对Handler工作原理的分析

一直对Handler的工作原理非常好奇，看了网上很多帖子还是感觉似懂非懂。今天专门亲自从源码角度分析一下Handler的工作过程。

 

private Handler mHandler;

@Override
protected void onCreate(BundlesavedInstanceState) {
    super.onCreate(savedInstanceState);
    setContentView(R.layout.activity_main);
    textView = (TextView)findViewById(R.id.tv01);
    mHandler = new Handler(){
        @Override
        public void handleMessage(Message msg) {
            super.handleMessage(msg);
        }
    };

 

这是很常规的一种写法，实际上这样写是不对的，这样的话Handler的会持有Activity的引用，导致Activity不能及时得到回收。并且从源码中可以看到

if (FIND_POTENTIAL_LEAKS) {
    final Class<? extends Handler> klass =getClass();
    if ((klass.isAnonymousClass()|| klass.isMemberClass() || klass.isLocalClass()) &&
            (klass.getModifiers() &Modifier.STATIC) == 0) {
        Log.w(TAG, "The followingHandler class should be static or leaks might occur: " +
            klass.getCanonicalName());
    }
}

 

Handler在构造的时候会检查你是否将它声明为单独的一个类，如果没有单独去静态继承Handler类去实现的话发出警告，因为这个问题不是致命的。

正确的方式是这样做，

private static class MyHandler extends Handler{
    @Override
    public void handleMessage(Message msg) {
        super.handleMessage(msg);
    }
}

 

接下来会发生这个

mLooper = Looper.myLooper();
if (mLooper == null) {
    throw new RuntimeException(
        "Can't createhandler inside thread that has not called Looper.prepare()");
}

 

Looper.myLooper()去拿当前线程的Looper对象。如果是在主线程中Looper.prepare()方法在创建Activity之前已经执行，所以myLooper()会返回一个Looper对象。但是如果是在子线程去调myLooper()是没法拿到Looper对象的。看看Looper::prepare()的源码，

public static void prepare() {
    prepare(true);
}

private static void prepare(boolean quitAllowed) {
    if (sThreadLocal.get() != null) {
        throw new RuntimeException("Only one Loopermay be created per thread");
    }
    sThreadLocal.set(new Looper(quitAllowed));
}

Looper::prepare()方法向ThreadLocal变量中保存了一个Looper对象。

public static @Nullable Looper myLooper() {
    return sThreadLocal.get();
}

而Looper::myLooper()就是将ThreadLocal变量中保存的值返回出去。接下来这句

mQueue = mLooper.mQueue;

Handler用的Queue是Looper的Queue。

Handler中的发送消息最终都会调到这个方法

public boolean sendMessageAtTime(Message msg, long uptimeMillis) {
    MessageQueue queue = mQueue;
    if (queue == null) {
        RuntimeException e = new RuntimeException(
                this + "sendMessageAtTime() called with no mQueue");
        Log.w("Looper", e.getMessage(), e);
        return false;
    }
    return enqueueMessage(queue, msg, uptimeMillis);
}

 

而这个方法返回时执行了Handler::enqueueMessage方法

private boolean enqueueMessage(MessageQueue queue, Message msg, long uptimeMillis) {
    msg.target = this;
    if (mAsynchronous) {
        msg.setAsynchronous(true);
    }
    return queue.enqueueMessage(msg, uptimeMillis);
}

该方法再返回时又调用了MessageQueue::enqueueMessage方法。此时我们应该去到MessageQueue类中查看源码，

boolean enqueueMessage(Message msg, long when) {
    ......

    if (p == null || when == 0 || when < p.when) {
            // New head, wake up theevent queue if blocked.
            msg.next = p;
            mMessages = msg;
            needWake = mBlocked;
        } else {
            // Inserted within themiddle of the queue.  Usually we don'thave to wake
            // up the event queue unlessthere is a barrier at the head of the queue
            // and the message is theearliest asynchronous message in the queue.
            needWake = mBlocked && p.target == null &&msg.isAsynchronous();
            Message prev;
            for (;;) {
                prev = p;
                p = p.next;
                if (p == null || when < p.when) {
                    break;
                }
                if (needWake &&p.isAsynchronous()) {
                    needWake = false;
                }
            }
            msg.next = p; // invariant: p ==prev.next
            prev.next = msg;
        }
       ......
    }
    return true;
}

 

这个就是将发送来的消息保存到消息队列等待调度。之后的事情就交给Looper来做了。

public static void loop() {
    ......   

    final MessageQueue queue =me.mQueue;

    // Make sure the identityof this thread is that of the local process,
    // and keep track of what thatidentity token actually is.
    Binder.clearCallingIdentity();
    final long ident = Binder.clearCallingIdentity();

    for (;;) {
        Message msg = queue.next(); // might block
        if (msg == null) {
            // No message indicatesthat the message queue is quitting.
            return;
        }

        // This must be in a localvariable, in case a UI event sets the logger
        Printer logging =me.mLogging;
        if (logging != null) {
            logging.println(">>>>>Dispatching to " + msg.target + " " +
                    msg.callback + ": " + msg.what);
        }

        msg.target.dispatchMessage(msg);
       ......
    }
}

当Looper::loop()方法被调用以后队列就活起来了，不断地获取消息并分发消息。最后就是Handler的回调了

public void dispatchMessage(Message msg) {
    if (msg.callback != null) {
        handleCallback(msg);
    } else {
        if (mCallback != null) {
            if (mCallback.handleMessage(msg)) {
                return;
            }
        }
        handleMessage(msg);
    }
}

之后就回到了最开始我们提到的关于如何实例化一个Handler对象，之后覆写里面的Handler::handleMessage(msg)方法。整个消息循环就跑起来了。

这里最大的一个问题在于Looper::loop()中的死循环是如何跑起来的，为什么没有开启新线程执行Looper::loop(),但是也没有阻塞主线程？这个可能与Binder机制有关，待进一步研究。