1.0.0-RC3

• Merging fixes from 0.20.4 into 1.x branch
  • Pull 1667 Fix the bug that Switch doesn't propagate 'unsubscribe'
  • Pull 1659 OperatorScan should check for MAX_VALUE on request
  • Pull 1657 Ignore furthur messages after entering terminate state
  • Pull 1669 Error Handling Unsubscribe and Terminal State
  • Pull 1656 Make TakeUntil obey Rx serialization contract

Artifacts: Maven Central

0.20.4

• Pull 1667 Fix the bug that Switch doesn't propagate 'unsubscribe'
• Pull 1659 OperatorScan should check for MAX_VALUE on request
• Pull 1657 Ignore furthur messages after entering terminate state
• Pull 1669 Error Handling Unsubscribe and Terminal State
• Pull 1656 Make TakeUntil obey Rx serialization contract
• Pull 1664 StringObservable.split NPE fixes

Artifacts: Maven Central

1.0.0-RC2

• Pull 1652 Operator Scan Backpressure Fix
• Pull 1645 Remove ParallelMerge

Artifacts: Maven Central

0.20.3

• Pull 1648 Operator Scan Backpressure Fix
• Pull 1651 RxScala: Fix the problem that Subscriber.onStart isn't called
• Pull 1641 RxScala: Fix infinite recursive onStart call in Subscriber
• Pull 1646 Deprecate ParallelMerge

Artifacts: Maven Central

1.0.0-RC1

The first release candidate for 1.0.0.

This is the same code as version 0.20.2 except:

• all deprecated methods and types are deleted
• now published to groupId io.reactivex instead of com.netflix.rxjava
• artifactId is now rxjava instead of rxjava-core

io.reactivex:rxjava:1.0.0-rc.1

• all sub-projects are separated into their own projects and no longer released along with RxJava

The artifacts can be found on maven Central at: http://repo1.maven.org/maven2/io/reactivex/rxjava/1.0.0-rc.1/

Artifacts: Maven Central

0.20.2

• Pull 1637 Optimize single BlockingObservable operations

Artifacts: Maven Central

0.20.1

• Pull 1631 Handle Fatal Exceptions in doOn* operators
• Pull 1625 RxScala: Mark superfluous from/empty methods with scheduler parameter as deprecated
• Pull 1623 RxScala: Add more operators to match RxJava
• Pull 1632 Composite Exception - Circular Reference Handling

Artifacts: Maven Central

0.20.0

RxJava 0.20.0 is a major release that adds "reactive pull" support for backpressure along with several other enhancements leading into the 1.0 release.

Reactive Pull for Backpressure

Solutions for backpressure was the major focus of this release. A "reactive pull" implementation was implemented. Documentation on this and other options for backpressure are found in the wiki: https://github.com/ReactiveX/RxJava/wiki/Backpressure

The reactive pull solution evolved out of several prototypes and interaction with many people over the months.

Signature Changes

A new type Producer has been added:

public interface Producer {
    public void request(long n);
}

The Subscriber type now has these methods added:

public abstract class Subscriber<T> implements Observer<T>, Subscription {
    public void onStart();
    protected final void request(long n);
    public final void setProducer(Producer producer);
}

Examples

This trivial example shows requesting values one at a time:

Observable.from(1, 2, 3, 4).subscribe(new Subscriber<Integer>() {

    @Override
    public void onStart() {
        // on start this tells it to request 1
        // otherwise it defaults to request(Long.MAX_VALUE)
        request(1);
    }

    @Override
    public void onCompleted() {
    }

    @Override
    public void onError(Throwable e) {
    }

    @Override
    public void onNext(Integer t) {
        System.out.println(t);
        // as each onNext is consumed, request another 
        // otherwise the Producer will not send more
        request(1);
    }

});

The OnSubscribeFromIterable operator shows how an Iterable is consumed with backpressure.

Some hi-lights (modified for simplicity rather than performance and completeness):

public final class OnSubscribeFromIterable<T> implements OnSubscribe<T> {

    @Override
    public void call(final Subscriber<? super T> o) {
        final Iterator<? extends T> it = is.iterator();
        // instead of emitting directly to the Subscriber, it emits a Producer
        o.setProducer(new IterableProducer<T>(o, it));
    }

    private static final class IterableProducer<T> implements Producer {

        public void request(long n) {
            int _c = requested.getAndAdd(n);
            if (_c == 0) {
                while (it.hasNext()) {
                    if (o.isUnsubscribed()) {
                        return;
                    }
                    T t = it.next();
                    o.onNext(t);
                    if (requested.decrementAndGet() == 0) {
                        // we're done emitting the number requested so return
                        return;
                    }
                }

                o.onCompleted();
            }

        }
    }
}

The observeOn operator is a sterotypical example of queuing on one side of a thread and draining on the other, now with backpressure.

private static final class ObserveOnSubscriber<T> extends Subscriber<T> {
        @Override
        public void onStart() {
            // signal that this is an async operator capable of receiving this many
            request(RxRingBuffer.SIZE);
        }

        @Override
        public void onNext(final T t) {
            try {
                // enqueue
                queue.onNext(t);
            } catch (MissingBackpressureException e) {
                // fail if the upstream has not obeyed our backpressure requests
                onError(e);
                return;
            }
            // attempt to schedule draining if needed
            schedule();
        }

        // the scheduling polling will then drain the queue and invoke `request(n)` to request more after draining
}

Many use cases will be able to use Observable.from, Observable.onBackpressureDrop and Observable.onBackpressureBuffer to achieve "reactive pull backpressure" without manually implementing Producer logic. Also, it is optional to make an Observable support backpressure. It can remain completely reactive and just push events as it always has. Most uses of RxJava this works just fine. If backpressure is needed then it can be migrated to use a Producer or several other approaches to flow control exist such as throttle, sample, debounce, window, buffer, onBackpressureBuffer, and onBackpressureDrop.

The wiki provides further documentation.

Relation to Reactive Streams

Contributors to RxJava are involved in defining the Reactive Streams spec. RxJava 1.0 is trying to comply with the semantic rules but is not attempting to comply with the type signatures. It will however have a separate module that acts as a bridge between the RxJava Observable and the Reactive Stream types.

The reasons for this are:

• Rx has Observer.onCompleted whereas Reactive Streams has onComplete. This is a massive breaking change to remove a "d".
• The RxJava Subscription is used also a "Closeable"/"Disposable" and it does not work well to make it now also be used for request(n), hence the separate type Producer in RxJava. It was attempted to reuse rx.Subscription but it couldn't be done without massive breaking changes.
• Reactive Streams uses onSubscribe(Subscription s) whereas RxJava injects the Subscription as the Subscriber. Again, this change could not be done without major breaking changes.
• RxJava 1.0 needs to be backwards compatible with the major Rx contracts established during the 0.x roadmap.

Considering these things, the major semantics of request(long n) for backpressure are compatible and this will allow interop with a bridge between the interfaces.

New Features

Compose/Transformer

The compose operator is similar to lift but allows custom operator implementations that are chaining Observable operators whereas lift is directly implementing the raw Subscriber logic.

Here is a trival example demonstrating how using compose is a better option than lift when existing Observable operators can be used to achieve the custom behavior.

import rx.Observable;
import rx.Observable.Operator;
import rx.Observable.Transformer;
import rx.Subscriber;

public class ComposeExample {

    public static void main(String[] args) {
        Observable.just("hello").compose(appendWorldTransformer()).forEach(System.out::println);
        Observable.just("hello").lift(appendWorldOperator()).forEach(System.out::println);
    }

    // if existing operators can be used, compose with Transformer is ideal
    private static Transformer<? super String, String> appendWorldTransformer() {
        return o -> o.map(s -> s + " world!").finallyDo(() -> {
            System.out.println("  some side-effect");
        });
    }

    // whereas lift is more low level
    private static Operator<? super String, String> appendWorldOperator() {
        return new Operator<String, String>() {

            @Override
            public Subscriber<? super String> call(Subscriber<? super String> child) {
                return new Subscriber<String>(child) {

                    @Override
                    public void onCompleted() {
                        child.onCompleted();
                    }

                    @Override
                    public void onError(Throwable e) {
                        child.onError(e);
                    }

                    @Override
                    public void onNext(String t) {
                        child.onNext(t + " world!");
                        System.out.println("  some side-effect");
                    }

                };
            }

        };
    }
}

retryWhen/repeatWhen

New operators retryWhen and repeatWhen were added which offer support for more advanced recursion such as retry with exponential backoff.

Here is an example that increases delay between each retry:

Observable.create((Subscriber<? super String> s) -> {
    System.out.println("subscribing");
    s.onError(new RuntimeException("always fails"));
}).retryWhen(attempts -> {
    return attempts.zipWith(Observable.range(1, 3), (n, i) -> i).flatMap(i -> {
        System.out.println("delay retry by " + i + " second(s)");
        return Observable.timer(i, TimeUnit.SECONDS);
    });
}).toBlocking().forEach(System.out::println);

Breaking Changes

The use of Producer has been added in such a way that it is optional and additive, but some operators that used to have unbounded queues are now bounded. This means that if a source Observable emits faster than the Observer can consume them, a MissingBackpressureException can be emitted via onError.

This semantic change can break existing code.

There are two ways of resolving this:

1. Modify the source Observable to use Producer and support backpressure.
2. Use newly added operators such as onBackpressureBuffer or onBackpressureDrop to choose a strategy for the source Observable of how to behave when it emits more data than the consuming Observer is capable of handling. Use of onBackpressureBuffer effectively returns it to having an unbounded buffer and behaving like version 0.19 or earlier.

Example:

sourceObservable.onBackpressureBuffer().subscribe(slowConsumer);

Deprecations

Various methods, operators or classes have been deprecated and will be removed in 1.0. Primarily they have been done to remove ambiguity, remove nuanced functionality that is easy to use wrong, clear out superfluous methods and eliminate cruft that was add...

0.20.0-RC6

Further fixes and enhancements bringing us close to completing 0.20.0 and almost ready for 1.0.

A more major change in this release is the deprecation of Observable.from(T). The full discussion can be seen in #1563.

• Pull 1575 combineLatest with backpressure
• Pull 1569 Compose/Transform Covariance
• Pull 1577 Fix the compose covariance
• Pull 1581 zip(Iterable) -> zipWith(Iterable)
• Pull 1582 Deprecate GroupedObservable.from
• Pull 1583 Redo/Repeat Backpressure
• Pull 1576 from(T) -> just(T)
• Pull 1545 Make Android ViewObservable.input observe TextView instead of String

Artifacts: Maven Central

0.20.0-RC5

Version 0.20.0-RC5 updates parallel, buffer(size), switchOnNext, repeat, and retry to support "reactive pull" backpressure. It adds a groupBy overload with an element selector, a new compose method as an alternative to lift for custom operators, fixes bugs and other general improvements.

There are still oustanding items being tracked for 0.20 that need to be completed for the final release.

• Pull 1573 Backpressure: parallel
• Pull 1572 Remove Timeout in Blocking Iterator
• Pull 1570 RxClojure: Fix for mapcat
• Pull 1568 Compose/Transformer
• Pull 1567 groupBy with element selector
• Pull 1565 Fixing Kotlin Defer
• Pull 1507 BufferWithSize with Backpressure Support
• Pull 1557 SwitchOnNext with backpressure support
• Pull 1562 TakeLastTimed with backpressure support
• Pull 1564 RxScala: Fix errors in Completeness.scala and also improve it
• Pull 1548 Adding backpressure to OnSubscribeRedo
• Pull 1559 More consistent hooks for scheduler plugins
• Pull 1561 Remove Variance on Defer
• Pull 1537 recursive scheduling in RxScala
• Pull 1560 flatMap overloads
• Pull 1558 mergeMap generics
• Pull 1552 Fixing a bug and a potential for other concurrency issues
• Pull 1555 RxScala: Add retryWhen/repeatWhen methods

Artifacts: Maven Central