Function definition inside let-block impacting performance?

Yup! My proposal here is simply to recognize that there are no syntactical assignments to fib inside nor after its definition(s) nor inside the definition of any other function, and therefore it doesn’t need to capture itself. Within its functor’s definition it can simply refer directly to itself. It should work for closures that capture variables from their environment too.

This is actually a different clever trick from what I suggested; what you wrote causes the function to be declared as a singleton object, which has interesting differences in how it captures variables from its environment, but also it doesn’t work inside function definitions because those often need to be closures (non-singleton). For example:

julia> genfib(a) = @stable fib(n) = n ≤ a ? n : fib(n-1)+fib(n-2)
ERROR: syntax: Global method definition around REPL[7]:9 needs to be placed at the top level, or use "eval".

My proposal here should work with recursive closures defined inside functions, not just locally-scoped within let statements.

However, I don’t currently see a way to make my proposal work with two mutually recursive closures. It might be the case that efficient locally-declared mutually recursive functions can only exist in let statements using something similar to your trick, and not be defined inside a function?