我知道这是一个老问题，但通过快速阅读这里的回答，我并没有看到任何人提到同步方法有时可能是错误的锁。 摘自Java并发实践(第72页):

public class ListHelper<E> {
  public List<E> list = Collections.syncrhonizedList(new ArrayList<>());
...

public syncrhonized boolean putIfAbsent(E x) {
 boolean absent = !list.contains(x);
if(absent) {
 list.add(x);
}
return absent;
}

上面的代码看起来是线程安全的。然而，现实并非如此。在这种情况下，锁在类的实例上获得。但是，该列表可能被另一个不使用该方法的线程修改。正确的方法是使用

public boolean putIfAbsent(E x) {
 synchronized(list) {
  boolean absent = !list.contains(x);
  if(absent) {
    list.add(x);
  }
  return absent;
}
}

上面的代码将阻止所有试图修改list的线程修改列表，直到同步块完成。

注意:静态同步方法和块工作在Class对象上。

public class MyClass {
   // locks MyClass.class
   public static synchronized void foo() {
// do something
   }

   // similar
   public static void foo() {
      synchronized(MyClass.class) {
// do something
      }
   }
}

In general these are mostly the same other than being explicit about the object's monitor that's being used vs the implicit this object. One downside of synchronized methods that I think is sometimes overlooked is that in using the "this" reference to synchronize on you are leaving open the possibility of external objects locking on the same object. That can be a very subtle bug if you run into it. Synchronizing on an internal explicit Object or other existing field can avoid this issue, completely encapsulating the synchronization.

当java编译器将源代码转换为字节码时，它处理同步方法和同步块的方式非常不同。

当JVM执行一个同步方法时，执行线程识别出该方法的method_info结构体设置了ACC_SYNCHRONIZED标志，然后它自动获取对象的锁，调用该方法，并释放锁。如果发生异常，线程自动释放锁。

另一方面，同步方法块绕过了JVM对获取对象锁和异常处理的内置支持，并要求显式地用字节代码编写功能。如果读取带有同步块的方法的字节代码，您将看到十多个额外的操作来管理此功能。

这显示了生成同步方法和同步块的调用:

public class SynchronizationExample {
    private int i;

    public synchronized int synchronizedMethodGet() {
        return i;
    }

    public int synchronizedBlockGet() {
        synchronized( this ) {
            return i;
        }
    }
}

synchronizedMethodGet()方法生成以下字节代码:

0:  aload_0
1:  getfield
2:  nop
3:  iconst_m1
4:  ireturn

下面是synchronizedBlockGet()方法的字节代码:

0:  aload_0
1:  dup
2:  astore_1
3:  monitorenter
4:  aload_0
5:  getfield
6:  nop
7:  iconst_m1
8:  aload_1
9:  monitorexit
10: ireturn
11: astore_2
12: aload_1
13: monitorexit
14: aload_2
15: athrow

One significant difference between synchronized method and block is that, Synchronized block generally reduce scope of lock. As scope of lock is inversely proportional to performance, its always better to lock only critical section of code. One of the best example of using synchronized block is double checked locking in Singleton pattern where instead of locking whole getInstance() method we only lock critical section of code which is used to create Singleton instance. This improves performance drastically because locking is only required one or two times.

在使用同步方法时，如果混合使用静态同步方法和非静态同步方法，则需要格外小心。

Synchronizing with threads. 1) NEVER use synchronized(this) in a thread it doesn't work. Synchronizing with (this) uses the current thread as the locking thread object. Since each thread is independent of other threads, there is NO coordination of synchronization. 2) Tests of code show that in Java 1.6 on a Mac the method synchronization does not work. 3) synchronized(lockObj) where lockObj is a common shared object of all threads synchronizing on it will work. 4) ReenterantLock.lock() and .unlock() work. See Java tutorials for this.

The following code shows these points. It also contains the thread-safe Vector which would be substituted for the ArrayList, to show that many threads adding to a Vector do not lose any information, while the same with an ArrayList can lose information. 0) Current code shows loss of information due to race conditions A) Comment the current labeled A line, and uncomment the A line above it, then run, method loses data but it shouldn't. B) Reverse step A, uncomment B and // end block }. Then run to see results no loss of data C) Comment out B, uncomment C. Run, see synchronizing on (this) loses data, as expected. Don't have time to complete all the variations, hope this helps. If synchronizing on (this), or the method synchronization works, please state what version of Java and OS you tested. Thank you.

import java.util.*;

/** RaceCondition - Shows that when multiple threads compete for resources 
     thread one may grab the resource expecting to update a particular 
     area but is removed from the CPU before finishing.  Thread one still 
     points to that resource.  Then thread two grabs that resource and 
     completes the update.  Then thread one gets to complete the update, 
     which over writes thread two's work.
     DEMO:  1) Run as is - see missing counts from race condition, Run severa times, values change  
            2) Uncomment "synchronized(countLock){ }" - see counts work
            Synchronized creates a lock on that block of code, no other threads can 
            execute code within a block that another thread has a lock.
        3) Comment ArrayList, unComment Vector - See no loss in collection
            Vectors work like ArrayList, but Vectors are "Thread Safe"
         May use this code as long as attribution to the author remains intact.
     /mf
*/ 

public class RaceCondition {
    private ArrayList<Integer> raceList = new ArrayList<Integer>(); // simple add(#)
//  private Vector<Integer> raceList = new Vector<Integer>(); // simple add(#)

    private String countLock="lock";    // Object use for locking the raceCount
    private int raceCount = 0;        // simple add 1 to this counter
    private int MAX = 10000;        // Do this 10,000 times
    private int NUM_THREADS = 100;    // Create 100 threads

    public static void main(String [] args) {
    new RaceCondition();
    }

    public RaceCondition() {
    ArrayList<Thread> arT = new ArrayList<Thread>();

    // Create thread objects, add them to an array list
    for( int i=0; i<NUM_THREADS; i++){
        Thread rt = new RaceThread( ); // i );
        arT.add( rt );
    }

    // Start all object at once.
    for( Thread rt : arT ){
        rt.start();
    }

    // Wait for all threads to finish before we can print totals created by threads
    for( int i=0; i<NUM_THREADS; i++){
        try { arT.get(i).join(); }
        catch( InterruptedException ie ) { System.out.println("Interrupted thread "+i); }
    }

    // All threads finished, print the summary information.
    // (Try to print this informaiton without the join loop above)
    System.out.printf("\nRace condition, should have %,d. Really have %,d in array, and count of %,d.\n",
                MAX*NUM_THREADS, raceList.size(), raceCount );
    System.out.printf("Array lost %,d. Count lost %,d\n",
             MAX*NUM_THREADS-raceList.size(), MAX*NUM_THREADS-raceCount );
    }   // end RaceCondition constructor



    class RaceThread extends Thread {
    public void run() {
        for ( int i=0; i<MAX; i++){
        try {
            update( i );        
        }    // These  catches show when one thread steps on another's values
        catch( ArrayIndexOutOfBoundsException ai ){ System.out.print("A"); }
        catch( OutOfMemoryError oome ) { System.out.print("O"); }
        }
    }

    // so we don't lose counts, need to synchronize on some object, not primitive
    // Created "countLock" to show how this can work.
    // Comment out the synchronized and ending {, see that we lose counts.

//    public synchronized void update(int i){   // use A
    public void update(int i){                  // remove this when adding A
//      synchronized(countLock){            // or B
//      synchronized(this){             // or C
        raceCount = raceCount + 1;
        raceList.add( i );      // use Vector  
//          }           // end block for B or C
    }   // end update

    }   // end RaceThread inner class


} // end RaceCondition outter class

同步方法可以使用反射API进行检查。这对于测试某些契约很有用，比如模型中的所有方法都是同步的。

下面的代码段打印哈希表的所有同步方法:

for (Method m : Hashtable.class.getMethods()) {
        if (Modifier.isSynchronized(m.getModifiers())) {
            System.out.println(m);
        }
}