简述

因为Android UI组件通常会频繁性的创建销毁，在搭配异步操作时，子线程持有UI组件引用，而子线程通常用来执行耗时操作，当子线程运行期间UI组件被销毁时，UI组件实例会因为被引用而无法被释放内存占用，所以很容易导致内存泄露。

之前分析过RxJava和Retrofit结合使用的源码，其中最后调用了compose(ObservableTransFormer)方法，传入lifecycleProvider.bindToLifecycle()实现和Android UI组件生命周期绑定，以解决UI组件内存泄漏问题，这篇文章就来分析一下是如何实现的。

源码分析

前面分析过，ViewModel中调用subscribe的是一个compose返回的Observable，我们就从这里开始分析：

@CheckReturnValue
@SchedulerSupport(SchedulerSupport.NONE)
public final <R> Observable<R> compose(ObservableTransformer<? super T, ? extends R> composer) {
    return wrap(((ObservableTransformer<T, R>) ObjectHelper.requireNonNull(composer, "composer is null")).apply(this));
}

wrap：

@CheckReturnValue
@SchedulerSupport(SchedulerSupport.NONE)
public static <T> Observable<T> wrap(ObservableSource<T> source) {
    ObjectHelper.requireNonNull(source, "source is null");
    if (source instanceof Observable) {
        return RxJavaPlugins.onAssembly((Observable<T>)source);
    }
    return RxJavaPlugins.onAssembly(new ObservableFromUnsafeSource<T>(source));
}

会返回一个ObservableFromUnsafeSource对象，当我们调用subscribe的时候就是调用它的subscribeActual：

@Override
protected void subscribeActual(Observer<? super T> observer) {
    source.subscribe(observer);
}

observer就是前面ViewModel传入的HttpObserver，source是((ObservableTransformer<T, R>) ObjectHelper.requireNonNull(composer, "composer is null")).apply(this)，composer也就是前面传入的lifecycle.bindToLifecycle()，lifecycle是啥，是LifecycleProvider接口的实例，那LifecycleProvider的实现类是什么呢？RxJava针对AndroidUI组件（Activity、Fragment）实现了继承自AppCompatActivity、Fragment等的LifecycleProvider的实现类RxAppCompatActivity、RxFragment等。

RxAppCompatActivity中的bindToLifecycle是：

private final BehaviorSubject<ActivityEvent> lifecycleSubject = BehaviorSubject.create();

@Override
@NonNull
@CheckResult
public final <T> LifecycleTransformer<T> bindToLifecycle() {
    return RxLifecycleAndroid.bindActivity(lifecycleSubject);
}

RxFragment中的是：

private final BehaviorSubject<FragmentEvent> lifecycleSubject = BehaviorSubject.create();

@Override
@NonNull
@CheckResult
public final <T> LifecycleTransformer<T> bindToLifecycle() {
    return RxLifecycleAndroid.bindFragment(lifecycleSubject);
}

BehaviorSubject.create：

@CheckReturnValue
@NonNull
public static <T> BehaviorSubject<T> create() {
    return new BehaviorSubject<T>();
}

再看bindActivity和bindFragment：

@NonNull
@CheckResult
public static <T> LifecycleTransformer<T> bindActivity(@NonNull final Observable<ActivityEvent> lifecycle) {
    return bind(lifecycle, ACTIVITY_LIFECYCLE);
}

@NonNull
@CheckResult
public static <T> LifecycleTransformer<T> bindFragment(@NonNull final Observable<FragmentEvent> lifecycle) {
    return bind(lifecycle, FRAGMENT_LIFECYCLE);
}

ACTIVITY_LIFECYCLE是：

private static final Function<ActivityEvent, ActivityEvent> ACTIVITY_LIFECYCLE =
    new Function<ActivityEvent, ActivityEvent>() {
        @Override
        public ActivityEvent apply(ActivityEvent lastEvent) throws Exception {
            switch (lastEvent) {
                case CREATE:
                    return ActivityEvent.DESTROY;
                case START:
                    return ActivityEvent.STOP;
                case RESUME:
                    return ActivityEvent.PAUSE;
                case PAUSE:
                    return ActivityEvent.STOP;
                case STOP:
                    return ActivityEvent.DESTROY;
                case DESTROY:
                    throw new OutsideLifecycleException("Cannot bind to Activity lifecycle when outside of it.");
                default:
                    throw new UnsupportedOperationException("Binding to " + lastEvent + " not yet implemented");
            }
        }
    };

FRAGMENT_LIFECYCLE是：

private static final Function<FragmentEvent, FragmentEvent> FRAGMENT_LIFECYCLE =
    new Function<FragmentEvent, FragmentEvent>() {
        @Override
        public FragmentEvent apply(FragmentEvent lastEvent) throws Exception {
            switch (lastEvent) {
                case ATTACH:
                    return FragmentEvent.DETACH;
                case CREATE:
                    return FragmentEvent.DESTROY;
                case CREATE_VIEW:
                    return FragmentEvent.DESTROY_VIEW;
                case START:
                    return FragmentEvent.STOP;
                case RESUME:
                    return FragmentEvent.PAUSE;
                case PAUSE:
                    return FragmentEvent.STOP;
                case STOP:
                    return FragmentEvent.DESTROY_VIEW;
                case DESTROY_VIEW:
                    return FragmentEvent.DESTROY;
                case DESTROY:
                    return FragmentEvent.DETACH;
                case DETACH:
                    throw new OutsideLifecycleException("Cannot bind to Fragment lifecycle when outside of it.");
                default:
                    throw new UnsupportedOperationException("Binding to " + lastEvent + " not yet implemented");
            }
        }
    };

bind方法：

@Nonnull
@CheckReturnValue
public static <T, R> LifecycleTransformer<T> bind(@Nonnull Observable<R> lifecycle,
                                                  @Nonnull final Function<R, R> correspondingEvents) {
    checkNotNull(lifecycle, "lifecycle == null");
    checkNotNull(correspondingEvents, "correspondingEvents == null");
    return bind(takeUntilCorrespondingEvent(lifecycle.share(), correspondingEvents));
}

private static <R> Observable<Boolean> takeUntilCorrespondingEvent(final Observable<R> lifecycle,
                                                                   final Function<R, R> correspondingEvents) {
    return Observable.combineLatest(
        lifecycle.take(1).map(correspondingEvents),
        lifecycle.skip(1),
        new BiFunction<R, R, Boolean>() {
            @Override
            public Boolean apply(R bindUntilEvent, R lifecycleEvent) throws Exception {
                return lifecycleEvent.equals(bindUntilEvent);
            }
        })
        .onErrorReturn(Functions.RESUME_FUNCTION)
        .filter(Functions.SHOULD_COMPLETE);
}

lifecycle.share()实际上调用的是其父类Observable的share方法：

@CheckReturnValue
@SchedulerSupport(SchedulerSupport.NONE)
public final Observable<T> share() {
    return publish().refCount();
}

最终返回的是ObservableRefCount：

@NonNull
@CheckReturnValue
@SchedulerSupport(SchedulerSupport.NONE)
public Observable<T> refCount() {
    return RxJavaPlugins.onAssembly(new ObservableRefCount<T>(onRefCount()));
}

封装的太多了，我们直接看filter返回的：

@CheckReturnValue
@SchedulerSupport(SchedulerSupport.NONE)
public final Observable<T> filter(Predicate<? super T> predicate) {
    ObjectHelper.requireNonNull(predicate, "predicate is null");
    return RxJavaPlugins.onAssembly(new ObservableFilter<T>(this, predicate));
}

最终bind(Observable)返回：

@Nonnull
@CheckReturnValue
public static <T, R> LifecycleTransformer<T> bind(@Nonnull final Observable<R> lifecycle) {
    return new LifecycleTransformer<>(lifecycle);
}

调用LifecycleTransformer的apply方法：

@Override
public ObservableSource<T> apply(Observable<T> upstream) {
    return upstream.takeUntil(observable);
}

upstream就是ObservableObserveOn，observable是takeUntilCorrespondingEvent返回的对象。所以看一下ObservableObserveOn的takeUntil(在父类Observable中)：

@CheckReturnValue
@SchedulerSupport(SchedulerSupport.NONE)
public final <U> Observable<T> takeUntil(ObservableSource<U> other) {
    ObjectHelper.requireNonNull(other, "other is null");
    return RxJavaPlugins.onAssembly(new ObservableTakeUntil<T, U>(this, other));
}

所以ObservableTakeUntil就是ObservableFromUnsafeSource的source，那么在其subscribeActual中就是调用了ObservableTakeUntil的subscribeActual:

@Override
public void subscribeActual(Observer<? super T> child) {
    TakeUntilMainObserver<T, U> parent = new TakeUntilMainObserver<T, U>(child);
    child.onSubscribe(parent);

    other.subscribe(parent.otherObserver);
    source.subscribe(parent);
}

child是自定义的HttpObserver，根据ObservableTakeUntil构造方法追溯到source就是ObservableObserveOn，other是takeUntilCorrespondingEvent返回的对象。

首先看other.subscribe方法，就是调用ObservableFilter的subscribeActual方法，parent.otherObserver是一个new OtherObserver()：

@Override
public void subscribeActual(Observer<? super T> observer) {
    source.subscribe(new FilterObserver<T>(observer, predicate));
}

source是调用filter方法的ObservableOnErrorReturn：

@Override
public void subscribeActual(Observer<? super T> t) {
    source.subscribe(new OnErrorReturnObserver<T>(t, valueSupplier));
}

再往上就是：

Observable.combineLatest(
    lifecycle.take(1).map(correspondingEvents),
    lifecycle.skip(1),
    new BiFunction<R, R, Boolean>() {
        @Override
        public Boolean apply(R bindUntilEvent, R lifecycleEvent) throws Exception {
            return lifecycleEvent.equals(bindUntilEvent);
        }
    })

@CheckReturnValue
@NonNull
@SchedulerSupport(SchedulerSupport.NONE)
public static <T, R> Observable<R> combineLatest(ObservableSource<? extends T>[] sources,
        Function<? super Object[], ? extends R> combiner, int bufferSize) {
    ObjectHelper.requireNonNull(sources, "sources is null");
    if (sources.length == 0) {
        return empty();
    }
    ObjectHelper.requireNonNull(combiner, "combiner is null");
    ObjectHelper.verifyPositive(bufferSize, "bufferSize");

    // the queue holds a pair of values so we need to double the capacity
    int s = bufferSize << 1;
    return RxJavaPlugins.onAssembly(new ObservableCombineLatest<T, R>(sources, null, combiner, s, false));
}

ObservableCombineLatest的subscribeActual：

@Override
@SuppressWarnings("unchecked")
public void subscribeActual(Observer<? super R> observer) {
    ObservableSource<? extends T>[] sources = this.sources;
    int count = 0;
    if (sources == null) {
        sources = new ObservableSource[8];
        for (ObservableSource<? extends T> p : sourcesIterable) {
            if (count == sources.length) {
                ObservableSource<? extends T>[] b = new ObservableSource[count + (count >> 2)];
                System.arraycopy(sources, 0, b, 0, count);
                sources = b;
            }
            sources[count++] = p;
        }
    } else {
        count = sources.length;
    }

    if (count == 0) {
        EmptyDisposable.complete(observer);
        return;
    }

    LatestCoordinator<T, R> lc = new LatestCoordinator<T, R>(observer, combiner, count, bufferSize, delayError);
    lc.subscribe(sources);
}

整理一下它持有的属性，combiner是：

static final class Array2Func<T1, T2, R> implements Function<Object[], R> {
    final BiFunction<? super T1, ? super T2, ? extends R> f;

    Array2Func(BiFunction<? super T1, ? super T2, ? extends R> f) {
        this.f = f;
    }

    @SuppressWarnings("unchecked")
    @Override
    public R apply(Object[] a) throws Exception {
        if (a.length != 2) {
            throw new IllegalArgumentException("Array of size 2 expected but got " + a.length);
        }
        return f.apply((T1)a[0], (T2)a[1]);
    }
}

它的f是：

new BiFunction<R, R, Boolean>() {
    @Override
    public Boolean apply(R bindUntilEvent, R lifecycleEvent) throws Exception {
        return lifecycleEvent.equals(bindUntilEvent);
    }
}

sources有两个，一个是ObservableMap（通过lifecycle.take(1).map(correspondingEvents)获得），一个是ObservableSkip(通过lifecycle.skip(1)获得)。

最后会走到LatestCoordinator.subscribe:

public void subscribe(ObservableSource<? extends T>[] sources) {
    Observer<T>[] as = observers;
    int len = as.length;
    downstream.onSubscribe(this);
    for (int i = 0; i < len; i++) {
        if (done || cancelled) {
            return;
        }
        sources[i].subscribe(as[i]);
    }
}

downstream.onSubscribe会沿着调用链最终调用HttpObserver中的onSubscribe里，这里就不贴了，主要看一下sources[i].subscribe(as[i])。as是在构造时生成的一个同sources一样长度的数组：

CombinerObserver<T, R>[] as = new CombinerObserver[count];
for (int i = 0; i < count; i++) {
    as[i] = new CombinerObserver<T, R>(this, i);
}

先看第一个source，ObservableMap：

@Override
public void subscribeActual(Observer<? super U> t) {
    source.subscribe(new MapObserver<T, U>(t, function));
}

它的source是ObservableTake：

@Override
protected void subscribeActual(Observer<? super T> observer) {
    source.subscribe(new TakeObserver<T>(observer, limit));
}

而它的source又是ObservableRefCount：

@Override
protected void subscribeActual(Observer<? super T> observer) {

    RefConnection conn;

    boolean connect = false;
    synchronized (this) {
        conn = connection;
        if (conn == null) {
            conn = new RefConnection(this);
            connection = conn;
        }

        long c = conn.subscriberCount;
        if (c == 0L && conn.timer != null) {
            conn.timer.dispose();
        }
        conn.subscriberCount = c + 1;
        if (!conn.connected && c + 1 == n) {
            connect = true;
            conn.connected = true;
        }
    }

    source.subscribe(new RefCountObserver<T>(observer, this, conn));

    if (connect) {
        source.connect(conn);
    }
}

它的source是ObservablePublish:

@Override
protected void subscribeActual(Observer<? super T> observer) {
    onSubscribe.subscribe(observer);
}

onSubscribe是PublishSource：

@Override
public void subscribe(Observer<? super T> child) {
    // create the backpressure-managing producer for this child
    InnerDisposable<T> inner = new InnerDisposable<T>(child);
    child.onSubscribe(inner);
    // concurrent connection/disconnection may change the state,
    // we loop to be atomic while the child subscribes
    for (;;) {
        // get the current subscriber-to-source
        PublishObserver<T> r = curr.get();
        // if there isn't one or it is disposed
        if (r == null || r.isDisposed()) {
            // create a new subscriber to source
            PublishObserver<T> u = new PublishObserver<T>(curr);
            // let's try setting it as the current subscriber-to-source
            if (!curr.compareAndSet(r, u)) {
                // didn't work, maybe someone else did it or the current subscriber
                // to source has just finished
                continue;
            }
            // we won, let's use it going onwards
            r = u;
        }

        /*
         * Try adding it to the current subscriber-to-source, add is atomic in respect
         * to other adds and the termination of the subscriber-to-source.
         */
        if (r.add(inner)) {
            inner.setParent(r);
            break; // NOPMD
        }
        /*
         * The current PublishObserver has been terminated, try with a newer one.
         */
        /*
         * Note: although technically correct, concurrent disconnects can cause
         * unexpected behavior such as child observers never receiving anything
         * (unless connected again). An alternative approach, similar to
         * PublishSubject would be to immediately terminate such child
         * observers as well:
         *
         * Object term = r.terminalEvent;
         * if (r.nl.isCompleted(term)) {
         *     child.onComplete();
         * } else {
         *     child.onError(r.nl.getError(term));
         * }
         * return;
         *
         * The original concurrent behavior was non-deterministic in this regard as well.
         * Allowing this behavior, however, may introduce another unexpected behavior:
         * after disconnecting a previous connection, one might not be able to prepare
         * a new connection right after a previous termination by subscribing new child
         * observers asynchronously before a connect call.
         */
    }
}

很遗憾，走到这里就结束了，并没有发现和我们主题相关的验证代码，我们得再回到ObservableRefCount的subscribeActual，下面会执行source.connect(conn)，这里的source是ObservablePublish，它的connect方法如下：

@Override
public void connect(Consumer<? super Disposable> connection) {
    boolean doConnect;
    PublishObserver<T> ps;
    // we loop because concurrent connect/disconnect and termination may change the state
    for (;;) {
        // retrieve the current subscriber-to-source instance
        ps = current.get();
        // if there is none yet or the current has been disposed
        if (ps == null || ps.isDisposed()) {
            // create a new subscriber-to-source
            PublishObserver<T> u = new PublishObserver<T>(current);
            // try setting it as the current subscriber-to-source
            if (!current.compareAndSet(ps, u)) {
                // did not work, perhaps a new subscriber arrived
                // and created a new subscriber-to-source as well, retry
                continue;
            }
            ps = u;
        }
        // if connect() was called concurrently, only one of them should actually
        // connect to the source
        doConnect = !ps.shouldConnect.get() && ps.shouldConnect.compareAndSet(false, true);
        break; // NOPMD
    }
    /*
     * Notify the callback that we have a (new) connection which it can dispose
     * but since ps is unique to a connection, multiple calls to connect() will return the
     * same Disposable and even if there was a connect-disconnect-connect pair, the older
     * references won't disconnect the newer connection.
     * Synchronous source consumers have the opportunity to disconnect via dispose on the
     * Disposable as subscribe() may never return in its own.
     *
     * Note however, that asynchronously disconnecting a running source might leave
     * child observers without any terminal event; PublishSubject does not have this
     * issue because the dispose() was always triggered by the child observers
     * themselves.
     */
    try {
        connection.accept(ps);
    } catch (Throwable ex) {
        Exceptions.throwIfFatal(ex);
        throw ExceptionHelper.wrapOrThrow(ex);
    }
    if (doConnect) {
        source.subscribe(ps);
    }
}

最终会走到source.subscribe(ps)，source是调用share()的那个对象，也就是BehaviorSubject：

@Override
protected void subscribeActual(Observer<? super T> observer) {
    BehaviorDisposable<T> bs = new BehaviorDisposable<T>(observer, this);
    observer.onSubscribe(bs);
    if (add(bs)) {
        if (bs.cancelled) {
            remove(bs);
        } else {
            bs.emitFirst();
        }
    } else {
        Throwable ex = terminalEvent.get();
        if (ex == ExceptionHelper.TERMINATED) {
            observer.onComplete();
        } else {
            observer.onError(ex);
        }
    }
}

关键代码bs.emitFirst()：

void emitFirst() {
    if (cancelled) {
        return;
    }
    Object o;
    synchronized (this) {
        if (cancelled) {
            return;
        }
        if (next) {
            return;
        }

        BehaviorSubject<T> s = state;
        Lock lock = s.readLock;

        lock.lock();
        index = s.index;
        o = s.value.get();
        lock.unlock();

        emitting = o != null;
        next = true;
    }

    if (o != null) {
        if (test(o)) {
            return;
        }

        emitLoop();
    }
}

关键代码test(o)，往下看:

@Override
public boolean test(Object o) {
    return cancelled || NotificationLite.accept(o, downstream);
}

NotificationLite.accept(o, downstream)中：

@SuppressWarnings("unchecked")
public static <T> boolean accept(Object o, Observer<? super T> observer) {
    if (o == COMPLETE) {
        observer.onComplete();
        return true;
    } else
    if (o instanceof ErrorNotification) {
        observer.onError(((ErrorNotification)o).e);
        return true;
    }
    observer.onNext((T)o);
    return false;
}

observer.onNext((T)o)，observer就是ps，ps就是PublishObserver：

@Override
public void onNext(T t) {
    for (InnerDisposable<T> inner : observers.get()) {
        inner.child.onNext(t);
    }
}

observers是什么时候赋值的呢，就是在source.connect之前的subscribe的调用中：

if (r.add(inner)) {
    inner.setParent(r);
    break; // NOPMD
}

最终经过RefCountObserver、TakeObserver、MapObserver会回调到MapObserver的onNext中：

@Override
public void onNext(T t) {
    if (done) {
        return;
    }

    if (sourceMode != NONE) {
        downstream.onNext(null);
        return;
    }

    U v;

    try {
        v = ObjectHelper.requireNonNull(mapper.apply(t), "The mapper function returned a null value.");
    } catch (Throwable ex) {
        fail(ex);
        return;
    }
    downstream.onNext(v);
}

根据ObservableMap的subscribeActual中MapObserver的构造方法可知mapper就是function，function就是前面构造ObservableMap时map方法的参数correspondingEvents，correspondingEvents就是ACTIVITY_LIFECYCLE或FRAGMENT_LIFECYCLE，所以把t代入到他们的apply方法中，test(o)的o是什么，发现它来自this.value，this.value在setCurrent中通过value.lazySet(o)存值，setCurrent在BehaviorSubject的onNext中调用，onNext又在哪调用呢？看一下RxAppCompatActivity：

public abstract class RxAppCompatActivity extends AppCompatActivity implements LifecycleProvider<ActivityEvent> {

    private final BehaviorSubject<ActivityEvent> lifecycleSubject = BehaviorSubject.create();

    @Override
    @NonNull
    @CheckResult
    public final Observable<ActivityEvent> lifecycle() {
        return lifecycleSubject.hide();
    }

    @Override
    @NonNull
    @CheckResult
    public final <T> LifecycleTransformer<T> bindUntilEvent(@NonNull ActivityEvent event) {
        return RxLifecycle.bindUntilEvent(lifecycleSubject, event);
    }

    @Override
    @NonNull
    @CheckResult
    public final <T> LifecycleTransformer<T> bindToLifecycle() {
        return RxLifecycleAndroid.bindActivity(lifecycleSubject);
    }

    @Override
    @CallSuper
    protected void onCreate(@Nullable Bundle savedInstanceState) {
        super.onCreate(savedInstanceState);
        lifecycleSubject.onNext(ActivityEvent.CREATE);
    }

    @Override
    @CallSuper
    protected void onStart() {
        super.onStart();
        lifecycleSubject.onNext(ActivityEvent.START);
    }

    @Override
    @CallSuper
    protected void onResume() {
        super.onResume();
        lifecycleSubject.onNext(ActivityEvent.RESUME);
    }

    @Override
    @CallSuper
    protected void onPause() {
        lifecycleSubject.onNext(ActivityEvent.PAUSE);
        super.onPause();
    }

    @Override
    @CallSuper
    protected void onStop() {
        lifecycleSubject.onNext(ActivityEvent.STOP);
        super.onStop();
    }

    @Override
    @CallSuper
    protected void onDestroy() {
        lifecycleSubject.onNext(ActivityEvent.DESTROY);
        super.onDestroy();
    }
}

所以伴随着Activity或者Fragment的生命周期方法回调，对应的BehaviorSubject会调用onNext方法：

@Override
public void onNext(T t) {
    ObjectHelper.requireNonNull(t, "onNext called with null. Null values are generally not allowed in 2.x operators and sources.");

    if (terminalEvent.get() != null) {
        return;
    }
    Object o = NotificationLite.next(t);
    setCurrent(o);
    for (BehaviorDisposable<T> bs : subscribers.get()) {
        bs.emitNext(o, index);
    }
}

setCurrent的意义也就是保存当前Activity的状态，所以当发送请求的时候，会通过apply方法返回一个Event，然后调用downstream.onNext(event)，downstream就是CombinerObserver，它的onNext是：

@Override
public void onNext(T t) {
    parent.innerNext(index, t);
}

构造的时候传入的parent是LatestCoordinator.this，所以看一下LatestCoordinator的innerNext:

void innerNext(int index, T item) {
    boolean shouldDrain = false;
    synchronized (this) {
          //初始值空数组，但不是null
        Object[] latest = this.latest;
        if (latest == null) {
            return;
        }
        Object o = latest[index];
        int a = active;
        if (o == null) {
            active = ++a;
        }
          //若是第一次，此时active == a == 1，latest.length == 2
        latest[index] = item;
        if (a == latest.length) {
              //入队列
            queue.offer(latest.clone());
            shouldDrain = true;
        }
    }
      //drain
    if (shouldDrain) {
        drain();
    }
}

所以因为第一次innerNext的时候latest.length是2（初始化就是和sources长度一致，这里是两个source）而a此时是1，也就是ObservableMap.subscribe的时候是不会进行drain()的，当ObservableSkip调用subscribe第二次进来的时候才会走到drain()。这里记住this.latest有了第一个值，是ACTIVITY_LIFECYCLE的apply返回的值（以Activity为例）。

ObservableSkip的subscribeActual：

@Override
public void subscribeActual(Observer<? super T> observer) {
    source.subscribe(new SkipObserver<T>(observer, n));
}

直接调用其source（ObservableRefCount）的subscribeActual，流程和之前的ObservableMap一样，最终回到CombinerObserver，和之前相比，少了MapObserver的mapper（也就是ACTIVITY_LIFECYCLE或FRAGMENT_LIFECYCLE）的apply操作，所以此时的innerNext中的item就是生命周期组件当前的event。

drain操作：

void drain() {
    if (getAndIncrement() != 0) {
        return;
    }

    final SpscLinkedArrayQueue<Object[]> q = queue;
    final Observer<? super R> a = downstream;
    final boolean delayError = this.delayError;

    int missed = 1;
    for (;;) {

        for (;;) {
            if (cancelled) {
                clear(q);
                return;
            }

            if (!delayError && errors.get() != null) {
                cancelSources();
                clear(q);
                a.onError(errors.terminate());
                return;
            }

            boolean d = done;
            Object[] s = q.poll();
            boolean empty = s == null;

            if (d && empty) {
                clear(q);
                Throwable ex = errors.terminate();
                if (ex == null) {
                    a.onComplete();
                } else {
                    a.onError(ex);
                }
                return;
            }

            if (empty) {
                break;
            }

            R v;

            try {
                v = ObjectHelper.requireNonNull(combiner.apply(s), "The combiner returned a null value");
            } catch (Throwable ex) {
                Exceptions.throwIfFatal(ex);
                errors.addThrowable(ex);
                cancelSources();
                clear(q);
                ex = errors.terminate();
                a.onError(ex);
                return;
            }

            a.onNext(v);
        }

        missed = addAndGet(-missed);
        if (missed == 0) {
            break;
        }
    }
}

看到前面的验证通过后有一句combiner.apply(s)，combiner是前面的Functions.toFunction(combiner)，Functions.toFunction(combiner)返回的是Array2Func：

static final class Array2Func<T1, T2, R> implements Function<Object[], R> {
    final BiFunction<? super T1, ? super T2, ? extends R> f;

    Array2Func(BiFunction<? super T1, ? super T2, ? extends R> f) {
        this.f = f;
    }

    @SuppressWarnings("unchecked")
    @Override
    public R apply(Object[] a) throws Exception {
        if (a.length != 2) {
            throw new IllegalArgumentException("Array of size 2 expected but got " + a.length);
        }
        return f.apply((T1)a[0], (T2)a[1]);
    }
}

持有的f是Observable.combineLatest()方法的第三个参数，传入的是：

new BiFunction<R, R, Boolean>() {
    @Override
    public Boolean apply(R bindUntilEvent, R lifecycleEvent) throws Exception {
        return lifecycleEvent.equals(bindUntilEvent);
    }
})

a.onNext(v)往下传到了FilterObserver中的onNext中：

@Override
public void onNext(T t) {
    if (sourceMode == NONE) {
        boolean b;
        try {
            b = filter.test(t);
        } catch (Throwable e) {
            fail(e);
            return;
        }
        if (b) {
            downstream.onNext(t);
        }
    } else {
        downstream.onNext(null);
    }
}

filter为：

static final Predicate<Boolean> SHOULD_COMPLETE = new Predicate<Boolean>() {
    @Override
    public boolean test(Boolean shouldComplete) throws Exception {
        return shouldComplete;
    }
};

没有额外操作直接返回apply()传进来的参数。

这个时候如果apply判断为false，则说明当前状态还不是需要被取消的状态，则到此这个调用链就结束了，还记得调用开始的地方吗？就是ObservableTakeUntil的subscribeActual中的other.subscribe(parent.otherObserver)：

@Override
public void subscribeActual(Observer<? super T> child) {
    TakeUntilMainObserver<T, U> parent = new TakeUntilMainObserver<T, U>(child);
    child.onSubscribe(parent);

    other.subscribe(parent.otherObserver);
    source.subscribe(parent);
}

接下来就会去执行下面的source.subscribe(parent)了，就是正常订阅接收回调的过程了。

如果apply判断为true，则说明当前状态是需要被取消订阅的状态了，上面的代码也就是b是true，就会执行downstream.onNext(t)操作，这个downstream就是上面的parent.otherObserver，parent是TakeUntilMainObserver，otherOnserver在它的构造方法里创建：

TakeUntilMainObserver(Observer<? super T> downstream) {
    this.downstream = downstream;
    this.upstream = new AtomicReference<Disposable>();
    this.otherObserver = new OtherObserver();
    this.error = new AtomicThrowable();
}

所以downstream.onNext(t)就是OtherObserver的onNext方法：

@Override
public void onNext(U t) {
    DisposableHelper.dispose(this);
    otherComplete();
}

void otherComplete() {
    DisposableHelper.dispose(upstream);
    HalfSerializer.onComplete(downstream, this, error);
}

可以看到，在这里使用dispose取消了订阅。

总结

通过前面的分析，我们可以得到这样的结论：

• compose创建了一个ObservableFromUnsafeSource对象，我们调用它的subscribe方法开启一个异步调用；
• 它的subscribe方法实际上调用的是((ObservableTransformer<T, R>) ObjectHelper.requireNonNull(composer, "composer is null")).apply(this)的这个对象的subscribe方法，也就是composer的subscribe方法；
• composer是通过lifecycle.bindToLifecycle()创建的，lifecycle对应RxAppCompatActivity，bindToLifecycle()里通过RxLifecycleAndroid.bindActivity(lifecycleSubject)得到LifecycleTransformer<>(lifecycle)；
• LifecycleTransformer的apply返回一个ObservableTakeUntil对象，它的source就是compose的调用者ObservableObserveOn，other就是RxLifecycleAndroid.bindActivity的调用链最终返回的ObservableFilter对象；
• ObservableFilter的调用链是：ObservableCombineLatest->ObservableOnErrorReturn->ObservableFilter；
• ObservableFromUnsafeSource的subscribeActual中source就是ObservableTakeUntil，它的subsribeActual中先后调用了other.subscribe和source.subscribe(parent)；
• other.subscribe可以理解为验证当前订阅是否还在有效生命周期范围内，如果不在则进行dispose操作；
• source.subscribe(parent)在执行过程中如果检查到已经dispose了则结束调用，从而释放UI组件引用，达到解决内存泄露的目的。