[Chapter 3] Threads

Threads provide a way for a Java program to do multiple tasks concurrently. A thread is essentially a flow of control in a program and is similar to the more familiar concept of a process. An operating system that can run more than one program at the same time uses processes to keep track of the various programs that it is running. However, processes generally do not share any state, while multiple threads within the same application share much of the same state. In particular, all of the threads in an application run in the same address space, sharing all resources except the stack. In concrete terms, this means that threads share field variables, but not local variables.

When multiple processes share a single processor, there are times when the operating system must stop the processor from running one process and start it running another process. The operating system must execute a sequence of events called a context switch to transfer control from one process to another. When a context switch occurs, the operating system has to save a lot of information for the process that is being paused and load the comparable information for the process being resumed. A context switch between two processes can require the execution of thousands of machine instructions. The Java virtual machine is responsible for handling context switches between threads in a Java program. Because threads share much of the same state, a context switch between two threads typically requires the execution of less than 100 machine instructions.

There are a number of situations where it makes sense to use threads in a Java program. Some programs must be able to engage in multiple activities and still be able to respond to additional input from the user. For example, a web browser should be able to respond to user input while fetching an image or playing a sound. Because threads can be suspended and resumed, they can make it easier to control multiple activities, even if the activities do not need to be concurrent. If a program models real world objects that display independent, autonomous behavior, it makes sense to use a separate thread for each object. Threads can also implement asynchronous methods, so that a calling method does not have to wait for the method it calls to complete before continuing with its own activity.

Java applets make considerable use of threads. For example, an animation is generally implemented with a separate thread. If an applet has to download extensive information, such as an image or a sound, to initialize itself, the initialization can take a long time. This initialization can be done in a separate thread to prevent the initialization from interfering with the display of the applet. If an applet needs to process messages from the network, that work generally is done in a separate thread so that the applet can continue painting itself on the screen and responding to mouse and keyboard events. In addition, if each message is processed separately, the applet uses a separate thread for each message.

For all of the reasons there are to use threads, there are also some compelling reasons not to use them. If a program uses inherently sequential logic, where one operation starts another operation and then must wait for the other operation to complete before continuing, one thread can implement the entire sequence. Using multiple threads in such a case results in a more complex program with no accompanying benefits. There is considerable overhead in creating and starting a thread, so if an operation involves only a few primitive statements, it is faster to handle it with a single thread. This can even be true when the operation is conceptually asynchronous. When multiple threads share objects, the objects must use synchronization mechanisms to coordinate thread access and maintain consistent state. Synchronization mechanisms add complexity to a program, can be difficult to tune for optimal performance, and can be a source of bugs.

3.1 Using Thread Objects

The Thread class in the java.lang package creates and controls threads in Java programs. The execution of Java code is always under the control of a Thread object. The Thread class provides a static method called currentThread() that provides a reference to the Thread object that controls the current thread of execution.

Associating a Method with a Thread

The first thing you need to do to make a Thread object useful is to associate it with a method you want it to run. Java provides two ways of associating a method with a Thread:

Declare a subclass of Thread that defines a run() method.
Pass a reference to an object that implements the Runnable interface to a Thread constructor.

For example, if you need to load the contents of a URL as part of an applet's initialization, but the applet can provide other functionality before the content is loaded, you might want to load the content in a separate thread. Here is a class that does just that:

import java.net.URL;
class UrlData extends Thread   {
    private Object data;
    private URL url
    public UrlData(String urlName) throws MalformedURLException {
        url = new URL(urlName);
        start();
    }
    public void run(){
        try {
            data = url.getContent();
        } catch (java.io.IOException  e) {
        }
    }
    public Object getUrlData(){
        return data;
    }
}

The UrlData class is declared as a subclass of Thread so that it can get the contents of the URL in a separate thread. The constructor creates a java.net.URL object to fetch the contents of the URL, and then calls the start() method to start the thread. Once the thread is started, the constructor returns; it does not wait for the contents of the URL to be fetched. The run() method is executed after the thread is started; it does the real work of fetching the data. The getUrlData() method is an access method that returns the value of the data variable. The value of this variable is null until the contents of the URL have been fetched, at which time it contains a reference to the actual data.

Subclassing the Thread class is convenient when the method you want to run in a separate thread does not need to belong to a particular class. Sometimes, however, you need the method to be part of a particular class that is a subclass of a class other than Thread. Say, for example, you want a graphical object that is displayed in a window to alternate its background color between red and blue once a second. The object that implements this behavior needs to be a subclass of the java.awt.Canvas class. However, at the same time, you need a separate thread to alternate the color of the object once a second.

In this situation, you want to tell a Thread object to run code in another object that is not a subclass of the Thread class. You can accomplish this by passing a reference to an object that implements the Runnable interface to the constructor of the Thread class. The Runnable interface requires that an object has a public method called run() that takes no arguments. When a Runnable object is passed to the constructor of the Thread class, it creates a Thread object that calls the Runnable object's run() method when the thread is started. The following example shows part of the code that implements an object that alternates its background color between red and blue once a second:

class AutoColorChange extends java.awt.Canvas implements Runnable {
    private Thread myThread;
    AutoColorChange () {
        myThread = new Thread(this);
        myThread.start();
        ...
    }
    public void run() {
        while (true) {
            setBackground(java.awt.Color.red);
            repaint();
            try {
                myThread.sleep(1000);
            } catch (InterruptedException e) {}
            setBackground(java.awt.Color.blue);
            repaint();
            try {
                myThread.sleep(1000);
            } catch (InterruptedException e) {}
        }
    }
}

The AutoChangeColor class extends java.awt.Canvas, alternating the background color between red and blue once a second. The constructor creates a new Thread by passing the current object to the Thread constructor, which tells the Thread to call the run() method in the AutoChangeColor class. The constructor then starts the new thread by calling its start() method, so that the color change happens asynchronously of whatever else is going on. The class has an instance variable called myThread that contains a reference to the Thread object, so that can control the thread. The run() method takes care of changing the background color, using the sleep() method of the Thread class to temporarily suspend the thread and calling repaint() to redisplay the object after each color change.

Controlling a Thread

As shown in the previous section, you start a Thread by calling its start() method. Before the start() method is called, the isAlive() method of the Thread object always returns false. When the start() method is called, the Thread object becomes associated with a scheduled thread in the underlying environment. After the start() method has returned, the isAlive() method always returns true. The Thread is now scheduled to run until it dies, unless it is suspended or in another unrunnable state.

It is actually possible for isAlive() to return true before start() returns, but not before start() is called. This can happen because the start() method can return either before the started Thread begins to run or after it begins to run. In other words, the method that called start() and the new thread are now running concurrently. On a multiprocessor system, the start() method can even return at the same time the started Thread begins to run.

Thread objects have a parent-child relationship. The first thread created in a Java environment does not have a parent Thread. However, after the first Thread object is created, the Thread object that controls the thread used to create another Thread object is considered to be the parent of the newly created Thread. This parent-child relationship is used to supply some default values when a Thread object is created, but it has no further significance after a Thread has been created.

Stopping a thread

A thread dies when one of the following things happens:

The run() method called by the Thread returns.
An exception is thrown that causes the run() method to be exited.
The stop() method of the Thread is called.

The stop() method of the Thread class works by throwing a ThreadDeath object in the run() method of the thread. Normally, you should not catch ThreadDeath objects in a try statement. If you need to catch ThreadDeath objects to detect that a Thread is about to die, the try statement that catches ThreadDeath objects should rethrow them.

When an object (ThreadDeath or otherwise) is thrown out of the run() method for the Thread, the uncaughtException() method of the ThreadGroup for that Thread is called. If the thrown object is an instance of the ThreadDeath class, the thread dies, and the thrown object is ignored. Otherwise, if the thrown object is of any other class, uncaughtException() calls the thrown object's printStackTrace() method, the thread dies, and the thrown object is ignored. In either case, if there are other nondaemon threads running in the system, the current program continues to run.

Interrupting a thread

There are a number of methods in the Java API, such as wait() and join(), that are declared as throwing an InterruptedException. What these methods have in common is that they temporarily suspend the execution of a thread. In Java 1.1, if a thread is waiting for one of these methods to return and another thread calls interrupt() on the waiting thread, the method that is waiting throws an InterruptedException.

The interrupt() method sets an internal flag in a Thread object. Before the interrupt() method is called, the isInterrupted() method of the Thread object always returns false. After the interrupt() method is called, isInterrupted() returns true.

Prior to version 1.1, the methods in the Java API that are declared as throwing an InterruptedException do not actually do so. However, the isInterrupted() method does function as described above. Thus, if the code in the run() method for a thread periodically calls isInterrupted(), the thread can respond to a call to interrupt() by shutting down in an orderly fashion.

Thread priority

One of the attributes that controls the behavior of a thread is its priority. Although Java does not guarantee much about how threads are scheduled, it does guarantee that a thread with a priority that is higher than that of another thread will be scheduled to run at least as often, and possibly more often, than the thread with the lower priority. The priority of a thread is set when the Thread object is created, by passing an argument to the constructor that creates the Thread object. If an explicit priority is not specified, the Thread inherits the priority of its parent Thread object.

You can query the priority of a Thread object by calling its getPriority() method. Similarly, you can set the priority of a Thread using its setPriority() method. The priority you specify must be greater than or equal to Thread.MIN_PRIORITY and less than or equal to Thread.MAX_PRIORITY.

Before actually setting the priority of a Thread object, the setPriority() method checks the maximum allowable priority for the ThreadGroup that contains the Thread by calling getMaxPriority() on the ThreadGroup. If the call to setPriority() tries to set the priority to a value that is higher than the maximum allowable priority for the ThreadGroup, the priority is instead set to the maximum priority. It is possible for the current priority of a Thread to be greater than the maximum allowable priority for the ThreadGroup. In this case, an attempt to raise the priority of the Thread results in its priority being lowered to the maximum priority.

Daemon threads

A daemon thread is a thread that runs continuously to perform a service, without having any connection with the overall state of the program. For example, the thread that runs the garbage collector in Java is a daemon thread. The thread that processes mouse events for a Java program is also a daemon thread. In general, threads that run application code are not daemon threads, and threads that run system code are daemon threads. If a thread dies and there are no other threads except daemon threads alive, the Java virtual machine stops.

A Thread object has a boolean attribute that specifies whether or not a thread is a daemon thread. The daemon attribute of a thread is set when the Thread object is created, by passing an argument to the constructor that creates the Thread object. If the daemon attribute is not explicitly specified, the Thread inherits the daemon attribute of its parent Thread object.

The daemon attribute is queried using the isDaemon() method; it is set using the setDaemon() method.

Yielding

When a thread has nothing to do, it can call the yield() method of its Thread object. This method tells the scheduler to run a different thread. The value of calling yield() depends largely on whether the scheduling mechanism for the platform on which the program is running is preemptive or nonpreemptive.

By choosing a maximum length of time a thread can continuously, a preemptive scheduling mechanism guarantees that no single thread uses more than its fair share of the processor. If a thread runs for that amount of time without yielding control to another thread, the scheduler preempts the thread and causes it to stop running so that another thread can run.

A nonpreemptive scheduling mechanism cannot preempt threads. A nonpreemptive scheduler relies on the individual threads to yield control of the processor frequently, so that it can provide reasonable performance. A thread explicitly yields control by calling the Thread object's yield() method. More often, however, a thread implicitly yields control when it is forced to wait for something to happen elsewhere.

Calling a Thread object's yield() method during a lengthy computation can be quite valuable on a platform that uses a nonpreemptive scheduling mechanism, as it allows other threads to run. Otherwise, the lengthy computation can prevent other threads from running. On a platform that uses a preemptive scheduling mechanism, calling yield() does not usually make any noticeable difference in the responsiveness of threads.

Regardless of the scheduling algorithm that is being used, you should not make any assumptions about when a thread will be scheduled to run again after it has called yield(). If you want to prevent a thread from being scheduled to run until a specified amount of time has elapsed, you should call the sleep() method of the Thread object. The sleep() method takes an argument that specifies a minimum number of milliseconds that must elapse before the thread can be scheduled to run again.

Controlling groups of threads

Sometimes is it necessary to control multiple threads at the same time. Java provides the ThreadGroup class for this purpose. Every Thread object belongs to a ThreadGroup object. By passing an argument to the constructor that creates the Thread object, the ThreadGroup of a thread can be set when the Thread object is created. If an explicit ThreadGroup is not specified, the Thread belongs to the same ThreadGroup as its parent Thread object.