Question

For example if I defined F x y as a currying function and in an environment the partial function (F x) is used a lot, should i always let binding it to a named function in terms of performance? (say, compared to cached a property of an object, as it is often unnecessary to cache the property such as Array.Length)

 let Gofrom (board: int[][]) (color: int) (reelID, reelPosition) (direction: Direction) =
    let rec walkfrom x =
        match direction.Target x with
        | (a, _ | _, a) when a = -1 || a = board.Length -> x
        | a, b when board.[a].[b]= color -> walkfrom (a, b)
        | _ -> x
    walkfrom (reelID, reelPosition)

v.s.

 let rec Gofrom (board: int[][]) (color: int) (reelID, reelPosition) (direction: Direction) =
        match direction.Target (reelID, reelPosition) with
        | (a, _ | _, a) when a = -1 || a = board.Length -> (reelID, reelPosition)
        | a, b when board.[a].[b]= color -> Gofrom board color (a, b) direction
        | _ -> (reelID, reelPosition)

Answer 1

There should be no difference in performance, but I would definitely use the first form since it makes clear which args do not change between invocations.

Incidentally, this isn’t really about currying as you’re refactoring code into a separate function (closure) to avoid passing args around.