How to collect more than a single collection or a single scalar

A well-known characteristic of Java Stream is that it can be consumed only once, thus if we want to do some operations that require two passes or more through elements we have to create a new stream or develop a custom collector. Thankfully, Java 12 comes with Collectors::teeing that solves exactly this situation.

Imagine that we need a concise way to get both the min and max values of a stream.

We know how to do that if deal with specialized primitive streams by terminating it with operations like summarizingInt:

void shouldFindMinAndMaxInt() {
    IntSummaryStatistics statistics = IntStream.of(32, 42, 1, 2)

    assertEquals(42, statistics.getMax());
    assertEquals(1, statistics.getMin());

Similarly we have specialized collectors when dealing with objects:

void shouldFindMinAndMaxProduct() {
    record Product(String label, double price) {}
    var products = Stream.of(
            new Product("T-Shirt", 12.99),
            new Product("socks", 5.09),
            new Product("pants", 89.1)

    DoubleSummaryStatistics statistics = products.collect(
    assertEquals(89.01, statistics.getMax());
    assertEquals(5.09, statistics.getMin());

The drawback of these collectors is that we don’t know which object is the max or min we have just its value. Using Collectors::maxBy and Collectors::minBy will require to pass the stream twice.

Thankfully, the following collector alleviates this problem:

public static <T, R1, R2, R> Collector<T, ?, R> teeing(
    Collector<? super T, ?, R1> downstream1,
    Collector<? super T, ?, R2> downstream2,
    BiFunction<? super R1, ? super R2, R> merger) {

Collectors::teeing has three arguments first two are downstream collectors. Every stream element is processed by each of them. The third parameter is a BiFunction that merges two results into the single one, e.g. it can be a function that creates a pair.

Consider the following record representing a movie:

record Movie(String title, double rating) {}

Finding the best and worst movie in a single stream pass is as simple as:

void shouldFindWorstAndBestMovie() {
    var m1 = new Movie("Groundhog Day", 8);
    var m2 = new Movie("Stop! Or My Mom Will Shoot", 4.4);
    var m3 = new Movie("Forrest Gump", 8.8);

    record MovieStatistics(Movie worst, Movie best) {}

    var ratingComparator = Comparator.comparing(Movie::rating);

    var movieStatistics = Stream.of(m1, m2, m3)
                    (worst, best) -> new MovieStatistics(

    assertEquals(m2, movieStatistics.worst());

The best part about teeing collector is that we can handle much more complex scenarios by combining multiple collectors downstream. Consider the following enhanced Movie record:

record Movie(
    String title,
    double rating,
    List<Genre> genres
) {}

enum Genre {

We can find titles of drama movies and their average rating:

void shouldFindDramasTitlesAndAvgRating() {
    List<Movie> movies = readMoviesFromCSVFile();

    record MovieStatistics(List<String> titles, double avgRating) {}

    var movieStatistics =
            .filter(m -> m.genres().contains(DRAMA))
                    mapping(Movie::title, toList()),

    var expectedTitles = List.of(
            "Pulp Fiction",
            "You've Got Mail",
            "Forrest Gump"
    assertEquals(expectedTitles.size(), movieStatistics.titles().size());
    assertEquals(8.13, movieStatistics.avgRating(), /* delta */ 2);

In the main pipeline we filter movies by genre, then in teeing collector, for one downstream we extract movie titles and collect them in a list. For the second downstream, we produce a scalar - average rating. Finally, for the merger function, we pass a constructor reference that matches types returned by both downstream collectors.


We can solve most of the problems with one out-of-the-box collector, even when we need to collect more than a collection or a scalar.

Full code can be found over on GitHub.

Related Posts

A Feature Toggle Story

Unit Testing Anti-Patterns

REST Search API with QueryDSL

Running Java shebang with Kubernetes

How to Organize the Code in the Ports and Adapters Architecture

Functional Programming Concepts in Java

An Introduction to Java Sealed Classes and Interfaces