because it seems easier to
manage in the codebase. The alternative is to use data structures that can
be eval'd, like you would use in the body of a macro. I like how that
clearly separates the code that runs on the local machine from that which
All functions returned by a given arity of partial or comp are of the same
Java class under the hood, whereas defining an anonymous function always
generates a new class, and I've heard that fewer classes can result in
better performance from a JIT perspective. But I'm just speculating and
haven't