Comment by derefr
1 year ago
I don't get the problem. The generated code for your example would return the same function pointer + evaluated args list to the wrapping trampoline (i.e. the code that contains the loop), where that loop expects each thing it invokes to return a sum type NextInvoke(func, args) | Return(value).
And that's not a special case for passed-in function pointers; that's what a trampoline always does. Trampolines expect static-invoked tail-calls to be codegen'ed into function-pointer references too.
A trampoline is less efficient theoretically because it implies that you must check a condition and that you can't make an unconditional tail-call. It's also not quite equivalent since it's a non-local compilation technique requiring the caller to do something differently.
It's completely different than the story with tail recursion which can be essentially reduced to syntactic sugar over a normal loop (just change the parameters to be mutable variables).