Add Validation Applicative

[Andrae Muys]

I have sketched a possible approach to implementing a Validation Applicative for Fugue.

At the moment it is written using JDK8 lambdas; however, I don't believe this is required for correctness, just legibility.

I am interested in comments and critique.

https://bitbucket.org/snippets/atlassian/LAoK4

[Anund McKague]

Base classes:

class Validation<EE,A> {
    public abstract <B> B fold(Function<EE, B> onError, Function<A, B> onSuccess);
    public abstract <B> Validation<EE, B> map(Function<A, B> f);
    public abstract <B> Validation<EE, B> ap(SemiGroup<EE> s, Validation<EE, Function<A, B>> ff);
    protected abstract <B> B match(Function<Failure<EE,A>, B> onError, Function<Success<EE,A>, B> onSuccess);
}

// Builders
static <EE, A> Validation<EE, A> succeed (A a){
   return new Success<>(a);
}

static <EE, A> Validation<EE, A> fail (EE e){
    return new Failure<>(e);
}

Class Success extends Validation<EE,A>
Class Failure extends Validation<EE,A>

SemiGroup<A> {
   A add(A a1, A a2);
}

SemiGroupApp<A,B> extends SemiGroup<A> {
   A point(B b); 
}

[Anund McKague]

// lifts(?)
public static <EE, A, R> Validation<EE, R> lift1(
       Validation<EE, A> fa,
       Function<A,R> f) {
    return fa.map(f);
}
public static <EE, A, B, R> Validation<EE, R> lift2(
        SemiGroup<EE> s,
        Validation<EE, A> fa,
        Validation<EE, B> fb,
        Function<A,Function<B,R>> f) {
    return fa.match(
            failure -> failure.accumulate(s, fb),
            success -> lift1(fb, f.apply(success.getValue())));
}

// accumulate
protected <R> Validation<EE, R> accumulate(SemiGroup<EE> s, Validation<EE, ?> vb) {
    return accLoop(s, v -> v.failure(), vb);
}

protected <R> Validation<EE, R> accumulate(SemiGroup<EE> s, Validation<EE, ?> vb, Validation<EE, ?> vc, Validation<EE, ?> vd) {
    return accLoop(s, v -> v.accumulate(s, vc, vd), vb);
}

protected <R> Validation<EE, R> accLoop(SemiGroup<EE> s, Function<Failure<EE, ?>, Validation<EE, R>> acc, Validation<EE, ?> v) {
    return v.fold(ee -> acc.apply(failure(s.add(e, ee))), x -> acc.apply(this));
}

[Anund McKague]

Possible additional behaviours:

• leftmap(Function<A,B>)/bimap(Function<E,C>, Function<A,B)
• extend maybe interface
• success biased foreach
• <* or *> chooseleft, chooseright given two validations ??
• a swap Validation<A,B> to Validation<B,A>

[Andrae Muys]

An initial implementation of a Validation Applicative.

See Issue #35 for details.

→ <<cset 8e920fb3dde1>>

[Jean-Baptiste Giraudeau]

Since Validation is isomorphic to Either, I think the approach of FJ - Validation just being a wrapper around Either with specialized/renamed methods - is good to maximize code reuse.

[Matthew McMahon]

Is this likely to make it into a (soon) future version? Would be very useful

[Anund McKague]

I'm not currently looking at it. I'd be happy to work though an implementation in PR though.