Java 8 introduced the concept of collectors. Most of the time we barely use factory methods from Collectors class, e.g. collect(toList()), toSet() or maybe something more fancy like counting() or groupingBy(). Not many of us actually bother to look how collectors are defined and implemented. Let’s start from analysing what Collector<T, A, R> really is and how it works.
Collector<T, A, R> works as a “sink” for streams – stream pushes items (one after another) into a collector, which should produce some “collected” value in the end. Most of the time it means building a collection (like toList()) by accumulating elements or reducing stream into something smaller (e.g. counting() collector that barely counts elements). Every collector accepts items of type T and produces aggregated (accumulated) value of type R (e.g. R = List<T>). Generic type A simply defines the type of intermediate mutable data structure that we are going to use to accumulate items of type T in the meantime. Type A can, but doesn’t have to be the same as R – in simple words the mutable data structure that we use to collect items from input Stream<T> can be different than the actual output collection/value. That being said, every collector must implement the following methods:
interface Collector<T,A,R> {
Supplier<A> supplier()
BiConsumer<A,T> acumulator()
BinaryOperator<A> combiner()
Function<A,R> finisher()
Set<Characteristics> characteristics()
}
supplier()returns a function that creates an instance of accumulator – mutable data structure that we will use to accumulate input elements of typeT.accumulator()returns a function that will take accumulator and one item of typeT, mutating accumulator.combiner()is used to join two accumulators together into one. It is used when collector is executed in parallel, splitting inputStream<T>and collecting parts independently first.finisher()takes an accumulatorAand turns it into a result value, e.g. collection, of typeR. All of this sounds quite abstract, so let’s do a simple example.
Obviously Java 8 doesn’t provide a built-in collector for ImmutableSet<T> from Guava. However creating one is very simple. Remember that in order to iteratively build ImmutableSet we use ImmutableSet.Builder<T> – this is going to be our accumulator.
import com.google.common.collect.ImmutableSet;
public class ImmutableSetCollector<T>
implements Collector<T, ImmutableSet.Builder<T>, ImmutableSet<T>> {
@Override
public Supplier<ImmutableSet.Builder<T>> supplier() {
return ImmutableSet::builder;
}
@Override
public BiConsumer<ImmutableSet.Builder<T>, T> accumulator() {
return (builder, t) -> builder.add(t);
}
@Override
public BinaryOperator<ImmutableSet.Builder<T>> combiner() {
return (left, right) -> {
left.addAll(right.build());
return left;
};
}
@Override
public Function<ImmutableSet.Builder<T>, ImmutableSet<T>> finisher() {
return ImmutableSet.Builder::build;
}
@Override
public Set<Characteristics> characteristics() {
return EnumSet.of(Characteristics.UNORDERED);
}
}
First of all look carefully at generic types. Our ImmutableSetCollector takes input elements of type T, so it works for any Stream<T>. In the end it will produce ImmutableSet<T> – as expected. ImmutableSet.Builder<T> is going to be our intermediate data structure.
supplier()returns a function that creates newImmutableSet.Builder<T>. If you are not that familiar with lambdas in Java 8,ImmutableSet::builderis a shorthand for() -> ImmutableSet.builder().accumulator()returns a function that takesbuilderand one element of typeT. It simply adds said element to the builder.combiner()returns a function that will accept two builders and turn them into one by adding all elements from one of them into the other – and returning the latter. Finallyfinisher()returns a function that will turnImmutableSet.Builder<T>intoImmutableSet<T>. Again this is a shorthand syntax for:builder -> builder.build().- Last but not least,
characteristics()informs JDK what capabilities our collector has. For example ifImmutableSet.Builder<T>was thread-safe (it isn’t), we could sayCharacteristics.CONCURRENTas well.
We can now use our custom collector everywhere using collect():
final ImmutableSet<Integer> set = Arrays
.asList(1, 2, 3, 4)
.stream()
.collect(new ImmutableSetCollector<>());
However creating new instance is slightly verbose so I suggest creating static factory method, similar to what JDK does:
public class ImmutableSetCollector<T> implements Collector<T, ImmutableSet.Builder<T>, ImmutableSet<T>> {
//...
public static <T> Collector<T, ?, ImmutableSet<T>> toImmutableSet() {
return new ImmutableSetCollector<>();
}
}
From now on we can take full advantage of our custom collector by simply typing: collect(toImmutableSet()). In the second part we will learn how to write more complex and useful collectors.
Update
@akarazniewicz pointed out that collectors are just verbose implementation of folding. With my love and hate relationship with folds, I have to comment on that. Collectors in Java 8 are basically object-oriented encapsulation of the most complex type of fold found in Scala, namely GenTraversableOnce.aggregate[B](z: ⇒ B)(seqop: (B, A) ⇒ B, combop: (B, B) ⇒ B): B. aggregate() is like fold(), but requires extra combop to combine two accumulators (of type B) into one. Comparing this to collectors, parameter z comes from a supplier(), seqop()reduction operation is an accumulator() and combop is a combiner(). In pseudo-code we can write:
finisher(
seq.aggregate(collector.supplier())
(collector.accumulator(), collector.combiner()))
GenTraversableOnce.aggregate() is used when concurrent reduction is possible – just like with collectors.
| Reference: | Introduction to writing custom collectors in Java 8 from our JCG partner Tomasz Nurkiewicz at the Java and neighbourhood blog. |
