nested mel uncurry experiment

[anmonteiro]

putting this up to show how it can't really work at the melange level because nested mel.uncurry callbacks aren't arguments that we pass, but rather passed from the JS side.

[anmonteiro]

this function shows why we can't optimize nested [@mel.uncurry] implicitly. Here's the explanation:

• whenever we apply arguments to an external function, we walk both the external argument list (as declared by the user), and the params passed to the function
• nested [@mel.uncurry] are, however, and by definition, not parameters that we pass, but rather, arguments to some callback. Take the simplest example:

external foo : (((unit -> unit)[@mel.uncurry]) -> unit) -> unit = "foo"

let () = foo (fun f -> f ())

You can see how, in the function application, f is passed to us by the callback (the one we actually pass as a parameter). We don't have that information later on to apply f in the uncurried way.

Take, for example, a pathological case that could happen:

external foo : (((unit -> unit)[@mel.uncurry]) -> unit) -> unit = "foo"

let () = 
  let the_function = ref (fun () -> ()) in
  foo (fun f -> the_function := f);
  !the_function ()

[anmonteiro]

we merged #875 which links here to the explanation. Closing this.