Introduction to Spliterator in Java – Java中的Spliterator简介

最后修改: 2018年 1月 27日

1. Overview


The Spliterator interface, introduced in Java 8, can be used for traversing and partitioning sequences. It’s a base utility for Streams, especially parallel ones.

Java 8中引入的Spliterator接口可以用于遍历和分割序列。它是Streams的基本工具,尤其是并行的。

In this article, we’ll cover its usage, characteristics, methods and how to create our own custom implementations.


2. Spliterator API

2.Spliterator API

2.1. tryAdvance


This is the main method used for stepping through a sequence. The method takes a Consumer that’s used to consume elements of the Spliterator one by one sequentially and returns false if there’re no elements to be traversed.


Here, we’ll take a look at how to use it to traverse and partition elements.


First, let’s assume that we’ve got an ArrayList with 35000 articles and that Article class is defined as:


public class Article {
    private List<Author> listOfAuthors;
    private int id;
    private String name;
    // standard constructors/getters/setters

Now, let’s implement a task that processes the list of articles and adds a suffix of “– published by Baeldung” to each article name:

现在,让我们实现一个任务,处理文章列表,并为每篇文章的名称添加一个后缀”– 由Baeldung出版”

public String call() {
    int current = 0;
    while (spliterator.tryAdvance(a -> a.setName(article.getName()
      .concat("- published by Baeldung")))) {
    return Thread.currentThread().getName() + ":" + current;

Notice that this task outputs the number of articles processed when it finishes the execution.


Another key point is that we used tryAdvance() method to process the next element.


2.2. trySplit

2.2. trySplit

Next, let’s split Spliterators (hence the name) and process partitions independently.

接下来,让我们分割Spliterators (因此而得名)并独立处理分区。

The trySplit method tries to split it into two parts. Then the caller process elements, and finally, the returned instance will process the others, allowing the two to be processed in parallel.


Let’s generate our list first:


public static List<Article> generateElements() {
    return Stream.generate(() -> new Article("Java"))

Next, we obtain our Spliterator instance using the spliterator() method. Then we apply our trySplit() method:


public void givenSpliterator_whenAppliedToAListOfArticle_thenSplittedInHalf() {
    Spliterator<Article> split1 = Executor.generateElements().spliterator(); 
    Spliterator<Article> split2 = split1.trySplit(); 
    assertThat(new Task(split1).call()) 
      .containsSequence(Executor.generateElements().size() / 2 + ""); 
    assertThat(new Task(split2).call()) 
      .containsSequence(Executor.generateElements().size() / 2 + ""); 

The splitting process worked as intended and divided the records equally.


2.3. estimatedSize


The estimatedSize method gives us an estimated number of elements:

estimatedSize方法给我们一个估计的元素数量。"Size: " + split1.estimateSize());

This will output:


Size: 17500

2.4. hasCharacteristics


This API checks if the given characteristics match the properties of the Spliterator. Then if we invoke the method above, the output will be an int representation of those characteristics:

这个API检查给定的特征是否与Spliterator的属性相符。然后如果我们调用上面的方法,输出将是这些特性的int 代表。"Characteristics: " + split1.characteristics());
Characteristics: 16464

3. Spliterator Characteristics


It has eight different characteristics that describe its behavior. Those can be used as hints for external tools:


  • SIZED if it’s capable of returning an exact number of elements with the estimateSize() method
  • SORTED – if it’s iterating through a sorted source
  • SUBSIZED – if we split the instance using a trySplit() method and obtain Spliterators that are SIZED as well
  • CONCURRENT – if source can be safely modified concurrently
  • DISTINCT – if for each pair of encountered elements x, y, !x.equals(y)
  • IMMUTABLE – if elements held by source can’t be structurally modified
  • NONNULL – if source holds nulls or not
  • ORDERED – if iterating over an ordered sequence

4. A Custom Spliterator


4.1. When to Customize


First, let’s assume the following scenario:


We’ve got an article class with a list of authors, and the article that can have more than one author. Furthermore, we consider an author related to the article if his related article’s id matches article id.


Our Author class will look like the this:


public class Author {
    private String name;
    private int relatedArticleId;

    // standard getters, setters & constructors

Next, we’ll implement a class to count authors while traversing a stream of authors. Then the class will perform a reduction on the stream.


Let’s have a look at the class implementation:


public class RelatedAuthorCounter {
    private int counter;
    private boolean isRelated;
    // standard constructors/getters
    public RelatedAuthorCounter accumulate(Author author) {
        if (author.getRelatedArticleId() == 0) {
            return isRelated ? this : new RelatedAuthorCounter( counter, true);
        } else {
            return isRelated ? new RelatedAuthorCounter(counter + 1, false) : this;

    public RelatedAuthorCounter combine(RelatedAuthorCounter RelatedAuthorCounter) {
        return new RelatedAuthorCounter(
          counter + RelatedAuthorCounter.counter, 

Each method in the above class performs a specific operation to count while traversing.


First, the accumulate() method traverse the authors one by one in an iterative way, then combine() sums two counters using their values. Finally, the getCounter() returns the counter.


Now, to test what we’ve done so far. Let’s convert our article’s list of authors to a stream of authors:


Stream<Author> stream = article.getListOfAuthors().stream();

And implement a countAuthor() method to perform the reduction on the stream using RelatedAuthorCounter:


private int countAutors(Stream<Author> stream) {
    RelatedAuthorCounter wordCounter = stream.reduce(
      new RelatedAuthorCounter(0, true), 
    return wordCounter.getCounter();

If we used a sequential stream the output will be as expected “count = 9”, however, the problem arises when we try to parallelize the operation.

如果我们使用一个顺序流,输出将是预期的“count = 9”,然而,当我们试图将操作并行化时,问题就出现了。

Let’s take a look at the following test case:


  givenAStreamOfAuthors_whenProcessedInParallel_countProducesWrongOutput() {

Apparently, something has gone wrong – splitting the stream at a random position caused an author to be counted twice.


4.2. How to Customize


To solve this, we need to implement a Spliterator that splits authors only when related id and articleId matches. Here’s the implementation of our custom Spliterator:


public class RelatedAuthorSpliterator implements Spliterator<Author> {
    private final List<Author> list;
    AtomicInteger current = new AtomicInteger();
    // standard constructor/getters

    public boolean tryAdvance(Consumer<? super Author> action) {
        return current.get() < list.size();

    public Spliterator<Author> trySplit() {
        int currentSize = list.size() - current.get();
        if (currentSize < 10) {
            return null;
        for (int splitPos = currentSize / 2 + current.intValue();
          splitPos < list.size(); splitPos++) {
            if (list.get(splitPos).getRelatedArticleId() == 0) {
                Spliterator<Author> spliterator
                  = new RelatedAuthorSpliterator(
                  list.subList(current.get(), splitPos));
                return spliterator;
        return null;

   public long estimateSize() {
       return list.size() - current.get();
   public int characteristics() {
       return CONCURRENT;

Now applying countAuthors() method will give the correct output. The following code demonstrates that:


public void
  givenAStreamOfAuthors_whenProcessedInParallel_countProducesRightOutput() {
    Stream<Author> stream2 =, true);

Also, the custom Spliterator is created from a list of authors and traverses through it by holding the current position.


Let’s discuss in more details the implementation of each method:


  • tryAdvance passes authors to the Consumer at the current index position and increments its position
  • trySplit defines the splitting mechanism, in our case, the RelatedAuthorSpliterator is created when ids matched, and the splitting divides the list into two parts
  • estimatedSize – is the difference between the list size and the position of currently iterated author
  • characteristics – returns the Spliterator characteristics, in our case SIZED as the value returned by the estimatedSize() method is exact; moreover, CONCURRENT indicates that the source of this Spliterator may be safely modified by other threads

5. Support for Primitive Values


The Spliterator API supports primitive values including double, int and long.

Spliterator API支持包括doubleintlong的原始值。

The only difference between using a generic and a primitive dedicated Spliterator is the given Consumer and the type of the Spliterator.


For example, when we need it for an int value we need to pass an intConsumer. Furthermore, here’s a list of primitive dedicated Spliterators:


  • OfPrimitive<T, T_CONS, T_SPLITR extends Spliterator.OfPrimitive<T, T_CONS, T_SPLITR>>: parent interface for other primitives
  • OfInt: A Spliterator specialized for int
  • OfDouble: A Spliterator dedicated for double
  • OfLong: A Spliterator dedicated for long

6. Conclusion


In this article, we covered Java 8 Spliterator usage, methods, characteristics, splitting process, primitive support and how to customize it.

在这篇文章中,我们介绍了Java 8 Spliterator的用法、方法、特征、分割过程、基元支持以及如何定制它。

As always, the full implementation of this article can be found over on Github.