下面是我在Hibernate/SQL项目的服务器中编写的断言。一个实体bean有两个有效的布尔属性，称为isActive和isDefault。每个都可以有一个值“Y”或“N”或null，被视为“N”。我们希望确保浏览器客户端仅限于这三个值。所以，在这两个属性的setter中，我添加了这样的断言:

assert new HashSet<String>(Arrays.asList("Y", "N", null)).contains(value) : value;

注意以下几点。

This assertion is for the development phase only. If the client sends a bad value, we will catch that early and fix it, long before we reach production. Assertions are for defects that you can catch early. This assertion is slow and inefficient. That's okay. Assertions are free to be slow. We don't care because they're development-only tools. This won't slow down the production code because assertions will be disabled. (There's some disagreement on this point, which I'll get to later.) This leads to my next point. This assertion has no side effects. I could have tested my value against an unmodifiable static final Set, but that set would have stayed around in production, where it would never get used. This assertion exists to verify the proper operation of the client. So by the time we reach production, we will be sure that the client is operating properly, so we can safely turn the assertion off. Some people ask this: If the assertion isn't needed in production, why not just take them out when you're done? Because you'll still need them when you start working on the next version.

Some people have argued that you should never use assertions, because you can never be sure that all the bugs are gone, so you need to keep them around even in production. And so there's no point in using the assert statement, since the only advantage to asserts is that you can turn them off. Hence, according to this thinking, you should (almost) never use asserts. I disagree. It's certainly true that if a test belongs in production, you should not use an assert. But this test does not belong in production. This one is for catching a bug that's not likely to ever reach production, so it may safely be turned off when you're done.

顺便说一句，我可以这样写:

assert value == null || value.equals("Y") || value.equals("N") : value;

对于只有三个值是可以的，但是如果可能值的数量变大，HashSet版本就会变得更方便。我选择HashSet版本来说明我关于效率的观点。

这是另一个例子。我写了一个方法来查找两个排序数组中值的中位数。该方法假设数组已经排序。出于性能考虑，它不应该首先对数组排序，甚至不应该检查以确保它们已排序。然而，对未排序的数据调用此方法是一个严重的错误，我们希望在开发阶段尽早发现这些错误。下面是我处理这些看似矛盾的目标的方法:

public static int medianOf(int[] a, int[] b) {
  assert assertionOnlyIsSorted(a);      // Assertion is order n
  assert assertionOnlyIsSorted(b);
  ... // rest of implementation goes here. Algorithm is order log(n)
}

public static boolean assertionOnlyIsSorted(int[] array) {
  for (int i=1; i<array.length; ++i) {
    if (array[i] < array[i-1]) {
      return false;
    }
    return true;
  }
}

这样，缓慢的测试只在开发阶段执行，在开发阶段，速度没有捕获错误重要。您希望medianOf()方法具有log(n)性能，但“is sorted”测试是o (n)。因此，我将其放在断言中，以限制其在开发阶段的使用，并为其命名，以明确表示它不适合生产。

这样我就两全其美了。在开发过程中，我知道任何不正确地调用这个函数的方法都会被捕获并修复。而且我知道这样做的缓慢测试不会影响生产中的性能。(这也很好地说明了为什么要在生产环境中关闭断言，而在开发环境中启用断言。)

除了这里提供的所有很棒的答案之外，官方Java SE 7编程指南还提供了关于使用assert的非常简洁的手册;通过几个准确的例子，说明了何时使用断言是一个好主意(重要的是，也是坏主意)，以及它与抛出异常有何不同。

Link

在Java中assert关键字是做什么的?

让我们看看编译后的字节码。

我们将得出结论:

public class Assert {
    public static void main(String[] args) {
        assert System.currentTimeMillis() == 0L;
    }
}

生成几乎完全相同的字节码:

public class Assert {
    static final boolean $assertionsDisabled =
        !Assert.class.desiredAssertionStatus();
    public static void main(String[] args) {
        if (!$assertionsDisabled) {
            if (System.currentTimeMillis() != 0L) {
                throw new AssertionError();
            }
        }
    }
}

其中Assert.class.desiredAssertionStatus()在命令行传递-ea时为true，否则为false。

我们使用System.currentTimeMillis()来确保它不会被优化掉(assert true;所做的那样)。

合成字段生成后，Java只需要在加载时调用Assert.class.desiredAssertionStatus()一次，然后将结果缓存到那里。参见:“静态合成”是什么意思?

我们可以用以下方法验证:

javac Assert.java
javap -c -constants -private -verbose Assert.class

在Oracle JDK 1.8.0_45中，生成了一个合成的静态字段(参见:“静态合成”是什么意思?):

static final boolean $assertionsDisabled;
  descriptor: Z
  flags: ACC_STATIC, ACC_FINAL, ACC_SYNTHETIC

与静态初始化项一起使用:

 0: ldc           #6                  // class Assert
 2: invokevirtual #7                  // Method java/lang Class.desiredAssertionStatus:()Z
 5: ifne          12
 8: iconst_1
 9: goto          13
12: iconst_0
13: putstatic     #2                  // Field $assertionsDisabled:Z
16: return

主要方法是:

 0: getstatic     #2                  // Field $assertionsDisabled:Z
 3: ifne          22
 6: invokestatic  #3                  // Method java/lang/System.currentTimeMillis:()J
 9: lconst_0
10: lcmp
11: ifeq          22
14: new           #4                  // class java/lang/AssertionError
17: dup
18: invokespecial #5                  // Method java/lang/AssertionError."<init>":()V
21: athrow
22: return

我们的结论是:

assert没有字节码级别的支持:它是Java语言的概念 assert可以用系统属性-Pcom.me很好地模拟。assert=true替换命令行上的-ea，并抛出新的AssertionError()。

断言是用于捕获代码中的错误的开发阶段工具。它们被设计为易于删除，因此它们不会存在于生产代码中。因此断言不是您交付给客户的“解决方案”的一部分。它们是内部检查，以确保你所做的假设是正确的。最常见的例子是测试是否为空。很多方法都是这样写的:

void doSomething(Widget widget) {
  if (widget != null) {
    widget.someMethod(); // ...
    ... // do more stuff with this widget
  }
}

在这样的方法中，小部件通常不应该是空的。所以如果它是空的，在你的代码中有一个bug，你需要追踪。但是上面的代码永远不会告诉你这一点。因此，在编写“安全”代码的善意努力中，您也隐藏了一个错误。这样写代码会更好:

/**
 * @param Widget widget Should never be null
 */
void doSomething(Widget widget) {
  assert widget != null;
  widget.someMethod(); // ...
    ... // do more stuff with this widget
}

这样，您一定能尽早发现这个错误。(在合同中指定这个参数永远不应该为空也是有用的。)在开发过程中测试代码时，一定要打开断言。(说服你的同事这样做通常也很困难，我觉得这很烦人。)

现在，您的一些同事会反对这段代码，认为您仍然应该放入null检查，以防止生产中出现异常。在这种情况下，断言仍然有用。你可以这样写:

void doSomething(Widget widget) {
  assert widget != null;
  if (widget != null) {
    widget.someMethod(); // ...
    ... // do more stuff with this widget
  }
}

这样，您的同事就会高兴地看到产品代码有空检查，但在开发过程中，当小部件为空时，您就不再隐藏错误了。

这里有一个真实的例子:我曾经写过一个方法，比较两个任意值是否相等，其中任何一个值都可以为空:

/**
 * Compare two values using equals(), after checking for null.
 * @param thisValue (may be null)
 * @param otherValue (may be null)
 * @return True if they are both null or if equals() returns true
 */
public static boolean compare(final Object thisValue, final Object otherValue) {
  boolean result;
  if (thisValue == null) {
    result = otherValue == null;
  } else {
    result = thisValue.equals(otherValue);
  }
  return result;
}

这段代码在thisValue不为空的情况下委托equals()方法的工作。但它假设equals()方法通过正确处理空参数正确地实现了equals()的契约。

一位同事反对我的代码，告诉我我们的许多类都有不测试null的equals()方法，所以我应该把这个检查放到这个方法中。这是否是明智的，或者我们是否应该强制错误，这样我们就可以发现并修复它，这是有争议的，但我听从了我同事的意见，放入了一个空检查，我已经标记了一个注释:

public static boolean compare(final Object thisValue, final Object otherValue) {
  boolean result;
  if (thisValue == null) {
    result = otherValue == null;
  } else {
    result = otherValue != null && thisValue.equals(otherValue); // questionable null check
  }
  return result;
}

这里的额外检查other != null仅在equals()方法不能按照其契约要求检查null时才有必要。

我没有与同事就让有bug的代码留在代码库中是否明智进行毫无结果的争论，而是简单地在代码中放入了两个断言。这些断言会让我知道，在开发阶段，如果我们的一个类不能正确地实现equals()，所以我可以修复它:

public static boolean compare(final Object thisValue, final Object otherValue) {
  boolean result;
  if (thisValue == null) {
    result = otherValue == null;
    assert otherValue == null || otherValue.equals(null) == false;
  } else {
    result = otherValue != null && thisValue.equals(otherValue);
    assert thisValue.equals(null) == false;
  }
  return result;
}

需要记住的要点如下:

Assertions are development-phase tools only. The point of an assertion is to let you know if there's a bug, not just in your code, but in your code base. (The assertions here will actually flag bugs in other classes.) Even if my colleague was confident that our classes were properly written, the assertions here would still be useful. New classes will be added that might fail to test for null, and this method can flag those bugs for us. In development, you should always turn assertions on, even if the code you've written doesn't use assertions. My IDE is set to always do this by default for any new executable. The assertions don't change the behavior of the code in production, so my colleague is happy that the null check is there, and that this method will execute properly even if the equals() method is buggy. I'm happy because I will catch any buggy equals() method in development.

此外，您应该通过放入一个将失败的临时断言来测试断言策略，这样您就可以确定通过日志文件或输出流中的堆栈跟踪来通知您。

总结一下(不仅仅是Java，很多语言都是如此):

“assert”主要被软件开发人员在调试过程中用作调试辅助。断言消息永远不会出现。许多语言提供了一个编译时选项，该选项将导致所有“断言”被忽略，用于生成“生产”代码。

"exceptions" are a handy way to handle all kinds of error conditions, whether or not they represent logic errors, because, if you run into an error-condition such that you cannot continue, you can simply "throw them up into the air," from wherever you are, expecting someone else out there to be ready to "catch" them. Control is transferred in one step, straight from the code that threw the exception, straight to the catcher's mitt. (And the catcher can see the complete backtrace of calls that had taken place.)

此外，该子例程的调用者不必检查子例程是否成功:“如果我们现在在这里，它一定成功了，因为否则它会抛出异常，我们现在就不会在这里!”这种简单的策略使得代码设计和调试变得非常非常容易。

异常方便地允许致命错误条件保持原样:“规则的异常”。而且，对于它们来说，由代码路径来处理，这也是“规则的例外……“高飞球!”