Lambda Calculus via C# (13) Encoding Church Pairs (2-Tuples) and Generic Church Booleans

[LINQ via C# series]

[Lambda Calculus via C# series]

Latest version: https://weblogs.asp.net/dixin/lambda-calculus-via-csharp-4-tuple-and-signed-numeral

Church pair is the Church encoding of the pair type, aka 2-tuple. Unlike the Tuple<T1, T2> class in .NET, in lambda calculus Church pair will be represented by lambda expression. To avoid 2 naming systems, here in all the code, Church pair will be called tuple.

Church pair (2-tuple)

A Church pair can be constructed with 2 values x y:

CreateTuple := λx.λy.λf.f x y

And it return a tuple - another lambda expression (λf.f x y). So tuple is a higher order function that takes a function and apply it with x and y.

Tuple := λf.f x y

Notice:

  • tuple is a closure of x and y
  • f is supposed to be in the format of λx.λy.E

So, to get the first item x, a f like λx.λy.x can be applied to a tuple.

Item1 := λt.t (λx.λy.x)

Item1 takes a tuple as parameter, applies it with a (λx.λy.x), and returns the first item x. This is how Item1 works:

  Item1 (CreateTuple x y)
≡ Item1 (λf.f x y)
≡ (λt.t (λx.λy.x)) (λf.f x y)
≡ (λf.f x y) (λx.λy.x)
≡ (λx.λy.x) x y
≡ (λy.x) y
≡ x

So to get the second item y, a tuple can be applied with a f of λx.λy.y:

Item2 := λt.t (λx.λy.y)

And just like Item1:

  Item2 (CreateTuple x y)
≡ Item2 (λf.f x y)
≡ (λt.t (λx.λy.y)) (λf.f x y)
≡ (λf.f x y) (λx.λy.y)
≡ (λx.λy.y) x y
≡ (λy.y) y
≡ y

Based on above definitions, here is  the C# implementation:

// Tuple = f => f(item1)(item1)
public delegate object Tuple<out T1, out T2>(Func<T1, Func<T2, object>> f);
// Tuple is an alias of Func<Func<T1, Func<T2, object>>, object>

public static class ChurchTuple
{
    // CreateTuple = item1 => item2 => f => f(item1)(item2)
    public static Func<T2, Tuple<T1, T2>> Create<T1, T2>
        (T1 item1) => item2 => f => f(item1)(item2);

    // Item1 => tuple => tuple(x => y => x)
    public static T1 Item1<T1, T2>
        (this Tuple<T1, T2> tuple) => (T1)tuple(x => y => x);

    // Item2 => tuple => tuple(x => y => y)
    public static T2 Item2<T1, T2>
        (this Tuple<T1, T2> tuple) => (T2)tuple(x => y => y);
}

Tuple’s Item1 is of type T1, Item2 is of type T2. And, f is λx.λy.E, so its type is Func<T1, Func<T2, object>>. Again, just like the object in Church Boolean Func<object, Func<object, object>>, object here does not mean System.Object is introduced. It just mean λx.λy.E can return any type. For example:

  • in function Item1, f is λx.λy.x or x => y => x, so f returns a T1
  • in function Item2, f is λx.λy.y or x => y => y, so f returns a T2

Generic Church Booleans

If observing above definition:

Item1 := λt.t (λx.λy.x)
Item2 := λt.t (λx.λy.y)

In Item1 f is actually True, and in Item2 f becomes False. So above definition can be simplified to:

Item1 := λt.t True
Item2 := λt.t False

In C# more work need to be done for this substitution. As fore mentioned, f is Func<T1, Func<T2, object>> but currently implemented Church Boolean is Func<object, Func<object, object>>. So a more specific Church Boolean is needed.

// Curried from: object Boolean(TTrue @true, TFalse @TFalse)
public delegate Func<TFalse, object> Boolean<in TTrue, in TFalse>(TTrue @true);
// Boolean is alias of Func<TTrue, Func<TFalse, object>>

public static partial class ChurchBoolean
{
    // True = @true => @false => @true
    public static Func<TFalse, object> True<TTrue, TFalse>
        (TTrue @true) => @false => @true;

    // False = @true => @false => @false
    public static Func<TFalse, object> False<TTrue, TFalse>
        (TTrue @true) => @false => @false;
}

With this generic version of Church Booleans, above Church tuple can be re-implemented:

public delegate object Tuple<out T1, out T2>(Boolean<T1, T2> f);

public static partial class ChurchTuple
{
    // CreateTuple = item1 => item2 => f => f(item1)(item2)
    public static Func<T2, Tuple<T1, T2>> Create<T1, T2>
        (T1 item1) => item2 => f => f(item1)(item2);

    // Item1 = tuple => tuple(x => y => x)
    public static T1 Item1<T1, T2>
        (this Tuple<T1, T2> tuple) => (T1)tuple(ChurchBoolean.True<T1, T2>);

    // Item2 = tuple => tuple(x => y => y)
    public static T2 Item2<T1, T2>
        (this Tuple<T1, T2> tuple) => (T2)tuple(ChurchBoolean.False<T1, T2>);
}

Back to Church Boolean - why not using generic Church Booleans from the beginning?

If the Boolean logic is implemented with this generic version of Church Booleans, then:

public static partial class ChurchBoolean
{
    // And = a => b => a(b)(False)
    public static Boolean<TTrue, TFalse> And<TTrue, TFalse>
        (this Boolean<Boolean<TTrue, TFalse>, Boolean<TTrue, TFalse>> a, Boolean<TTrue, TFalse> b) => 
            (Boolean<TTrue, TFalse>)a(b)(False<TTrue, TFalse>);

    // Or = a => b => a(True)(b)
    public static Boolean<TTrue, TFalse> Or<TTrue, TFalse>
        (this Boolean<Boolean<TTrue, TFalse>, Boolean<TTrue, TFalse>> a, Boolean<TTrue, TFalse> b) => 
            (Boolean<TTrue, TFalse>)a(True<TTrue, TFalse>)(b);

    // Not = boolean => boolean(False)(True)
    public static Boolean<TTrue, TFalse> Not<TTrue, TFalse>
        (this Boolean<Boolean<TTrue, TFalse>, Boolean<TTrue, TFalse>> boolean) => 
            (Boolean<TTrue, TFalse>)boolean(False<TTrue, TFalse>)(True<TTrue, TFalse>);

    // Xor = a => b => a(b(False)(True))(b(True)(False))
    public static Boolean<TTrue, TFalse> Xor<TTrue, TFalse>
        (this Boolean<Boolean<TTrue, TFalse>, Boolean<TTrue, TFalse>> a, Boolean<Boolean<TTrue, TFalse>, Boolean<TTrue, TFalse>> b) => 
            (Boolean<TTrue, TFalse>)a((Boolean<TTrue, TFalse>)b(False<TTrue, TFalse>)(True<TTrue, TFalse>))((Boolean<TTrue, TFalse>)b(True<TTrue, TFalse>)(False<TTrue, TFalse>));
}

The type parameter becomes too noisy. It is difficult to read or use these functions.

Currying and type inference

The part of currying mentioned currying may cause some noise for type inference in C#. Here is an example:

Swap = λt.CreateTuple (Item2 t) (Item1 t)

C# logic is simple, but the type information has to be given so it is noisy:

// Swap = tuple => Create(tuple.Item2())(tuple.Item1())
public static Tuple<T2, T1> Swap<T1, T2>
    (this Tuple<T1, T2> tuple) => Create<T2, T1>(tuple.Item2())(tuple.Item1());

When invoking the curried Create function, the type arguments cannot be omitted. This is signature of Create:

Func<T2, Tuple<T1, T2>> Create<T1, T2>(T1 item1)

After currying, T2’s appearances are all relocated to Create’s returned type. So during the 2 applications of Create(item1)(item2), C# compiler does not even know how to compile first application Create(item1). It cannot infer what return type is wanted. The application code will always end up as:

ChurchTuple.Create<int, string>(1)("a");

So, only for convenience of C# coding and less noise for readability, this uncurried helper method can be created:

public static Tuple<T1, T2> _Create<T1, T2>
    (T1 item1, T2 item2) => Create<T1, T2>(item1)(item2);

Now T2 is relocated back to parameter, so type arguments are not mandatory:

ChurchTuple._Create(1, "a");

Much less noise. _Create is also tagged with underscore since its uncurrying is for adapting C# type inference feature.

2 Comments

Add a Comment

As it will appear on the website

Not displayed

Your website