Transient 和volatile

(原文来自http://www.devx.com/tips/Tip/13726)

Expertise: Intermediate
Language: Java

January 28, 2000

Be Careful With Transient Data

Java's serialization provides an elegant, and easy to use mechanism for making an object's state persistent. While controlling object serialization, we might have a particular object data member that we do not want the serialization mechanism to save.

To turn off serialization on a certain field of an object, we tag that field of the class of our object with the Java's "transient" keyword. This, to low-level parts of the Java virtual machine, is an indication that the transient variable is not part of the persistent state of an object.

First, let's have some backgrounder code with Java's serialization.
Suppose we define a class as:

 
public class LoggingInfo implements java.io.Serializable 
{ 
    private Date loggingDate = new Date(); 
    private String uid; 
    private transient String pwd; 
    
    LoggingInfo(String user, String password) 
    { 
        uid = user; 
        pwd = password; 
    } 
    public String toString() 
    { 
        String password=null; 
        if(pwd == null) 
        { 
        password = "NOT SET"; 
        } 
        else 
        { 
            password = pwd; 
        } 
        return "logon info: \n   " + "user: " + uid + 
            "\n   logging date : " + loggingDate.toString() + 
            "\n   password: " + password; 
    } 
}
 

Now we can create an instance of this class and serialize it, and write the serialized object to disk as in:

 
LoggingInfo logInfo = new LoggingInfo("MIKE", "MECHANICS"); 
System.out.println(logInfo.toString()); 
try 
{ 
   ObjectOutputStream o = new ObjectOutputStream( 
                new FileOutputStream("logInfo.out")); 
   o.writeObject(logInfo); 
   o.close(); 
} 
catch(Exception e) {//deal with exception} 

To read the object back, we can write

 
try 
{ 
   ObjectInputStream in =new ObjectInputStream( 
                new FileInputStream("logInfo.out")); 
   LoggingInfo logInfo = (LoggingInfo)in.readObject(); 
   System.out.println(logInfo.toString()); 
} 
catch(Exception e) {//deal with exception} 

If we run this code, we notice that the read-back object prints password as "NOT SET". This is exactly the effect we should have expected when we declared the pwd field as transient.

Now, let's see a potential problem that careless treatment of transient fields may cause. Suppose we modify our class definition and provide default values for the transient field, say we write:

 
public class GuestLoggingInfo implements java.io.Serializable 
{ 
    private Date loggingDate = new Date(); 
    private String uid; 
    private transient String pwd; 
    
    GuestLoggingInfo() 
    { 
        uid = "guest"; 
        pwd = "guest"; 
    } 
    public String toString() 
    { 
        //same as above 
     } 
} 

Now, if we serialize an instance of GuestLoggingInfo, write it to disk, and read it back, we still see that the read-back object prints password as "NOT SET". In effect, the process of reading back (de-serializing) totally ignores the constructor of GuestLoggingInfo. So what happened?

The answer lies in the fact that the initialization code is not called because we are not initializing, in other words, we are not constructing a brand new object, but loading back the persistent state of an object of a class, and assigning that state to another object of the same class. Declaring the pwd field as transient, excludes the data for that field from the persistent state of our object. Then, upon de-serialization, since there is no data preserved for the pwd field, the field gets Java's default value for its type (null for String).

So, if you mark a field of an object as transient, and write that object to disk, expect to have the default value of the type of that field when you de-serialize the object, and not the actual value that the field had before its state was serialized. If a default value (or any meaningful value) is essential for a transient field of a de-serialized object, you have to assign it yourself either directly (if the field is public) or via a setter method.

Behrouz Fallahi

Java语言的关键字，用来表示一个域不是该对象串行化的一部分。当一个对象被串行化的时候，transient型变量的值不包括在串行化的表示中，然而非transient型的变量是被包括进去的。Serializable接口不实现任何方法，仅仅是做个标识。

  Volatile修饰的成员变量在每次被线程访问时，都强迫从共享内存中重读该成员变量的值。而且，当成员变量发生变化时，强迫线程将变化值回写到共享内存。这样在任何时刻，两个不同的线程总是看到某个成员变量的同一个值。
       Java语言规范中指出：为了获得最佳速度，允许线程保存共享成员变量的私有拷贝，而且只当线程进入或者离开同步代码块时才与共享成员变量的原始值对比。
       这样当多个线程同时与某个对象交互时，就必须要注意到要让线程及时的得到共享成员变量的变化。
而volatile关键字就是提示VM：对于这个成员变量不能保存它的私有拷贝，而应直接与共享成员变量交互。
使用建议：在两个或者更多的线程访问的成员变量上使用volatile。当要访问的变量已在synchronized代码块中，或者为常量时，不必使用。
       由于使用volatile屏蔽掉了VM中必要的代码优化，所以在效率上比较低，因此一定在必要时才使用此关键字。
=================================================================================
       线程为了提高效率，将某成员变量(如A)拷贝了一份（如B），线程中对A的访问其实访问的是B。
      只在某些动作时才进行A和B的同步。因此存在A和B不一致的情况。volatile就是用来避免这种情况的。
      volatile告诉jvm， 它所修饰的变量不保留拷贝，直接访问主内存中的（也就是上面说的A)