Walkingshaw's work on "variational programming" might be relevant

But this may not be what you had in mind, and I don't think this type of thinking is really even in scope for an AOT optimizer without profiling feedback.

As always, there's a chance that code duplication improves branch prediction (e.g. suppose that List.map<MyType>(f,x) was only ever invoked with a particular f).

I more strongly associate rustc with the specific binary `rustc`, compared to rustdoc which as far as I'm concerned could be a library with a different name. Maybe this explains why I care more about case for the former?

"cause"

Are there situations (due to associated types?) where polymorphic recursion that is not syntactically guaranteed to terminate, will in fact terminate for some choices of type argument?