本篇文章已授权微信公众号
YYGeeker
独家发布转载请标明出处
CSDN学院课程地址
- RxJava2从入门到精通-初级篇:https://edu.csdn.net/course/detail/10036
- RxJava2从入门到精通-中级篇:https://edu.csdn.net/course/detail/10037
- RxJava2从入门到精通-进阶篇:https://edu.csdn.net/course/detail/10038
- RxJava2从入门到精通-源码分析篇:https://edu.csdn.net/course/detail/10138
10. RxJava源码分析
RxJava源码分析最主要的点在于
- RxJava是如何从事件流的发送方发送到事件流的接收方的
- RxJava是如何对操作符进行封装和操作的
- RxJava是如何随意切换线程的
在分析的过程中,部分源码分析我们会通过手写RxJava的部分代码进行分析,当然也会结合实际RxJava的代码进行分析。其中,手写RxJava的原因是为了简化源代码,让读者方便阅读到主要代码,更快的看懂RxJava的实现思路。在阅读源码之前,我们需要对RxJava的大体概念进行简单的梳理
- 发射器:Emitter,发射数据的对象
- 被观察者:Observable,被观察的对象
- 观察者:Observer,观察的对象
- 被观察者被订阅时:ObservableOnSubscribe,被订阅时的回调,同时创建出发射器
- 释放者:Disposable,释放RxJava的对象
RxJava的分析三步骤
- 创建:被观察者创建的过程
- 订阅:被观察者订阅观察者的过程
- 发射:发射器发射的过程
RxJava原理图解
- 第一排表示各个对象的创建关系,A->B->C->D
- 第二排表示各个对象的订阅关系,D->C->B->A
- 第三排表示各个对象的发射关系,A->B->C->D
10.1 RxJava的事件发射原理
知识点:
- 理解发射数据的过程
- 理解接收数据的过程
以下是手写RxJava的代码
Observable.create(new ObservableOnSubscribe<String>() {
@Override
public void subscribe(Emitter<String> emitter) {
emitter.onNext("Hello RxJava");
emitter.onError();
emitter.onNext("Hello RxJava");
}
}).subscribe(new Observabler<String>() {
@Override
public void onSubscribe() {
}
@Override
public void onNext(String string) {
System.out.println("onNext=" + string);
}
@Override
public void onError() {
System.out.println("onError");
}
@Override
public void onComplete() {
}
});
输出结果:在输出onError后,就不会继续收到新的事件流,表示事件已经被释放了
onNext=Hello RxJava
onError
1、定义接口
发射器
public interface Emitter<T> {
void onNext(T t);
void onError();
}
观察者
public interface Observer<T> {
void onSubscribe();
void onNext(T t);
void onError();
void onComplete();
}
被观察者被订阅时
public interface ObservableOnSubscribe<T> {
void subscribe(Emitter<T> emitter);
}
2、实现被观察者
被观察者Observable负责创建、订阅,发射由发射器负责
- 创建:创建的过程只是将传递进来的参数交给新的ObservableCreate进行管理
- 订阅:订阅的过程只是实现创建出来的ObservableCreate的subscribeActual方法
public abstract class Observable<T> {
public static <T> ObservableCreate create(ObservableOnSubscribe<T> observableOnSubscribe) {
return new ObservableCreate<T>(observableOnSubscribe);
}
public void subscribe(Observer<T> observer) {
subscribeActual(observer);
}
public abstract void subscribeActual(Observer<T> observer);
}
3、ObservableCreate
ObservableCreate继承自Observable,由于Observable.create返回当前ObservableCreate,所以在subscribe的时候,走的是这里的subscribeActual,subscribeActual中会去创建发射器,并给发射器传递进去observer
public class ObservableCreate<T> extends Observable{
private ObservableOnSubscribe source;
public ObservableCreate(ObservableOnSubscribe observableOnSubscribe) {
this.source = observableOnSubscribe;
}
@Override
public void subscribeActual(Observer observer) {
//固定的三步曲分析法(个人创建,基本都是这个步骤)
//1、创建发射器
EmitterCreate<T> emitterCreate = new EmitterCreate<>(observer);
//2、回调observer的onSubscribe
observer.onSubscribe();
//3、回调上一个的subscribe
source.subscribe(emitterCreate);
}
}
4、EmitterCreate
传递进来的observer即是我们最开始订阅时候new出来的,此时发射数据,就会去调用Observer的onNext方法,这样数据就从发射器中传递到观察者中了。DisposableHelper在后面会讲到,这里只是用作判断是否被释放的一个工具类
public class EmitterCreate<T>
extends AtomicReference<Disposable>
implements Emitter<T>, Disposable {
private Observer<T> observer;
public EmitterCreate(Observer<T> observer) {
this.observer = observer;
}
@Override
public void onNext(T t) {
if (!isDisposed()) {
observer.onNext(t);
}
}
@Override
public void onError() {
if (!isDisposed()) {
try {
observer.onError();
} finally {
dispose();
}
}
}
@Override
public void dispose() {
DisposableHelper.dispose(this);
}
@Override
public boolean isDisposed() {
return DisposableHelper.isDisposed(get());
}
}
以下是RxJava源代码
1、Observable.create
public static <T> Observable<T> create(ObservableOnSubscribe<T> source) {
ObjectHelper.requireNonNull(source, "source is null");//判空
return RxJavaPlugins.onAssembly(new ObservableCreate<T>(source));//返回自身
}
RxJavaPlugins.onAssembly只是对传递进来的参数做判断处理,最终还是返回ObservableCreate,有关RxJavaPlugins的东西最终都是返回自身,RxJavaPlugins后面分析会说到,这里只需要知道他是返回参数本身即可
2、Observable.subscribe
public final void subscribe(Observer<? super T> observer) {
ObjectHelper.requireNonNull(observer, "observer is null");//判空
try {
observer = RxJavaPlugins.onSubscribe(this, observer);//返回自身
ObjectHelper.requireNonNull(observer, "Plugin returned null Observer");
subscribeActual(observer);//回调ObservableCreate的subscribeActual
} catch (NullPointerException e) { // NOPMD
throw e;
} catch (Throwable e) {
......
throw npe;
}
}
observable.subscribe和我们手写代码一样,最终调用的是ObservableCreate的subscribeActual方法
3、ObservableCreate
public final class ObservableCreate<T> extends Observable<T> {
final ObservableOnSubscribe<T> source;
public ObservableCreate(ObservableOnSubscribe<T> source) {
this.source = source;
}
@Override
protected void subscribeActual(Observer<? super T> observer) {
//1、创建发射器
CreateEmitter<T> parent = new CreateEmitter<T>(observer);
//2、回调observer的onSubscribe
observer.onSubscribe(parent);
try {
//3、回调ObservableOnSubscribe的subscribe
source.subscribe(parent);
} catch (Throwable ex) {
Exceptions.throwIfFatal(ex);
parent.onError(ex);
}
}
static final class CreateEmitter<T>
extends AtomicReference<Disposable>
implements ObservableEmitter<T>, Disposable {
private static final long serialVersionUID = -3434801548987643227L;
final Observer<? super T> observer;
CreateEmitter(Observer<? super T> observer) {
this.observer = observer;
}
@Override
public void onNext(T t) {
if (t == null) {
onError(new NullPointerException("onNext called with null. Null values are generally not allowed in 2.x operators and sources."));
return;
}
if (!isDisposed()) {
observer.onNext(t);
}
}
@Override
public void onError(Throwable t) {
if (!tryOnError(t)) {
RxJavaPlugins.onError(t);
}
}
@Override
public boolean tryOnError(Throwable t) {
if (t == null) {
t = new NullPointerException("onError called with null. Null values are generally not allowed in 2.x operators and sources.");
}
if (!isDisposed()) {
try {
observer.onError(t);
} finally {
dispose();
}
return true;
}
return false;
}
@Override
public void onComplete() {
if (!isDisposed()) {
try {
observer.onComplete();
} finally {
dispose();
}
}
}
@Override
public void setDisposable(Disposable d) {
DisposableHelper.set(this, d);
}
@Override
public void setCancellable(Cancellable c) {
setDisposable(new CancellableDisposable(c));
}
@Override
public ObservableEmitter<T> serialize() {
return new SerializedEmitter<T>(this);
}
@Override
public void dispose() {
DisposableHelper.dispose(this);
}
@Override
public boolean isDisposed() {
return DisposableHelper.isDisposed(get());
}
}
}
ObservableCreate和我们手写代码一样,创建发射器,并在发射器中做发射数据等操作
小结
如图所示
10.2 RxJava的事件释放原理
知识点:
- 理解释放事件的原理
有关RxJava的释放原理是基于
Observable
可以返回
Disposable
对象,只有调用
dispose()
才能释放事件,通过上面的例子,我们知道在发射器里面有
isDisposed()
和
dispose()
操作,在发射完
onError
事件的情况下,我们会将事件释放,所以在finally会做释放操作,防止后面的事件再次发射
以下是手写RxJava的代码
public class EmitterCreate<T>
extends AtomicReference<Disposable>
implements Emitter<T>, Disposable {
private Observer<T> observer;
public EmitterCreate(Observer<T> observer) {
this.observer = observer;
}
@Override
public void onNext(T t) {
if (!isDisposed()) {
observer.onNext(t);
}
}
@Override
public void onError() {
if (!isDisposed()) {
try {
observer.onError();
} finally {
dispose();
}
}
}
@Override
public void dispose() {
DisposableHelper.dispose(this);
}
@Override
public boolean isDisposed() {
return DisposableHelper.isDisposed(get());
}
}
以下是RxJava源代码
@Override
public void dispose() {
DisposableHelper.dispose(this);
}
@Override
public boolean isDisposed() {
return DisposableHelper.isDisposed(get());
}
可以发现事件的释放都是通过
DisposableHelper
去触发的,不管是手写RxJava还是源代码,释放RxJava都是通过
DisposableHelper
进行释放,具体看
DisposableHelper
。在我们的演示程序中,我们通过发射
onNext->onError->onNext
的过程,去挖掘事件是怎么被释放掉的
public enum DisposableHelper implements Disposable {
DISPOSED
;
public static boolean isDisposed(Disposable d) {
return d == DISPOSED;
}
public static boolean dispose(AtomicReference<Disposable> field) {
Disposable current = field.get();//获取参数的Disposable对象
Disposable d = DISPOSED;//声明一个已经释放的Disposable对象
if (current != d) {//如果当前未被释放
current = field.getAndSet(d);//则将当前的Disposable赋值成已经释放过的Disposable对象
if (current != d) {//如果当前还未被释放
if (current != null) {//且不为空
current.dispose();//则释放当前Disposable对象
}
return true;
}
}
return false;
}
}
在事件释放的过程中,
EmitterCreate
本身是个
AtomicReference<Disposable>
,代码通过
get()
去获取
Disposable
对象,其中代码会通过双层判断去做释放,防止在多线程的时候出现抢夺的情况
- onNext:第一次发射数据时,
会获取一个null对象,所以不符合get()
d == DISPOSED
- onEroor:这时候会调用
去比较当前和释放过的对象,如果不等于,则将当前的对象设置为释放过的值dispose()
- onNext:第二次发射数据时,
会获取一个已经释放过的对象,这个时候符合get()
d == DISPOSED
其实这里的操作如同设置一个Flag,但由于
Disposable
是对象的形式,且需要保证原子性,
AtomicReference
类型是个最佳选择,能保证对象的原子性
10.3 RxJava的背压原理
知识点:
- 理解背压实现的本质
- 理解背压数据项丢弃的本质
背压原理有一部分和RxJava事件发射原理相似,其背压的过程就是在不同策略的发射器去处理当前的数据项而已。在分析背压策略的时候,我们都知道背压是需要手动进行请求才可以将数据发射到观察者中,所以我们会调用
s.request(Long.MAX_VALUE)
让观察者能接收到数据。有些人就会有疑问,为什么有些人平时用背压的时候,不需要去调用
request()
就能接收到数据,原因是有些背压已经在内部默认调用了
s.request(Long.MAX_VALUE)
,所以这里是不用多想的,是一定要调用
s.request(Long.MAX_VALUE)
才能收到数据的。由于不同背压的策略的原理大同小异,主要以Drop策略去分析背压的原理
public static void drop(View view) {
Flowable.create(new FlowableOnSubscribe<Integer>() {
@Override
public void subscribe(FlowableEmitter<Integer> emitter) throws Exception {
for (int i = 0; i < 1000; i++) {
emitter.onNext(i);
}
}
}, BackpressureStrategy.DROP)
.subscribeOn(Schedulers.io())
.observeOn(AndroidSchedulers.mainThread())
.subscribe(new FlowableSubscriber<Integer>() {
@Override
public void onSubscribe(Subscription s) {
s.request(Long.MAX_VALUE);
}
@Override
public void onNext(Integer integer) {
Log.e("TAG", "onNext=" + integer);
}
@Override
public void onError(Throwable t) {
t.printStackTrace();
}
@Override
public void onComplete() {
}
});
}
以下是RxJava源代码
1、Flowable.create
public static <T> Flowable<T> create(FlowableOnSubscribe<T> source, BackpressureStrategy mode) {
ObjectHelper.requireNonNull(source, "source is null");//判空
ObjectHelper.requireNonNull(mode, "mode is null");//判空
return RxJavaPlugins.onAssembly(new FlowableCreate<T>(source, mode));//返回自身
}
Flowable.create跟我们前面是一样的,最后还是会交给新对象FlowableCreate去处理
2、Flowable.subscribe
public final void subscribe(FlowableSubscriber<? super T> s) {
ObjectHelper.requireNonNull(s, "s is null");
try {
Subscriber<? super T> z = RxJavaPlugins.onSubscribe(this, s);
ObjectHelper.requireNonNull(z, "Plugin returned null Subscriber");
subscribeActual(z);
} catch (NullPointerException e) { // NOPMD
throw e;
} catch (Throwable e) {
......
throw npe;
}
}
Flowable.subscribe跟我们前面是一样的,最终调用的是FlowableCreate的subscribeActual方法
3、FlowableCreate.subscribeActual
public final class FlowableCreate<T> extends Flowable<T> {
final FlowableOnSubscribe<T> source;
final BackpressureStrategy backpressure;
public FlowableCreate(FlowableOnSubscribe<T> source, BackpressureStrategy backpressure) {
this.source = source;
this.backpressure = backpressure;
}
@Override
public void subscribeActual(Subscriber<? super T> t) {
//使用三步曲分析法
BaseEmitter<T> emitter;
switch (backpressure) {
case MISSING: {
emitter = new MissingEmitter<T>(t);
break;
}
case ERROR: {
emitter = new ErrorAsyncEmitter<T>(t);
break;
}
case DROP: {
emitter = new DropAsyncEmitter<T>(t);
break;
}
case LATEST: {
emitter = new LatestAsyncEmitter<T>(t);
break;
}
default: {
emitter = new BufferAsyncEmitter<T>(t, bufferSize());
break;
}
}
t.onSubscribe(emitter);
try {
source.subscribe(emitter);
} catch (Throwable ex) {
Exceptions.throwIfFatal(ex);
emitter.onError(ex);
}
}
}
subscribeActual会根据不同的策略生成不同的发射器,具体的所有策略逻辑都在发射器中体现的
4、DropAsyncEmitter
static final class DropAsyncEmitter<T> extends NoOverflowBaseAsyncEmitter<T> {
private static final long serialVersionUID = 8360058422307496563L;
DropAsyncEmitter(Subscriber<? super T> actual) {
super(actual);
}
@Override
void onOverflow() {
// nothing to do
}
}
DropAsyncEmitter其实没做什么事情,主要都在其父类中实现了,onOverflow的回调表示事件流溢出的时候的处理,很明显Drop策略就把溢出的数据项直接不做处理,意思就是抛弃掉这个数据项了
static final class ErrorAsyncEmitter<T> extends NoOverflowBaseAsyncEmitter<T> {
private static final long serialVersionUID = 338953216916120960L;
ErrorAsyncEmitter(Subscriber<? super T> actual) {
super(actual);
}
@Override
void onOverflow() {
onError(new MissingBackpressureException("create: could not emit value due to lack of requests"));
}
}
再看看Error策略,溢出之后就会抛出溢出的异常,其他策略也类似分析,具体父类是如何处理溢出函数的呢
5、NoOverflowBaseAsyncEmitter
abstract static class NoOverflowBaseAsyncEmitter<T> extends BaseEmitter<T> {
private static final long serialVersionUID = 4127754106204442833L;
NoOverflowBaseAsyncEmitter(Subscriber<? super T> actual) {
super(actual);
}
@Override
public final void onNext(T t) {
if (isCancelled()) {
return;
}
if (t == null) {
onError(new NullPointerException("onNext called with null. Null values are generally not allowed in 2.x operators and sources."));
return;
}
//这里暂时将get()函数当作是类似于List这种的容器,存储的是当前需要处理的数据项
if (get() != 0) { //从数据项容器中取值,如果当前有数据项需要处理
actual.onNext(t); //发射数据项
BackpressureHelper.produced(this, 1); //对当前存在需要处理的数据项进行-1操作
} else {
onOverflow(); //从数据项容器中取值,如果当前没有数据项需要处理,则回调溢出函数
}
}
abstract void onOverflow();
}
NoOverflowBaseAsyncEmitter在发射数据项的时候,会去BaseEmitter中的数据项容器去取出数据项,如果存在则处理,不存在则表示溢出,回调溢出函数,那么具体的数据项容器时候怎么存储需要处理的数据项的呢
6、BaseEmitter
abstract static class BaseEmitter<T>
extends AtomicLong
implements FlowableEmitter<T>, Subscription {
private static final long serialVersionUID = 7326289992464377023L;
final Subscriber<? super T> actual;
final SequentialDisposable serial;
BaseEmitter(Subscriber<? super T> actual) {
this.actual = actual;
this.serial = new SequentialDisposable();
}
@Override
public void onComplete() {
complete();
}
protected void complete() {
if (isCancelled()) {
return;
}
try {
actual.onComplete();
} finally {
serial.dispose();
}
}
@Override
public final void onError(Throwable e) {
if (!tryOnError(e)) {
RxJavaPlugins.onError(e);
}
}
@Override
public final void request(long n) {
if (SubscriptionHelper.validate(n)) {
BackpressureHelper.add(this, n);
}
}
}
BaseEmitter就是一个AtomicLong,如果没学过AtomicLong的话,可以简单理解为一个计数器,get()就是获取当前的Long值,只要不等于0就表示有值。主要还是在
request()
,
request()
表示此时需要处理的数据项。结合上面NoOverflowBaseAsyncEmitter的中的
BackpressureHelper.produced(this, 1)
和当前BaseEmitter中的
BackpressureHelper.add(this, n)
,可得数据项的容器完全都是由BackpressureHelper去控制,我们只需要对BackpressureHelper的存储和获取做分析,就可以知道当前是否有数据项需要处理
7、BackpressureHelper
public final class BackpressureHelper {
public static long add(AtomicLong requested, long n) {
for (;;) {
long r = requested.get(); //获取当前数据项
if (r == Long.MAX_VALUE) {
return Long.MAX_VALUE;
}
long u = addCap(r, n);//当前数据项 + 新增的数据项
if (requested.compareAndSet(r, u)) { //设置最新的数据项
return r;
}
}
}
public static long addCap(long a, long b) {
long u = a + b;
if (u < 0L) {
return Long.MAX_VALUE;
}
return u;
}
public static long produced(AtomicLong requested, long n) {
for (;;) {
long current = requested.get(); //获取当前数据项
if (current == Long.MAX_VALUE) {
return Long.MAX_VALUE;
}
long update = current - n; //当前数据项 - 需要发射的数据项(从源码上,n为1)
if (update < 0L) { //不能为负数
RxJavaPlugins.onError(new IllegalStateException("More produced than requested: " + update));
update = 0L;
}
if (requested.compareAndSet(current, update)) { //设置最新的数据项
return update;
}
}
}
}
BackpressureHelper就是利用AtomicLong的原子性就行简单的计数器操作而已,并没有什么复杂的操作。至此,我们就知道背压的原理原来就是利用AtomicLong计数器和生产消费的模式去决定是否发射当前的数据项而已
10.4 RxJava的常用操作符原理
知识点:
- 理解map操作符的原理
RxJava常用操作符的代表就是map,分析map源码后,其他的操作符的思想是一样的,只不过是实现逻辑不一致而已。下面我们通过分析map的主要流程去分析map是如何转换字符串的,从上面我们知道Observable的创建、订阅、发射的过程,这次对于重复的内容就不再继续分析,主要是分析中间map是如何回调
apply()
去将数据项转换成字符串的
public void map() {
//创建被观察者
Observable
.create(new ObservableOnSubscribe<String>() {
@Override
//默认在主线程里执行该方法
public void subscribe(@NonNull ObservableEmitter<String> e) throws Exception {
e.onNext("俊俊俊很帅");
e.onNext("你值得拥有");
e.onNext("取消关注");
e.onNext("但还是要保持微笑");
e.onComplete();
}
})
.map(new Function<String, String>() {
@Override
public String apply(String s) throws Exception {
return "Hello";
}
})
//创建观察者并订阅
.subscribe(new Observer<String>() {
@Override
public void onSubscribe(Disposable d) {
if (!d.isDisposed()) {
d.dispose();
}
}
@Override
public void onNext(String s) {
System.out.println("onNext=" + s);
}
@Override
public void onError(Throwable e) {
System.out.println("onNext=" + e.getMessage());
}
@Override
public void onComplete() {
}
});
}
以下是RxJava源代码
1、Observable.map
public static <T> Observable<T> create(ObservableOnSubscribe<T> source) {
ObjectHelper.requireNonNull(source, "source is null");
return RxJavaPlugins.onAssembly(new ObservableCreate<T>(source));
}
public final <R> Observable<R> map(Function<? super T, ? extends R> mapper) {
ObjectHelper.requireNonNull(mapper, "mapper is null");
return RxJavaPlugins.onAssembly(new ObservableMap<T, R>(this, mapper));
}
从create到map的过程中,create的时候,当前的Observable已经被转换成
ObservableCreate
,再次map的时候,
当前的Observable已经被转换成
ObservableMap
,而且在
ObservableMap
中传递的参数包含
this
,所以当前
ObservableMap
中是嵌套着
ObservableCreate
2、Observable.subscribe
由于当前的Observable是
ObservableMap
,所以Observable.subscribe会回调
ObservableMap
中的
subscribeActual
public final class ObservableMap<T, U> extends AbstractObservableWithUpstream<T, U> {
final Function<? super T, ? extends U> function;
public ObservableMap(ObservableSource<T> source, Function<? super T, ? extends U> function) {
super(source);
this.function = function;
}
@Override
public void subscribeActual(Observer<? super U> t) {
source.subscribe(new MapObserver<T, U>(t, function));//source是传递进来的ObservableCreate
}
}
ObservableMap
中的
subscribeActual
,会去调用
ObservableCreate
的
subscribe
方法,最后还是会去回调
ObservableCreate
的
subscribeActual
,不过这里在回调的过程中增加了一个参数
MapObserver
,这个参数只有在
ObservableCreate
发射器发射的时候才会被调用
3、ObservableCreate.subscribeActual
@Override
protected void subscribeActual(Observer<? super T> observer) {
CreateEmitter<T> parent = new CreateEmitter<T>(observer);
observer.onSubscribe(parent);
try {
source.subscribe(parent);
} catch (Throwable ex) {
Exceptions.throwIfFatal(ex);
parent.onError(ex);
}
}
在
ObservableCreate.subscribeActual
中,会接收一个
Observer
的参数,这个时候的
Observer
的参数是从
ObservableMap
中传递过来的
MapObserver
,当
CreateEmitter
发射
onNext
的时候,就会在当前的
MapObserver
对象
onNext
进行处理
4、MapObserver.onNext
static final class MapObserver<T, U> extends BasicFuseableObserver<T, U> {
final Function<? super T, ? extends U> mapper;
MapObserver(Observer<? super U> actual, Function<? super T, ? extends U> mapper) {
super(actual);
this.mapper = mapper;
}
@Override
public void onNext(T t) {
if (done) {
return;
}
if (sourceMode != NONE) {
actual.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;
}
actual.onNext(v);
}
......
}
onNext
主要做了两个事情,一个是
mapper.apply(t)
,这个就是
map
操作符所实现的方法,这里就将原来的值转换成新的值,一个是
actual.onNext(v)
,将转换出来的新值
v
继续
onNext
出去,这里的
actual
就是在构造函数中传递进来的
ObservableCreate
,这里就已经将数据项经过map的操作符后继续执行后面正常的发射流程
小结
如图所示
10.5 RxJava的线程切换原理
知识点:
- 理解在工作线程上为什么能执行耗时操作
- 理解在UI线程为什么能执行更新UI的操作
沿用上面的例子,在线程切换的过程中,无非就是相当于不同的操作符继续操作数据项而已,根本的实现思路和map等操作符是一样的,也是通过嵌套Observable的过程来执行的,只不过是线程切换的操作符内部实现的逻辑有区别而已。通过我们以往的思路去想,这两个知识点无非就是启动线程池去执行耗时任务,而UI线程则是交给Handler去处理,RxJava线程切换的原理就是这样的
Observable
.create(new ObservableOnSubscribe<String>() {
@Override
//默认在主线程里执行该方法
public void subscribe(@NonNull ObservableEmitter<String> e) throws Exception {
e.onNext("俊俊俊很帅");
e.onNext("你值得拥有");
e.onNext("取消关注");
e.onNext("但还是要保持微笑");
e.onComplete();
}
})
.map(new Function<String, String>() {
@Override
public String apply(String s) throws Exception {
return "Hello";
}
})
//将被观察者切换到子线程
.subscribeOn(Schedulers.io())
//将观察者切换到主线程 需要在Android环境下运行
.observeOn(AndroidSchedulers.mainThread())
//创建观察者并订阅
.subscribe(new Observer<String>() {
@Override
public void onSubscribe(Disposable d) {
if (!d.isDisposed()) {
d.dispose();
}
}
@Override
public void onNext(String s) {
System.out.println("onNext=" + s);
}
@Override
public void onError(Throwable e) {
System.out.println("onNext=" + e.getMessage());
}
@Override
public void onComplete() {
}
});
基础概念:
- Schedulers:调度器的管理者。管理着多种不同种类的Scheduler
- Scheduler:调度器。负责线程Worker的创建
,调度Worker的执行createWorker()
schedule()
- Worker:抽象的工作线程。被线程调度器管理,负责线程的创建和执行
在源码中,我们需要先熟悉这三者之间的关系到底是如何运作的
以下是RxJava源代码
1、observeOn()
@CheckReturnValue
@SchedulerSupport("custom")
public final Observable<T> observeOn(Scheduler scheduler) {
return this.observeOn(scheduler, false, bufferSize());
}
@CheckReturnValue
@SchedulerSupport("custom")
public final Observable<T> observeOn(Scheduler scheduler, boolean delayError, int bufferSize) {
ObjectHelper.requireNonNull(scheduler, "scheduler is null");
ObjectHelper.verifyPositive(bufferSize, "bufferSize");
return RxJavaPlugins.onAssembly(new ObservableObserveOn(this, scheduler, delayError, bufferSize));
}
当前的Observable已经被转换成
ObservableObserveOn
public final class ObservableObserveOn<T> extends AbstractObservableWithUpstream<T, T> {
final Scheduler scheduler;
final boolean delayError;
final int bufferSize;
public ObservableObserveOn(ObservableSource<T> source, Scheduler scheduler, boolean delayError, int bufferSize) {
super(source);
this.scheduler = scheduler;
this.delayError = delayError;
this.bufferSize = bufferSize;
}
protected void subscribeActual(Observer<? super T> observer) {
if (this.scheduler instanceof TrampolineScheduler) {
this.source.subscribe(observer);
} else {
//1、创建工作线程
Worker w = this.scheduler.createWorker();
//2、订阅之后,在发射的过程中
this.source.subscribe(new ObservableObserveOn.ObserveOnObserver(observer, w, this.delayError, this.bufferSize));
}
}
static final class ObserveOnObserver<T> extends BasicIntQueueDisposable<T> implements Observer<T>, Runnable {
private static final long serialVersionUID = 6576896619930983584L;
final Observer<? super T> actual;
final Worker worker;
final boolean delayError;
final int bufferSize;
SimpleQueue<T> queue;
public void onNext(T t) {
if (!this.done) {
if (this.sourceMode != 2) {
this.queue.offer(t);
}
//3、在OnNext中执行
this.schedule();
}
}
void schedule() {
if (this.getAndIncrement() == 0) {
//4、执行工作线程
this.worker.schedule(this);
}
}
}
}
其三者的关系简单的说就是在每次订阅的时候,都会去创建出对应的工作线程,这个工作线程取决于你传递的参数是哪个Worker,在发射器发射的过程中,这个工作线程总会去执行它的回调
schedule
,其实大部分的操作就是在
schedule
里面执行线程。搞懂了三者的关系之后,分析线程切换就简单多了,就相当于工厂一样,给个具体的任务给到具体的工人去执行,很像工厂的流水线,我们已经确定下来了流水线的流程了,这个时候我们就需要去关心参数具体是什么东西了。在阅读subscribeOn、observeOn前,我们先看看这两个方法中的参数都是什么
1、Schedulers.io()
public final class Schedulers {
@NonNull
static final Scheduler IO;
static {
SINGLE = RxJavaPlugins.initSingleScheduler(new SingleTask());
COMPUTATION = RxJavaPlugins.initComputationScheduler(new ComputationTask());
//1、在初始化的时候就构建出了IOTask,initIoScheduler会去执行IOTask的call方法
IO = RxJavaPlugins.initIoScheduler(new IOTask());
TRAMPOLINE = TrampolineScheduler.instance();
NEW_THREAD = RxJavaPlugins.initNewThreadScheduler(new NewThreadTask());
}
static final class IOTask implements Callable<Scheduler> {
@Override
public Scheduler call() throws Exception {
//2、IOTask的call方法会去获取IoHolder的值
return IoHolder.DEFAULT;
}
}
static final class IoHolder {
//3、创建IoScheduler
static final Scheduler DEFAULT = new IoScheduler();
}
public static Scheduler io() {
//Schedulers.io():它会去获取前面3步创建出来的IoScheduler对象
return RxJavaPlugins.onIoScheduler(IO); //返回IO自身
}
}
其正在实现在IoScheduler,其表示管理Io线程的管理者
public final class IoScheduler extends Scheduler {
final AtomicReference<CachedWorkerPool> pool;
static final CachedWorkerPool NONE;
static {
......
//1、创建CachedWorkerPool
NONE = new CachedWorkerPool(0, null, WORKER_THREAD_FACTORY);
}
static final class CachedWorkerPool implements Runnable {
......
private final ScheduledExecutorService evictorService;
private final Future<?> evictorTask;
CachedWorkerPool(long keepAliveTime, TimeUnit unit, ThreadFactory threadFactory) {
......
Future<?> task = null;
if (unit != null) {
evictor = Executors.newScheduledThreadPool(1, EVICTOR_THREAD_FACTORY);
task = evictor.scheduleWithFixedDelay(this, this.keepAliveTime, this.keepAliveTime, TimeUnit.NANOSECONDS);
}
evictorService = evictor;
evictorTask = task;
}
}
@NonNull
@Override
public Worker createWorker() {
//2、创建具体的线程
return new EventLoopWorker(pool.get());
}
static final class EventLoopWorker extends Scheduler.Worker {
private final CompositeDisposable tasks;
private final CachedWorkerPool pool;
private final ThreadWorker threadWorker;
final AtomicBoolean once = new AtomicBoolean();
EventLoopWorker(CachedWorkerPool pool) {
this.pool = pool;
this.tasks = new CompositeDisposable();
this.threadWorker = pool.get();
}
@NonNull
@Override
public Disposable schedule(@NonNull Runnable action, long delayTime, @NonNull TimeUnit unit) {
if (tasks.isDisposed()) {
// don't schedule, we are unsubscribed
return EmptyDisposable.INSTANCE;
}
//3、最终会去调用ThreadWorker的scheduleActual
return threadWorker.scheduleActual(action, delayTime, unit, tasks);
}
}
//4、由于ThreadWorker没有scheduleActual,在父类中找NewThreadWorker
static final class ThreadWorker extends NewThreadWorker {
private long expirationTime;
ThreadWorker(ThreadFactory threadFactory) {
super(threadFactory);
this.expirationTime = 0L;
}
public long getExpirationTime() {
return expirationTime;
}
public void setExpirationTime(long expirationTime) {
this.expirationTime = expirationTime;
}
}
}
NewThreadWorker,最终还是调用
executor.submit()
或
executor.schedule()
@NonNull
public ScheduledRunnable scheduleActual(final Runnable run, long delayTime, @NonNull TimeUnit unit, @Nullable DisposableContainer parent) {
Runnable decoratedRun = RxJavaPlugins.onSchedule(run);
ScheduledRunnable sr = new ScheduledRunnable(decoratedRun, parent);
if (parent != null) {
if (!parent.add(sr)) {
return sr;
}
}
Future<?> f;
try {
if (delayTime <= 0) {
f = executor.submit((Callable<Object>)sr);
} else {
f = executor.schedule((Callable<Object>)sr, delayTime, unit);
}
sr.setFuture(f);
} catch (RejectedExecutionException ex) {
if (parent != null) {
parent.remove(sr);
}
RxJavaPlugins.onError(ex);
}
return sr;
}
2、AndroidSchedulers.mainThread()
public final class AndroidSchedulers {
private static final class MainHolder {
static final Scheduler DEFAULT = new HandlerScheduler(new Handler(Looper.getMainLooper()));
}
private static final Scheduler MAIN_THREAD = RxAndroidPlugins.initMainThreadScheduler(
new Callable<Scheduler>() {
@Override public Scheduler call() throws Exception {
return MainHolder.DEFAULT;
}
});
/** A {@link Scheduler} which executes actions on the Android main thread. */
public static Scheduler mainThread() {
return RxAndroidPlugins.onMainThreadScheduler(MAIN_THREAD);
}
}
返回一个HandlerScheduler,创建单例模式的主线程Handler
final class HandlerScheduler extends Scheduler {
private final Handler handler;
HandlerScheduler(Handler handler) {
this.handler = handler;
}
public Disposable scheduleDirect(Runnable run, long delay, TimeUnit unit) {
if (run == null) {
throw new NullPointerException("run == null");
} else if (unit == null) {
throw new NullPointerException("unit == null");
} else {
run = RxJavaPlugins.onSchedule(run);
HandlerScheduler.ScheduledRunnable scheduled = new HandlerScheduler.ScheduledRunnable(this.handler, run);
this.handler.postDelayed(scheduled, unit.toMillis(delay));
return scheduled;
}
}
public Worker createWorker() {
//创建具体工作线程
return new HandlerScheduler.HandlerWorker(this.handler);
}
......
}
就好像我们上面分析的三者关系一样,Schedule最终还是会管理着具体的工作线程
private static final class HandlerWorker extends Worker {
private final Handler handler;
private volatile boolean disposed;
HandlerWorker(Handler handler) {
this.handler = handler;
}
@Override
public Disposable schedule(Runnable run, long delay, TimeUnit unit) {
if (run == null) throw new NullPointerException("run == null");
if (unit == null) throw new NullPointerException("unit == null");
if (disposed) {
return Disposables.disposed();
}
run = RxJavaPlugins.onSchedule(run);
//包装新的Runnable交给Handler
ScheduledRunnable scheduled = new ScheduledRunnable(handler, run);
Message message = Message.obtain(handler, scheduled);
message.obj = this; // Used as token for batch disposal of this worker's runnables.
handler.sendMessageDelayed(message, unit.toMillis(delay));
// Re-check disposed state for removing in case we were racing a call to dispose().
if (disposed) {
handler.removeCallbacks(scheduled);
return Disposables.disposed();
}
return scheduled;
}
@Override
public void dispose() {
disposed = true;
handler.removeCallbacksAndMessages(this /* token */);
}
@Override
public boolean isDisposed() {
return disposed;
}
}
3、subscribeOn()
理解完参数后,回到我们的分析重点
public final Observable<T> subscribeOn(Scheduler scheduler) {
ObjectHelper.requireNonNull(scheduler, "scheduler is null");
return RxJavaPlugins.onAssembly(new ObservableSubscribeOn<T>(this, scheduler));
}
subscribeOn就如同普通操作符一样,包装一层
ObservableSubscribeOn
,在subscribe的时候真正走的还是
subscribeActual
public final class ObservableSubscribeOn<T> extends AbstractObservableWithUpstream<T, T> {
@Override
public void subscribeActual(final Observer<? super T> s) {
//使用三步曲分析法
final SubscribeOnObserver<T> parent = new SubscribeOnObserver<T>(s);
s.onSubscribe(parent);
//3、将第三步的内容放到线程中去执行
parent.setDisposable(scheduler.scheduleDirect(new SubscribeTask(parent)));
}
final class SubscribeTask implements Runnable {
private final SubscribeOnObserver<T> parent;
SubscribeTask(SubscribeOnObserver<T> parent) {
this.parent = parent;
}
@Override
public void run() {
//3、回调ObservableOnSubscribe的subscribe
source.subscribe(parent);
}
}
}
scheduler.scheduleDirect
中会去执行Scheduler里的方法,这里的scheduler就是IoScheduler
@NonNull
public Disposable scheduleDirect(@NonNull Runnable run) {
return scheduleDirect(run, 0L, TimeUnit.NANOSECONDS);
}
@NonNull
public Disposable scheduleDirect(@NonNull Runnable run, long delay, @NonNull TimeUnit unit) {
final Worker w = createWorker();
final Runnable decoratedRun = RxJavaPlugins.onSchedule(run);
DisposeTask task = new DisposeTask(decoratedRun, w);
w.schedule(task, delay, unit);
return task;
}
回调IoScheduler的
createWorker()
并执行
w.schedule()
小结
如图所示
10.6 RxJava的自定义Operator原理
知识点:
- 自定义Operator是如何实现的
在讲解之前,让我们先回味下自定义Operator
public class CustomOperator implements ObservableOperator<String, List<String>> {
@Override
public Observer<? super List<String>> apply(final Observer<? super String> observer) throws Exception {
return new Observer<List<String>>() {
@Override
public void onSubscribe(Disposable d) {
observer.onSubscribe(d);
}
@Override
public void onNext(List<String> strings) {
observer.onNext(strings.toString());
}
@Override
public void onError(Throwable e) {
observer.onError(e);
}
@Override
public void onComplete() {
observer.onComplete();
}
};
}
}
Observable.create(new ObservableOnSubscribe<List<String>>() {
@Override
public void subscribe(@NonNull ObservableEmitter<List<String>> e) throws Exception {
}
}).lift(new CustomOperator())
自定义Operator如同普通的操作符原理差不多,用的是
lift
的操作符,只不过在
lift
里面将逻辑的执行回调到自定义的Operator的
apply()
以下是RxJava源代码
1、Observable.lift
public final <R> Observable<R> lift(ObservableOperator<? extends R, ? super T> lifter) {
ObjectHelper.requireNonNull(lifter, "onLift is null");
return RxJavaPlugins.onAssembly(new ObservableLift<R, T>(this, lifter));
}
2、ObservableLift.subscribeActual
public final class ObservableLift<R, T> extends AbstractObservableWithUpstream<T, R> {
/** The actual operator. */
final ObservableOperator<? extends R, ? super T> operator;
public ObservableLift(ObservableSource<T> source, ObservableOperator<? extends R, ? super T> operator) {
super(source);
this.operator = operator;
}
@Override
public void subscribeActual(Observer<? super R> s) {
Observer<? super T> observer;
try {
observer = ObjectHelper.requireNonNull(operator.apply(s), "Operator " + operator + " returned a null Observer");
} catch (NullPointerException e) { // NOPMD
throw e;
} catch (Throwable e) {
Exceptions.throwIfFatal(e);
// can't call onError because no way to know if a Disposable has been set or not
// can't call onSubscribe because the call might have set a Disposable already
RxJavaPlugins.onError(e);
NullPointerException npe = new NullPointerException("Actually not, but can't throw other exceptions due to RS");
npe.initCause(e);
throw npe;
}
source.subscribe(observer);
}
}
可以看到代码非常快的就将传递进来的参数
operator
执行
apply()
10.7 RxJava的自定义Transformer原理
知识点:
- 自定义Transformer是如何实现的
在讲解之前,让我们先回味下自定义Transformer
public class NetWorkTransformer implements ObservableTransformer {
@Override
public ObservableSource apply(Observable upstream) {
return upstream.subscribeOn(Schedulers.io()).observeOn(AndroidSchedulers.mainThread());
}
}
Observable.create(new ObservableOnSubscribe<Integer>() {
@Override
public void subscribe(ObservableEmitter<Integer> emitter) throws Exception {
}
}).compose(new CustomTransformer())
自定义Transformer如同普通的操作符原理差不多,用的是
compose
的操作符,只不过在
compose
里面将逻辑的执行回调到自定义的Transformer的
apply()
以下是RxJava源代码
1、Observable.compose
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));
}
可以看到代码非常快的就将传递进来的参数
composer
执行
apply()
,这里的
wrap()
只是将代码裹了一层,如果你想简单的理解的话,可以理解为作者的强迫症犯了,只是为了让所有代码看起来都比较规范,不然这里实在和其他操作符的实现不一样,我们可以追进去
wrap()
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));
}
wrap()
其实是对
composer
操作符做了Hook,因为所有操作符都会被RxJava去Hook住,这里会在下面讲到自定义Plugin原理的时候就明白了
10.8 RxJava的自定义Plugin原理
知识点:
- 自定义Plugin是如何实现AOP的
在讲解之前,让我们先回味下自定义Plugin
public class CustomObservableAssembly implements Function<Observable, Observable> {
@Override
public Observable apply(Observable observable) throws Exception {
System.out.println("CustomObservableAssembly observable.toString:" + observable.toString());
return observable;
}
}
RxJavaPlugins.setOnObservableAssembly(new CustomObservableAssembly());
在自定义Plugin中,类似于Android的术语Hook,但在这里并不是真正的Hook,而是作者在写RxJava的时候去限定一套规范,让后面的所有操作符或其他操作等,都可以实现Hook的原理
以下是RxJava源代码
1、RxJavaPlugins.setOnObservableAssembly
public static void setOnObservableAssembly(@Nullable Function<? super Observable, ? extends Observable> onObservableAssembly) {
if (lockdown) {
throw new IllegalStateException("Plugins can't be changed anymore");
}
RxJavaPlugins.onObservableAssembly = onObservableAssembly;
}
RxJavaPlugins.setOnObservableAssembly
只是对成员变量设置了自定义的值,这个时候
onObservableAssembly
就有了值,默认是为null的。设置完值就表示已经Hook成功了,当操作符执行的时候,是如何回调我们Hook的函数的
2、Observable.create
public static <T> Observable<T> create(ObservableOnSubscribe<T> source) {
ObjectHelper.requireNonNull(source, "source is null");
return RxJavaPlugins.onAssembly(new ObservableCreate<T>(source));
}
create相当于一个操作符,在每个操作符的里面都会去执行一段
RxJavaPlugins.onAssembly
,这里就是RxJava规定的规范,一开始我们只是说返回自身,但是有了Hook之后,就会回调Hook函数,返回已经经过二次加工的自身
3、RxJavaPlugins.onAssembly
public static <T> Observable<T> onAssembly(@NonNull Observable<T> source) {
Function<? super Observable, ? extends Observable> f = onObservableAssembly;
if (f != null) {
return apply(f, source);
}
return source;
}
由于我们已经设置了新值,这里的
onObservableAssembly
就不为null,不为null则执行
apply()
,
apply()
就是我们Hook传进去参数的回调方法
10.9 美团WhiteBoard
美团的WhiteBoard其实是取自美团的开源框架
Shield——开源的移动端页面模块化开发框架
中的代码,其主要作用是应用RxJava的Subject搭起组件间通讯的桥梁。实质上在WhiteBoard中,是将所有的组件的数据和Subject通讯的桥梁保存起来,通过key作为组件的唯一标志。不过比较可惜的是WhiteBoard使用的是RxJava1,不过关系不大,只要读懂里面的源码即可
1、WhiteBoard的初始化
初始化放在Activity/Fragment界面中,相当于通讯的桥梁,每个界面中仅有一个WhiteBoard的实例,并由所有组件共用
public abstract class ShieldFragment extends Fragment implements AgentCellBridgeInterface, DriverInterface {
static final String TAG = ShieldFragment.class.getSimpleName();
......
protected WhiteBoard whiteBoard;
public ShieldFragment() {
this.whiteBoard = new WhiteBoard();//初始化
}
@Override
public void onCreate(Bundle savedInstanceState) {
super.onCreate(savedInstanceState);
......
whiteBoard.onCreate(savedInstanceState);//对应生命周期
}
@Override
public void onDestroy() {
super.onDestroy();
......
whiteBoard.onDestory();//对应生命周期
}
@Override
public void onSaveInstanceState(Bundle outState) {
super.onSaveInstanceState(outState);
......
whiteBoard.onSaveInstanceState(outState);//对应生命周期
}
@Override
public WhiteBoard getWhiteBoard() {
return whiteBoard;//获取实例
}
}
2、WhiteBoard监听通知
组件只监听某个key的事件,有通知的时候就能收到
public class MixCellAgent extends LightAgent {
private MixCell mixCell;
private Subscription loadingSubscription;
private Subscription emptySubscription;
public MixCellAgent(Fragment fragment, DriverInterface bridge, PageContainerInterface pageContainer) {
super(fragment, bridge, pageContainer);
}
@Override
public void onCreate(Bundle savedInstanceState) {
super.onCreate(savedInstanceState);
mixCell = new MixCell(getContext(), this);
loadingSubscription = getWhiteBoard().getObservable(MixLoadingAgent.KEY_LOADING).filter(new Func1() {
@Override
public Object call(Object o) {
return o instanceof Boolean && ((Boolean) o);
}
}).subscribe(new Action1() {
@Override
public void call(Object o) {
loading();
}
});
emptySubscription = getWhiteBoard().getObservable(MixLoadingAgent.KEY_EMPTY).filter(new Func1<Object, Boolean>() {
@Override
public Boolean call(Object o) {
return o instanceof Boolean && ((Boolean) o);
}
}).subscribe(new Action1() {
@Override
public void call(Object o) {
mixCell.onEmpty();
}
});
......
}
@Override
public void onDestroy() {
if (loadingSubscription != null) {
loadingSubscription.unsubscribe();
loadingSubscription = null;
}
if (emptySubscription != null) {
emptySubscription.unsubscribe();
}
......
}
}
3、WhiteBoard的发送通知
WhiteBoard发送通知就是调用WhiteBoard提供的所有put方法,具体是如何收到消息的,还需要通过WhiteBoard的源码看下
public class MixLoadingAgent extends LightAgent implements MixLoadingCell.MixLoadingListener {
public static final String KEY_LOADING = "loading";
public static final String KEY_EMPTY = "empty";
public static final String KEY_FAILED = "failed";
public static final String KEY_MORE = "more";
public static final String KEY_DONE = "done";
private MixLoadingCell mixLoadingCell;
public MixLoadingAgent(Fragment fragment, DriverInterface bridge, PageContainerInterface pageContainer) {
super(fragment, bridge, pageContainer);
mixLoadingCell = new MixLoadingCell(getContext());
mixLoadingCell.setOnMixLoadingListener(this);
}
@Override
public SectionCellInterface getSectionCellInterface() {
return mixLoadingCell;
}
@Override
public void onLoading() {
getWhiteBoard().putBoolean(KEY_LOADING, true);
}
@Override
public void onEmpty() {
getWhiteBoard().putBoolean(KEY_EMPTY, true);
}
@Override
public void onFailed() {
getWhiteBoard().putBoolean(KEY_FAILED, true);
}
@Override
public void onMore() {
getWhiteBoard().putBoolean(KEY_MORE, true);
}
@Override
public void onDone() {
getWhiteBoard().putBoolean(KEY_DONE, true);
}
}
4、WhiteBoard的原理
最后只需要获取实例后发送通知即可,
getWhiteBoard().putBoolean(key)
。WhiteBoard原理是只要还是Subject的桥梁的作用
public class WhiteBoard {
public static final String WHITE_BOARD_DATA_KEY = "White_Board_Data";
protected Bundle mData;//保存所有组件的数据
protected HashMap<String, Subject> subjectMap;//保存所有组件的通讯桥梁
public WhiteBoard() {
this(null);
}
public WhiteBoard(Bundle data) {
mData = data;
if (mData == null) {
mData = new Bundle();//初始化
}
subjectMap = new HashMap<>();//初始化
}
public void onCreate(Bundle savedInstanceState) {
if (savedInstanceState != null) {
mData = savedInstanceState.getBundle(WHITE_BOARD_DATA_KEY);
}
if (mData == null) {
mData = new Bundle();
}
}
public void onSaveInstanceState(Bundle outState) {
if (outState != null) {
// here we must save a new copy of the mData into the outState
outState.putBundle(WHITE_BOARD_DATA_KEY, new Bundle(mData));
}
}
public void onDestory() {
subjectMap.clear();
mData.clear();
}
//通过key获取某组件的桥梁
public Observable getObservable(final String key) {
Subject res = null;
if (subjectMap.containsKey(key)) {
res = subjectMap.get(key);
} else {
res = PublishSubject.create();
subjectMap.put(key, res);
}
if (getData(key) != null) {
return res.startWith(getData(key));//带上已经存储过的数据
} else {
return res;
}
}
//通过key通知某组件
protected void notifyDataChanged(String key) {
if (subjectMap.containsKey(key)) {
subjectMap.get(key).onNext(mData.get(key));
}
}
//移除组件中的数据
public void removeData(String key) {
mData.remove(key);
notifyDataChanged(key);
}
//每次put值的时候,就会去通知对应的组件
public void putBoolean(@Nullable String key, boolean value) {
mData.putBoolean(key, value);
notifyDataChanged(key);
}
public void putInt(@Nullable String key, int value) {
mData.putInt(key, value);
notifyDataChanged(key);
}
public void putString(@Nullable String key, @Nullable String value) {
mData.putString(key, value);
notifyDataChanged(key);
}
......
public double getDouble(String key) {
return mData.getDouble(key);
}
public String getString(String key, String defaultValue) {
return mData.getString(key, defaultValue);
}
......
}