这个 Haskell 函数的 Java 等价物是什么？

Question

Philip Wadler's paper "Monads for functional programming" has an example of a function eval which performs division written in Haskell.

Here it is as adapted in "Programming in Haskell" by Graham Hutton:

data Expr = Val Int | Div Expr Expr

eval :: Expr -> Int
eval (Val n) = n
eval (Div x y) = eval x `div` eval y

My Java equivalent of this is:

abstract class IntegerExpression {
    abstract Integer evaluate();
}

class Value extends IntegerExpression {

    Integer value;

    public Value(Integer x) {
       value = x;
    }

    public  Integer evaluate() {
        return value;
    }
}

class DivisionExpression extends IntegerExpression {

    IntegerExpression x, y;

    public DivisionExpression(IntegerExpression a, IntegerExpression b) {
        x = a;
        y = b;
    }

    public Integer evaluate() {        
        return x.evaluate() / y.evaluate();
    }

}

   public class DivisionExample {

    public static void main(String[] args) {
        IntegerExpression two = new Value(2);
        IntegerExpression twenty = new DivisionExpression (new Value(100), new Value(5));
        IntegerExpression ten = new DivisionExpression(twenty, new Value(2));
        IntegerExpression five = new DivisionExpression(new Value(10), two);
        IntegerExpression expr = new DivisionExpression(ten, five);

        System.out.println(expr.evaluate());
    }
}

This seems fine but how do I develop this code so that I can demonstrate a Try monad (to catch division by zero) in Java?

Answer 1

Edit: a workaround to failures in this kind of cases is the use of the Maybe Monad , and his cousin in Java is the Optional class, where Option.of would be return and flatMap would be bind . On the other hand in Java and others OO languages there is a common pattern to use in this kind of cases called composite, basically your data type Expr will be an interface or abstract class, and the type constructors will be the leaves: So, with all of that in mind, a simple example working would be:

In haskell:

data Expr = Val Int | Div Expr Expr

eval :: Expr -> Maybe Int
eval (Val n) = Just n
eval (Div x y) = do
                  v1 <- eval x
                  v2 <- eval y
                  if v2 == 0 
                  then Nothing
                  else return (div v1 v2)
n1 = Val 8
n2 = Val 4
n3 = Val 0 
d1 = Div n1 n2
d2 = Div d1 d1
d3 = Div d2 n3

main = do
  putStrLn $ show (eval d2)
  putStrLn $ show (eval d3)

Equivalent approach in Java:

import java.util.Optional;

public interface Expr {

    public Optional<Integer> eval();

}

Then the leaves implementing Expr:

import java.util.Optional;

public class Val implements Expr{

    Optional<Integer> value;

    public Val(int value) {
        this.value = Optional.of(value);
    }

    @Override
    public Optional<Integer> eval() {
        return value;
    }
}

Then the recursive case:

import java.util.Optional;

public class Div implements Expr {

    Expr expr1;
    Expr expr2;

    public Div(Expr expr1, Expr expr2) {
        this.expr1 = expr1;
        this.expr2 = expr2;
    }

    @Override
    public Optional<Integer> eval() {
        return expr1.eval().flatMap(v1 ->
                expr2.eval().flatMap(v2 ->
                    (v2 == 0) ? Optional.empty() : Optional.of(v1 / v2)
                )
               );
    }

    public static void main(String[] args) {
        Expr iv1 = new Val(6);
        Expr iv2 = new Val(3);
        Expr iv3 = new Val(2);
        Expr iv4 = new Val(0);
        Expr div1 = new Div(iv1, iv2);
        Expr div2 = new Div(div1, iv3);
        Expr div3 = new Div(div2, iv4);

        System.out.println(div2.eval());
        System.out.println(div3.eval());

    }
}

The main function output will be:

Optional[1]
Optional.empty

Answer 2

Other answers have covered a more idiomatic way of implementing this in Java, and described how to use Optional to handle errors. But here I'd like to give the direct equivalent of Haskell pattern matching in Java, with the visitor pattern :

public class ExprTest {
    public static void main(String[] arguments) {

        // expr :: Expr
        // expr = Div
        //   (Div
        //     (Div (Val 100) (Val 5))
        //     (Val 2))
        //   (Div (Val 10) (Val 2))
        Expr two = new Val(2);
        Expr twenty = new Div(new Val(100), new Val(5));
        Expr ten = new Div(twenty, new Val(2));
        Expr five = new Div(new Val(10), two);
        Expr expr = new Div(ten, five);

        // eval :: Expr -> Int
        // eval expr = case expr of
        ExprVisitor<Integer> eval = new ExprVisitor<Integer>() {

            // Val value -> value
            public Integer visit(Val val) {
                return val.value;
            }

            // Div left right -> eval left `div` eval right
            public Integer visit(Div div) {
                return div.left.match(this) / div.right.match(this);
            }

        };

        // main = print (eval expr)
        System.out.println(expr.match(eval));
    }
}

// data Expr
abstract class Expr {
    abstract <T> T match(ExprVisitor<T> visitor);
}

// = Val Int
class Val extends Expr {

    public final int value;

    public Val(int value) {
        this.value = value;
    }

    <T> T match(ExprVisitor<T> visitor) {
        return visitor.visit(this);
    }

}

// | Div Expr Expr
class Div extends Expr {

    public final Expr left, right;

    public Div(Expr left, Expr right) {
        this.left = left;
        this.right = right;
    }

    <T> T match(ExprVisitor<T> visitor) {
        return visitor.visit(this);
    }

}

abstract class ExprVisitor<T> {
    abstract T visit(Val val);
    abstract T visit(Div div);
}

In the land of functional programming, this is called Böhm–Berarducci encoding—sometimes referred to as Church encoding, although they're different things. This is a fancy-sounding way of saying “representing data types and pattern matching with functions”. You can of course use this encoding of matching in Haskell:

match
  :: (Int -> t)           -- visit(Val)
  -> (Expr -> Expr -> t)  -- visit(Div)
  -> Expr
  -> t
match val div expr = case expr of
  Val x -> val x
  Div left right -> div left right

eval :: Expr -> Int
eval = match id (\ left right -> eval left `div` eval right)

Since eval is recursive, you can also write it using the fixed point combinator fix —and then the use of this in the ExprVisitor in the Java version may become more clear: it's how you make eval recursive!

import Data.Function (fix)

eval :: Expr -> Int
eval = fix $ \ this -> match
  (\ value -> value)
  (\ left right -> this left `div` this right)

And here's the other half of the encoding: we can do away with the data type altogether and just use functions:

{-# LANGUAGE RankNTypes #-}

newtype Expr = Expr
  { visit
    :: forall a.
       (Int -> a)     -- Val
    -> (a -> a -> a)  -- Div
    -> a }

valE :: Int -> Expr
valE x = Expr $ \ v _d -> v x

divE :: Expr -> Expr -> Expr
divE left right = Expr $ \ v d
  -> d (visit left v d) (visit right v d)

eval :: Expr -> Int
eval expr = visit expr
  (\ val -> val)
  (\ left right -> left `div` right)

eval (divE
  (divE (divE (valE 100) (valE 5)) (valE 2))
  (divE (valE 10) (valE 2)))
  == 2

And that implementation of eval can of course be written as just this:

eval = visit expr id div

Answer 3

You could evaluate both x and y before the division:

Integer xE = x.evaluate(), yE = y.evaluate();

And then see if yE is equal to 0 :

if(yE == 0){
    // your logic here if it is a division by 0
}

Which would yield you with the following function:

public Integer evaluate() {        
    Integer xE = x.evaluate(), yE = y.evaluate();

    if(yE == 0){
        // your logic here if it is a division by 0
    }
    return xE / yE;  
}