Contravariantly converting Guava Predicates

Question

I have a Predicate<Object> and need an equivalent Predicate<Animal>.

Predicate<Animal> provideIsSentientPredicate() {
    // Won't compile -- cannot convert from Predicate<Object> to Predicate<Animal>
    return Predicates.instanceOf(Human.class);
}

Predicates are contravariant, so converting a Predicate<Object> to a Predicate<Animal> is safe. Is there clean and readable way to convert from a Predicate<Object> to a Predicate<Animal> (e.g. without suppressing warnings)?

I'd prefer not to change my method's type signature to return a Predicate<? super Animal> or Predicate<Object> unless someone convinces me that is the correct thing to do.

Related question for Optional: http://stackoverflow.com/questions/7848789/how-to-use-guava-optional-as-naturally-covariant-object — Robert Cooper, Apr 22 '13 at 17:20
The type of `Predicates.instanceOf` is absolutely intended to push you to either return a `Predicate super Animal>`, or to return `Predicate — Louis Wasserman, Apr 22 '13 at 17:31
@Louis Wow! Really? That wasn't my expectation at all! I've always thought it was rude to make a client deal with a wildcard type and so I've never written a method with a wildcard return type. Are wildcard return types ever considered good practice? I can't recall any method in the Guava source tree that returns a wildcard type -- is there one? — Robert Cooper, Apr 22 '13 at 17:57
I don't have a serious preference one way or the other, but if you dislike returning wildcard types, then it's totally legit to return `Predicate — Louis Wasserman, Apr 22 '13 at 18:06
Okay, I've been convinced that the examples from `Predicates` as how they want you to use it: return specific types, even if they're more general than necessary, and _consume_ `Predicate super Foo>`. — Louis Wasserman, Apr 22 '13 at 18:18
@Louis So, in this case, return the specific type `Predicate`? — Robert Cooper, Apr 22 '13 at 18:20
let us [continue this discussion in chat](http://chat.stackoverflow.com/rooms/28682/discussion-between-robert-cooper-and-louis-wasserman) — Robert Cooper, Apr 22 '13 at 18:21

score 4 · Answer 1 · edited Apr 22 '13 at 20:39

4

Predicate<Animal> provideIsSentientPredicate() 
{
    return cast( Predicates.instanceOf(Human.class) );
}

static <A, B extends A> Predicate<B> cast(Predicate<A> pa)
{
    @SuppressWarnings("unchecked")
    Predicate<B> pb = (Predicate)(pa);
    return pb;

    // we know it works correctly in practice.
    // or if you are a theorist, create a wrapper predicate
    //
    //     Predicate<B>
    //         boolean apply(B b)
    //             return pa.apply(b);
}

BTW, there's no reason why Guava shouldn't declare the method as

static <T> Predicate<T> instanceOf(Class<?> clazz)

consistent with other sibling methods.

edited Apr 22 '13 at 20:39

Paul Bellora

54,340
18
130
181

answered Apr 22 '13 at 19:05

ZhongYu

19,446
5
33
61

1

Yeah, I couldn't figure out why it isn't `static Predicate instanceOf(Class> clazz)` either, but the Guava team usually has good (if subtle) reasons for this sort of thing. Guava avoids generics in other static factory methods (`Ordering.allEqual()`, `Ordering.arbitrary()`, `Ordering.usingToString()`), and @Louis comment seems to suggest that `Predicates.instanceOf()`'s return type is intentional. Still no clue why, though. – Robert Cooper Apr 22 '13 at 19:54
2

In reply to my previous comment, I would guess the most likely reasons the method signature of `Predicate – Robert Cooper Apr 22 '13 at 20:07

Contravariantly converting Guava Predicates

1 Answers1