我有一个这样的循环:

for (int i = 0; i < max; i++) {
    String myString = ...;
    float myNum = Float.parseFloat(myString);
    myFloats[i] = myNum;
}

这是一个方法的主要内容,该方法的唯一目的是返回浮点数数组。我想让这个方法在出现错误时返回null,所以我把循环放在try…Catch block,像这样:

try {
    for (int i = 0; i < max; i++) {
        String myString = ...;
        float myNum = Float.parseFloat(myString);
        myFloats[i] = myNum;
    }
} catch (NumberFormatException ex) {
    return null;
}

但后来我也想到试一试……Catch块在循环中,像这样:

for (int i = 0; i < max; i++) {
    String myString = ...;
    try {
        float myNum = Float.parseFloat(myString);
    } catch (NumberFormatException ex) {
        return null;
    }
    myFloats[i] = myNum;
}

是否有任何理由,性能或其他方面,更喜欢其中一个?


编辑:共识似乎是,将循环放在try/catch中更干净,可能是在它自己的方法中。然而,关于哪个速度更快仍存在争议。有人能测试一下并给出一个统一的答案吗?


当前回答

我想添加我自己的0.02c,关于在哪里放置异常处理的一般问题时两个相互竞争的考虑因素:

The "wider" the responsibility of the try-catch block (i.e. outside the loop in your case) means that when changing the code at some later point, you may mistakenly add a line which is handled by your existing catch block; possibly unintentionally. In your case, this is less likely because you are explicitly catching a NumberFormatException The "narrower" the responsibility of the try-catch block, the more difficult refactoring becomes. Particularly when (as in your case) you are executing a "non-local" instruction from within the catch block (the return null statement).

其他回答

好吧,在Jeffrey L Whitledge说没有性能差异之后(截至1997年),我去测试了一下。我运行了一个小的基准测试:

public class Main {

    private static final int NUM_TESTS = 100;
    private static int ITERATIONS = 1000000;
    // time counters
    private static long inTime = 0L;
    private static long aroundTime = 0L;

    public static void main(String[] args) {
        for (int i = 0; i < NUM_TESTS; i++) {
            test();
            ITERATIONS += 1; // so the tests don't always return the same number
        }
        System.out.println("Inside loop: " + (inTime/1000000.0) + " ms.");
        System.out.println("Around loop: " + (aroundTime/1000000.0) + " ms.");
    }
    public static void test() {
        aroundTime += testAround();
        inTime += testIn();
    }
    public static long testIn() {
        long start = System.nanoTime();
        Integer i = tryInLoop();
        long ret = System.nanoTime() - start;
        System.out.println(i); // don't optimize it away
        return ret;
    }
    public static long testAround() {
        long start = System.nanoTime();
        Integer i = tryAroundLoop();
        long ret = System.nanoTime() - start;
        System.out.println(i); // don't optimize it away
        return ret;
    }
    public static Integer tryInLoop() {
        int count = 0;
        for (int i = 0; i < ITERATIONS; i++) {
            try {
                count = Integer.parseInt(Integer.toString(count)) + 1;
            } catch (NumberFormatException ex) {
                return null;
            }
        }
        return count;
    }
    public static Integer tryAroundLoop() {
        int count = 0;
        try {
            for (int i = 0; i < ITERATIONS; i++) {
                count = Integer.parseInt(Integer.toString(count)) + 1;
            }
            return count;
        } catch (NumberFormatException ex) {
            return null;
        }
    }
}

我使用javap检查了结果字节码,以确保没有任何内容得到内联。

结果表明,假设JIT优化微不足道,Jeffrey是正确的;在Java 6、Sun客户端VM上绝对没有性能差异(我没有访问其他版本)。整个测试的总时间差在几毫秒的数量级上。

因此,唯一要考虑的是什么看起来最干净。我发现第二种方式很难看,所以我要么坚持第一种方式,要么坚持雷·海耶斯的方式。

如果它在内部,那么您将获得N次try/catch结构的开销,而不是只在外部获得一次。


每次调用Try/Catch结构都会增加方法执行的开销。只需要处理结构所需的一点点内存和处理器节拍。如果运行一个循环100次,假设每个try/catch调用的代价是1 tick,那么在循环内执行try/catch调用的代价是100 tick,而在循环外只执行1 tick。

我想添加我自己的0.02c,关于在哪里放置异常处理的一般问题时两个相互竞争的考虑因素:

The "wider" the responsibility of the try-catch block (i.e. outside the loop in your case) means that when changing the code at some later point, you may mistakenly add a line which is handled by your existing catch block; possibly unintentionally. In your case, this is less likely because you are explicitly catching a NumberFormatException The "narrower" the responsibility of the try-catch block, the more difficult refactoring becomes. Particularly when (as in your case) you are executing a "non-local" instruction from within the catch block (the return null statement).

为try/catch设置一个特殊的堆栈框架会增加额外的开销,但是JVM可能能够检测到您正在返回并优化它。

根据迭代次数的不同,性能差异可能可以忽略不计。

然而,我同意其他人的观点,把它放在循环之外会让循环体看起来更干净。

如果您希望继续处理,而不是在存在无效数字时退出,那么您将希望代码位于循环中。

你应该喜欢外面的版本而不是里面的版本。这只是规则的一个特定版本,将任何可以移动到循环外的东西移动到循环外。根据IL编译器和JIT编译器的不同,您的两个版本最终可能具有不同的性能特征,也可能没有。

另一方面,你可能应该看看float。TryParse或Convert.ToFloat。