Can I duplicate a Stream in Java 8?

小开

It is not possible to duplicate a stream in this way. However, you can avoid the code duplication by moving the common part into a method or lambda expression.

Supplier<IntStream> supplier = () ->
IntStream.range(1, 100).filter(n -> n % 2 == 0);
supplier.get().filter(...);
supplier.get().filter(...);

小开

Update: This doesn't work. See explanation below, after the text of the original answer.

How silly of me. All that I need to do is:

Stream desired_stream = IntStream.range(1, 100).filter(n -> n % 2 == 0);
Stream stream14 = desired_stream.filter(n -> n % 7 == 0); // multiples of 14
Stream stream10 = desired_stream.filter(n -> n % 5 == 0); // multiples of 10

Explanation why this does not work:

If you code it up and try to collect both streams, the first one will collect fine, but trying to stream the second one will throw the exception: java.lang.IllegalStateException: stream has already been operated upon or closed.

To elaborate, streams are stateful objects (which by the way cannot be reset or rewound). You can think of them as iterators, which in turn are like pointers. So stream14 and stream10 can be thought of as references to the same pointer. Consuming the first stream all the way will cause the pointer to go "past the end." Trying to consume the second stream is like trying to access a pointer that is already "past the end," Which naturally is an illegal operation.

As the accepted answer shows, the code to create the stream must be executed twice but it can be compartmentalized into a Supplier lambda or a similar construct.

Full test code: save into Foo.java, then javac Foo.java, then java Foo

import java.util.stream.IntStream;


public class Foo {
public static void main (String [] args) {
IntStream s = IntStream.range(0, 100).filter(n -> n % 2 == 0);
IntStream s1 = s.filter(n -> n % 5 == 0);
s1.forEach(n -> System.out.println(n));
IntStream s2 = s.filter(n -> n % 7 == 0);
s2.forEach(n -> System.out.println(n));
}
}

Output:

$ javac Foo.java
$ java Foo
0
10
20
30
40
50
60
70
80
90
Exception in thread "main" java.lang.IllegalStateException: stream has already been operated upon or closed
at java.util.stream.AbstractPipeline.<init>(AbstractPipeline.java:203)
at java.util.stream.IntPipeline.<init>(IntPipeline.java:91)
at java.util.stream.IntPipeline$StatelessOp.<init>(IntPipeline.java:592)
at java.util.stream.IntPipeline$9.<init>(IntPipeline.java:332)
at java.util.stream.IntPipeline.filter(IntPipeline.java:331)
at Foo.main(Foo.java:8)

小开

Either,

Move the initialisation into a method, and simply call the method again

This has the advantage of being explicit about what you are doing, and also works for infinite streams.

Collect the stream and then re-stream it

In your example:

final int[] arr = IntStream.range(1, 100).filter(n -> n % 2 == 0).toArray();

Then

final IntStream s = IntStream.of(arr);

小开

最佳答案

It is not possible in general.

If you want to duplicate an input stream, or input iterator, you have two options:

A. Keep everything in a collection, say a `List<>`

Suppose you duplicate a stream into two streams s1 and s2. If you have advanced n1 elements in s1 and n2 elements with s2, you must keep |n2 - n1| elements in memory, just to keep pace. If your stream is infinite, there may be no upper bound for the storage required.

Take a look at Python's tee() to see what it takes:

This itertool may require significant auxiliary storage (depending on how much temporary data needs to be stored). In general, if one iterator uses most or all of the data before another iterator starts, it is faster to use list() instead of tee().

B. When possible: Copy the state of the generator that creates the elements

For this option to work, you'll probably need access to the inner workings of the stream. In other words, the generator - the part that creates the elements - should support copying in the first place. [OP: See this great answer, as an example of how this can be done for the example in the question]

It will not work on input from the user, since you'll have to copy the state of the entire "outside world". Java's Stream do not support copying, since it is designed to be as general as possible; for example, to work with files, network, keyboard, sensors, randomness etc. [OP: Another example is a stream that reads a temperature sensor on demand. It cannot be duplicated without storing a copy of the readings]

This is not only the case in Java; this is a general rule. You can see that std::istream in C++ only supports move semantics, not copy semantics ("copy constructor (deleted)"), for this reason (and others).

小开

It's possible if you're buffering elements that you've consumed in one duplicate, but not in the other yet.

We've implemented a duplicate() method for streams in jOOλ, an Open Source library that we created to improve integration testing for jOOQ. Essentially, you can just write:

Tuple2<Seq<Integer>, Seq<Integer>> desired_streams = Seq.seq(
IntStream.range(1, 100).filter(n -> n % 2 == 0).boxed()
).duplicate();

_{(note: we currently need to box the stream, as we haven't implemented an IntSeq yet)}

Internally, there is a LinkedList buffer storing all values that have been consumed from one stream but not from the other. That's probably as efficient as it gets if your two streams are consumed about at the same rate.

Here's how the algorithm works:

static <T> Tuple2<Seq<T>, Seq<T>> duplicate(Stream<T> stream) {
final LinkedList<T> gap = new LinkedList<>();
final Iterator<T> it = stream.iterator();


@SuppressWarnings("unchecked")
final Iterator<T>[] ahead = new Iterator[] { null };


class Duplicate implements Iterator<T> {
@Override
public boolean hasNext() {
if (ahead[0] == null || ahead[0] == this)
return it.hasNext();


return !gap.isEmpty();
}


@Override
public T next() {
if (ahead[0] == null)
ahead[0] = this;


if (ahead[0] == this) {
T value = it.next();
gap.offer(value);
return value;
}


return gap.poll();
}
}


return tuple(seq(new Duplicate()), seq(new Duplicate()));
}

More source code here

In fact, using jOOλ, you'll be able to write a complete one-liner like so:

Tuple2<Seq<Integer>, Seq<Integer>> desired_streams = Seq.seq(
IntStream.range(1, 100).filter(n -> n % 2 == 0).boxed()
).duplicate()
.map1(s -> s.filter(n -> n % 7 == 0))
.map2(s -> s.filter(n -> n % 5 == 0));


// This will yield 14, 28, 42, 56...
desired_streams.v1.forEach(System.out::println)


// This will yield 10, 20, 30, 40...
desired_streams.v2.forEach(System.out::println);

小开

You can also move the stream generation into separate method/function that returns this stream and call it twice.

小开

For non-infinite streams, if you have access to the source, its straight forward:

@Test
public void testName() throws Exception {
List<Integer> integers = Arrays.asList(1, 2, 4, 5, 6, 7, 8, 9, 10);
Stream<Integer> stream1 = integers.stream();
Stream<Integer> stream2 = integers.stream();


stream1.forEach(System.out::println);
stream2.forEach(System.out::println);
}

prints

1 2 4 5 6 7 8 9 10

1 2 4 5 6 7 8 9 10

For your case:

Stream originalStream = IntStream.range(1, 100).filter(n -> n % 2 == 0)


List<Integer> listOf = originalStream.collect(Collectors.toList())


Stream stream14 = listOf.stream().filter(n -> n % 7 == 0);
Stream stream10 = listOf.stream().filter(n -> n % 5 == 0);

For performance etc, read someone else's answer ;)

小开

I used this great answer to write following class:

public class SplitStream<T> implements Stream<T> {
private final Supplier<Stream<T>> streamSupplier;


public SplitStream(Supplier<Stream<T>> t) {
this.streamSupplier = t;
}


@Override
public Stream<T> filter(Predicate<? super T> predicate) {
return streamSupplier.get().filter(predicate);
}


@Override
public <R> Stream<R> map(Function<? super T, ? extends R> mapper) {
return streamSupplier.get().map(mapper);
}


@Override
public IntStream mapToInt(ToIntFunction<? super T> mapper) {
return streamSupplier.get().mapToInt(mapper);
}


@Override
public LongStream mapToLong(ToLongFunction<? super T> mapper) {
return streamSupplier.get().mapToLong(mapper);
}


@Override
public DoubleStream mapToDouble(ToDoubleFunction<? super T> mapper) {
return streamSupplier.get().mapToDouble(mapper);
}


@Override
public <R> Stream<R> flatMap(Function<? super T, ? extends Stream<? extends R>> mapper) {
return streamSupplier.get().flatMap(mapper);
}


@Override
public IntStream flatMapToInt(Function<? super T, ? extends IntStream> mapper) {
return streamSupplier.get().flatMapToInt(mapper);
}


@Override
public LongStream flatMapToLong(Function<? super T, ? extends LongStream> mapper) {
return streamSupplier.get().flatMapToLong(mapper);
}


@Override
public DoubleStream flatMapToDouble(Function<? super T, ? extends DoubleStream> mapper) {
return streamSupplier.get().flatMapToDouble(mapper);
}


@Override
public Stream<T> distinct() {
return streamSupplier.get().distinct();
}


@Override
public Stream<T> sorted() {
return streamSupplier.get().sorted();
}


@Override
public Stream<T> sorted(Comparator<? super T> comparator) {
return streamSupplier.get().sorted(comparator);
}


@Override
public Stream<T> peek(Consumer<? super T> action) {
return streamSupplier.get().peek(action);
}


@Override
public Stream<T> limit(long maxSize) {
return streamSupplier.get().limit(maxSize);
}


@Override
public Stream<T> skip(long n) {
return streamSupplier.get().skip(n);
}


@Override
public void forEach(Consumer<? super T> action) {
streamSupplier.get().forEach(action);
}


@Override
public void forEachOrdered(Consumer<? super T> action) {
streamSupplier.get().forEachOrdered(action);
}


@Override
public Object[] toArray() {
return streamSupplier.get().toArray();
}


@Override
public <A> A[] toArray(IntFunction<A[]> generator) {
return streamSupplier.get().toArray(generator);
}


@Override
public T reduce(T identity, BinaryOperator<T> accumulator) {
return streamSupplier.get().reduce(identity, accumulator);
}


@Override
public Optional<T> reduce(BinaryOperator<T> accumulator) {
return streamSupplier.get().reduce(accumulator);
}


@Override
public <U> U reduce(U identity, BiFunction<U, ? super T, U> accumulator, BinaryOperator<U> combiner) {
return streamSupplier.get().reduce(identity, accumulator, combiner);
}


@Override
public <R> R collect(Supplier<R> supplier, BiConsumer<R, ? super T> accumulator, BiConsumer<R, R> combiner) {
return streamSupplier.get().collect(supplier, accumulator, combiner);
}


@Override
public <R, A> R collect(Collector<? super T, A, R> collector) {
return streamSupplier.get().collect(collector);
}


@Override
public Optional<T> min(Comparator<? super T> comparator) {
return streamSupplier.get().min(comparator);
}


@Override
public Optional<T> max(Comparator<? super T> comparator) {
return streamSupplier.get().max(comparator);
}


@Override
public long count() {
return streamSupplier.get().count();
}


@Override
public boolean anyMatch(Predicate<? super T> predicate) {
return streamSupplier.get().anyMatch(predicate);
}


@Override
public boolean allMatch(Predicate<? super T> predicate) {
return streamSupplier.get().allMatch(predicate);
}


@Override
public boolean noneMatch(Predicate<? super T> predicate) {
return streamSupplier.get().noneMatch(predicate);
}


@Override
public Optional<T> findFirst() {
return streamSupplier.get().findFirst();
}


@Override
public Optional<T> findAny() {
return streamSupplier.get().findAny();
}


@Override
public Iterator<T> iterator() {
return streamSupplier.get().iterator();
}


@Override
public Spliterator<T> spliterator() {
return streamSupplier.get().spliterator();
}


@Override
public boolean isParallel() {
return streamSupplier.get().isParallel();
}


@Override
public Stream<T> sequential() {
return streamSupplier.get().sequential();
}


@Override
public Stream<T> parallel() {
return streamSupplier.get().parallel();
}


@Override
public Stream<T> unordered() {
return streamSupplier.get().unordered();
}


@Override
public Stream<T> onClose(Runnable closeHandler) {
return streamSupplier.get().onClose(closeHandler);
}


@Override
public void close() {
streamSupplier.get().close();
}
}

When you call any method of it's class, it delegates call to

streamSupplier.get()

So, instead of:

Supplier<IntStream> supplier = () ->
IntStream.range(1, 100).filter(n -> n % 2 == 0);
supplier.get().filter(...);
supplier.get().filter(...);

You can do:

SplitStream<Integer> stream =
new SplitStream<>(() -> IntStream.range(1, 100).filter(n -> n % 2 == 0).boxed());
stream.filter(...);
stream.filter(...);

You can expand it to work with IntStream, DoubleStream, etc...

小开

I think that the use of Concat with an empty stream could attend your need. Try something like this:

Stream<Integer> concat = Stream.concat(Stream.of(1, 2), Stream.empty());

小开

Starting with Java 12 we have Collectors::teeing that allows us to pass elements of the main stream pipeline to 2 or more downstream collectors.

Based on your example we can do the following:

@Test
void shouldProcessStreamElementsInTwoSeparateDownstreams() {
class Result {
List<Integer> multiplesOf7;
List<Integer> multiplesOf5;


Result(List<Integer> multiplesOf7, List<Integer> multiplesOf5) {
this.multiplesOf7 = multiplesOf7;
this.multiplesOf5 = multiplesOf5;
}
}


var result = IntStream.range(1, 100)
.filter(n -> n % 2 == 0)
.boxed()
.collect(Collectors.teeing(
Collectors.filtering(n -> n % 7 == 0, Collectors.toList()),
Collectors.filtering(n -> n % 5 == 0, Collectors.toList()),
Result::new
));


assertTrue(result.multiplesOf7.stream().allMatch(n -> n % 7 == 0));
assertTrue(result.multiplesOf5.stream().allMatch( n -> n % 5 == 0));
}

There are many other collectors that allows to do other things e.g. by using Collectors::mapping in downstream you can obtain two different objects/types from the same source as shown in this article.

小开

Straight answer is: yes

There's no specific support for this but one can implement it. The possible approaches that I see are these:
a. copy the entire stream data and then create the stream copies based on it -> the RAM consumption might be an impediment
b. read the stream and relay each of its elements to the copies -> I'll detail this approach below

The Concept

Let's imagine b. solution:
<T> List<Stream<T>> copyStream(int copiesCount, Stream<T> originalStream)
allows one to create copiesCount copies of the originalStream.

To understand the solution one has to understand the difference between a stream and the data-elements that might flow through it: for example an apple, a carrot and a potato would be data-elements while a pipe through which they move to reach some destination would be the stream. Copying a Stream it's as if creating more pipes: one has then to connect the original pipe (i.e. originalStream) to the additional ones (aka streamCopies); while in real world one can't pass an apple-object from one pipe to more pipes (i.e. streamCopies) in programming this is possible: just pass the variable containing the apple-object reference to the stream copies.

Implementation Details

The Java implementation of the Stream has a great impact on the solution's shape. First impact is related to what happens when data-elements flow through a stream (aka pipe): to actually read (& process) the elements in a Stream a terminal method has to be used, e.g. forEach. In our case originalStream.forEach must be called so that each element is read and passed to the streamCopies (aka downstream pipes); this must happen before copyStream() method returns, which is bad because forEach would block till all originalStream elements are consumed. To solve this copyStream() implementation will spawn a thread in which to call originalStream.forEach. Consuming originalStream elements means passing them to the downstream pipes (i.e. streamCopies); because there's no cache one has to ensure that each originalStream element is transferred to each streamCopies before getting to the next one. This means that all streamCopies must consume the same time: if some streamCopies is not consuming it will block all other streamCopies because originalStream will stop transferring to downstream pipes till everyone consumed current element (aka it will cache nothing for the late streamCopies consumers). But to consume a Stream in Java implies calling a terminal operation on it (e.g. forEach) which blocks the execution till the entire stream is consumed; because we need all streamCopies to be consumed in parallel this must happen on a distinct thread for each! Well, as a miscellaneous fact, one of the streamCopies could in fact be consumed on the current (main) thread. Summarizing, the solution usage would look like below:

List<Stream<?>> streamCopies = copyStream(copiesCount, originalStream);`
// start a thread for each `streamCopies` into which consume the corresponding
// stream copy (one of them could be consumed on the current thread though)
// optionally join the consuming threads
// continue your whatever business logic you have

Final Considerations

Some of the limitations apparent above can be circumvented:

the copying process is destructive, i.e. originalStream will be unusable after calling copyStream() because it'll be in a pending-consumption. If one really wants to consume it he can create an additional copy which to maybe consume on the current (main) thread (but only after starting the consumption of all other copies)
streamCopies must consume all received originalStream elements, otherwise, if one stops, the others block too (read the "Implementation Details" part again to understand why). This means each streamCopies element consumption must occur in a try...catch to ensure the lack of failures (aka processing stop). A production implementation would in fact circumvent this by wrapping each Stream copy with something overwriting close() method such that to remove the failed stream copy from the originalStream-to-streamCopies transfer logic (aka discard the underlying blockingQueue used for the communication between originalStream thread and originalStream thread -> see the implementation below). This implies that the clients would be forced to originalStream1 the Stream copies but that’s not so uncommon, e.g. see Spring’s originalStream3 outcome having same requirement.
as pointed before, each streamCopies terminal operation must be executed in a distinct thread - there's no workaround for this

The Code

Below is the code implementing the b. solution and a test checking its correctness.

@Test
void streamCopyTest() throws ExecutionException, InterruptedException {
// streamCopies are valid/normal Stream
// instances (e.g. it is allowed to be infinite)
List<Stream<String>> streamCopies = copyStream(3, Stream.of("a", "b", "c", "d"));
// The 3 copies relay on the original stream which can’t be
// consumed more than once! Consuming the copies one by one
// in the same thread isn’t possible because 1st consumed
// copy would leave nothing to consume for the others,
// so they must be consumed in parallel.
ExecutorService executorService = Executors.newCachedThreadPool();
CompletableFuture<?>[] futures =
streamCopies.stream().map(stream -> CompletableFuture.runAsync(() -> {
// the same consumption logic for all streamCopies is
// used here because this is just an example; the
// actual consumption logic could be distinct (and anything)
String outcome = stream.collect(Collectors.joining(", "));
// check the thread name in the message to differentiate the outcome
log.info("\n{}", outcome);
}, executorService)).toArray(CompletableFuture[]::new);
CompletableFuture.allOf(futures).get();
executorService.shutdown();
}


@RequiredArgsConstructor
@Slf4j
public class StreamCopiesFactory {


/**
* The amount of elements to be stored in the blockingQueue used
* to transfer elements from the original stream to its copies.
* This is very different to the cache use for the a. solution:
* here is about the transfer between original stream and its
* copies instead of the entire original stream data-copy.
* Change or make this configurable.
*/
private static final int cacheSize = 1;


/**
* Each of these stream copies must execute (their terminal operation)
* on a distinct thread! One of them could actually execute on the
* main thread, but only after all the others were started on their
* distinct thread.
*/
public static <T> List<Stream<T>> copyStream(int copies, Stream<T> stream) {
List<BlockingQueue<Object>> blockingQueues = new ArrayList<>(copies);
// creating the queues used to relay the stream's elements to the stream's copies
for (int i = 0; i < copies; i++) {
blockingQueues.add(new LinkedBlockingQueue<>(cacheSize));
}
// consume the stream copies in a distinct thread, otherwise
// bq.put (transferring for the next stream copy) would block
// because the 2nd stream copy isn't yet consuming
Executors.newSingleThreadExecutor().execute(() -> {
stream.forEach(streamElement -> blockingQueues.forEach(bq -> {
try {
bq.put(streamElement);
} catch (InterruptedException e) {
log.error(e.getMessage(), e);
// nothing to do here other than maybe simple optimization related to the
// failed bq.put (e.g. sending END_SIGNAL into bq then skipping its next put calls)
}
}));
blockingQueues.forEach(bq -> {
try {
bq.put(END_SIGNAL);
} catch (InterruptedException e) {
log.error(e.getMessage(), e);
// nothing to do here
}
});
});
// creating the copies
// A production implementation would wrap each Stream copy with
// something overwriting close() which to remove from blockingQueues
// the blockingQueue corresponding to the closed Stream.
return blockingQueues.stream().map(bq -> new SpliteratorCopy<T>(bq))
.map(spliterator -> StreamSupport.stream(spliterator, false))
.collect(Collectors.toList());
}
}


@RequiredArgsConstructor
@Slf4j
public class SpliteratorCopy<T> implements Spliterator<T> {


public static final Object END_SIGNAL = new Object();


private final BlockingQueue<?> blockingQueue;


@Override
public boolean tryAdvance(final Consumer<? super T> action) {
Object nextElement;
try {
nextElement = blockingQueue.take();
} catch (InterruptedException e) {
log.error(e.getMessage(), e);
throw new RuntimeException(e);
}
if (nextElement == END_SIGNAL) {
return false;
}
action.accept((T) nextElement);
return true;
}


@Override
public Spliterator<T> trySplit() {
return null;
}


@Override
public long estimateSize() {
return Long.MAX_VALUE;
}


@Override
public int characteristics() {
return Spliterator.ORDERED;
}
}

Can I duplicate a Stream in Java 8?

A. Keep everything in a collection, say a List<>

B. When possible: Copy the state of the generator that creates the elements

A. Keep everything in a collection, say a `List<>`