Question

For a small AST parser i have a small discriminated union

type Numerical =
    | Int of int
    | Real of float

for the use in some other constructs like

type Vector = Numerical list
Vector [Int 42; Real 13.5]

When i have a method like this

let eval (e:Numerical) =
    match e with
    | Int n -> ... (* return int *)
    | Real r -> ... (* return float *)

I know that F# infers the int type and produces an error with the Real pattern in the second line, so I want to know, which code-design would be the best one to be able to handle such “generic” types and return their appropriate values with the given types.

EDIT

I have here an either or situation which leads to functions like

let getInt = function
| Int n -> n
| Real _ -> failwith "Given value doesn't represent an int"

let getReal = function
| Real n -> n
| Int _ -> failwith "Given value doesn't represent a real number"

however i would like a method which encapsulates both cases and “automatically chooses the correct one”.

This whole effort should lead to the possibility to be able to really run computations with the “boxed” values like Int 42 and Real 13. with primitive data types but to able to return the appropriate wrapper. If i want to add Real 1. and Real 1.5 I want to extract 1.0 + 1.5 = 2.5 and then go on with Real 2.5, however I do not want to handle everything as floats so that I have the distinction between ints and floats.

Answer 1

You could cast the results to obj:

let eval (e:Numerical) =
    match e with
    | Int n -> n :> obj
    | Real r -> r :> obj

But that’s probably not what you want.

Another option would be implement your own operations on Numerical:

let (+) a b =
    match (a,b) with
    | (Int an, Int bn) -> Int (an + bn)
    | (Real ar, Real br) -> Real (ar + br)
    | _ -> failwith "Can't add Int and Real"

You can’t have a function that has different compile-time return type based on some run-time value.