Java SoftReference 保证失败

Question

JDK 7 文档对SoftReference有这样的说法：

"在虚拟机引发内存不足错误之前，保证已清除对软可访问对象的所有软引用。"

但是，在我的测试程序中，我始终看到OutOfMemoryError（下面的"陌生人行为"部分除外）：

// RefObjectTest.java
import java.util.*;
import java.lang.ref.*;
public class RefObjectTest {
    public static void main(String[] args) {
        ArrayList<byte[]> leaks = new ArrayList<>();
        byte[] obj = new byte[10 * 1024 * 1024];
        SoftReference<byte[]> ref = new SoftReference<>(obj);
        // WeakReference is supposed to be eagerly GC'ed, but I see no
        // difference in terms of program behavior: still get OOME.
        //
        // WeakReference<byte[]> ref = new WeakReference<>(obj);
        obj = null;
        while(true) {
            byte[] x = ref.get();
            if(x == null) {
                System.out.println("Referent stands garbage collected!!");
                break;
            } else {
                System.out.println("Referent still alive.");
            }
            // Leak memory in small, 10k increments. If soft reference
            // worked the way it's advertized, then just before the OOME, the 
            // 10MB referent should have been GC'ed to accomodate this small
            // 10K new memory request. But it doesn't appear to work that way!
//          try {
                leaks.add(new byte[10 * 1024]);
//          } catch(OutOfMemoryError e) {
//              System.out.println(ref.get() == null ? "Referent cleared" : 
//                  "Referent still alive");
//          }
            // VERY STRANGE: If I re-instate the above try-catch block, then no OOME is
            // thrown, and the program keeps printing "Referent still alive" lines
            // forever until I finally kill it with a Ctrl+C.
            // Uncommenting gc() tends to delay the OOME in terms of time, 
            // but OOME eventually does happen, and after the same number of
            // iterations of this loop. 
            // 
            // System.gc();
        }
    }
}

下面是输出：

$ java -Xmx15m RefObjectTest
Referent still alive.
  ...
Referent still alive.
Exception in thread "main" java.lang.OutOfMemoryError: Java heap space
        at RefObjectTest.main(RefObjectTest.java:38)

陌生人行为

非常奇怪的是，如果我重新恢复try-catch块，程序似乎永远运行良好，打印"Referent Still Alive."行，直到我累了并杀死它。

$ java -version
java version "1.7.0_45"
Java(TM) SE Runtime Environment (build 1.7.0_45-b18)
Java HotSpot(TM) 64-Bit Server VM (build 24.45-b08, mixed mode)
$ 
$ uname -a
Linux ida 3.10.11-100.fc18.x86_64 #1 SMP Mon Sep 9 13:06:31 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux

在这一切中，我错过了什么？

Answer 1

一点也不奇怪，分配失败，引发异常。通过捕获异常，您说程序可以继续，不捕获异常，您说您无法处理它并且程序退出。

在你的 while 循环中，你正在做：

        byte[] x = ref.get();

这是从您的软引用创建新的强引用。一旦您这样做，它就不再有资格收集。在继续执行新分配之前，您不会清除该强引用。

完成测试后，请x = null。