Java基准测试与Ellipticgroup / Brent Boyer



我正在尝试使用这里提到的java基准测试:https://www.ibm.com/developerworks/java/library/j-benchmark2/可以在这里下载:http://www.ellipticgroup.com/html/benchmarkingArticle.html

我尝试使用上面文章中的基本示例:

import bb.util.Benchmark;
public class ShortIndexesLoop {
    public static void main(String[] args) throws Exception {
        Callable<Integer> task = 
            new Callable<Integer>() { public Integer call() { return fibonacci(35); } };
            System.out.println("fibonacci(35): " + new Benchmark(task));
    }
    protected static int fibonacci(int n) throws IllegalArgumentException {
        if (n < 0) throw new IllegalArgumentException("n = " + n + " < 0");
        if (n <= 1) return n;
        return fibonacci(n - 1) + fibonacci(n - 2);
    }
}

但是这个基准测试的每次执行在运行基准测试之前都从一个异常开始

2015-6-13下午01:45:37 StringUtil警告:字符串的行为不符合预期;查看原因

java.lang.Exception: substring does NOT share the same underlying char[] with its parent String
    at bb.util.StringUtil.inspectStringConstructor(StringUtil.java:84)
    at bb.util.StringUtil.<clinit>(StringUtil.java:75)
    at bb.io.ConsoleUtil.<clinit>(ConsoleUtil.java:81)
    at bb.util.Benchmark.sendUserMsg(Benchmark.java:1002)
    at bb.util.Benchmark.osSpecificPreparation(Benchmark.java:579)
    at bb.util.Benchmark.perform(Benchmark.java:541)
    at bb.util.Benchmark.<init>(Benchmark.java:464)
    at bb.util.Benchmark.<init>(Benchmark.java:439)
    at ShortIndexesLoop.main(fib.java:13)

似乎对基准测试的任何调用最终都会在 StringUtil 中使用以下方法

private static final boolean stringContructorTrimsGarbage = inspectStringConstructor();
private static boolean inspectStringConstructor() {
    try {
            // first prove that substring shares the same underlying char[] as the parent String:
        String s1 = "abc123def";
        String s2 = s1.substring(3, 6);
        char[] value1 = (char[]) ReflectUtil.get(s1, "value");
        char[] value2 = (char[]) ReflectUtil.get(s2, "value");
        if (value1 != value2) throw new Exception("substring does NOT share the same underlying char[] with its parent String");
        if (value2.length != s1.length()) throw new Exception("value2.length = " + value2.length + " != s1.length() = " + s1.length());
            // second prove that the String(String) constructor trims garbage chars:
        String s3 = new String(s2);
        char[] value3 = (char[]) ReflectUtil.get(s3, "value");
        if (value3 == value2) throw new Exception("new String shares the same underlying char[] with its String arg");
        if (!(value3.length < value2.length)) throw new Exception("value3.length = " + value3.length + " is not < value2.length = " + value2.length);
        return true;
    }
    catch (Exception e) {
        LogUtil.getLogger2().logp(Level.WARNING, "StringUtil", "<clinit>", "String does not behave as expected; see cause", e);
        return false;
    }
}

我真的不介意例外,因为基准测试库确实给出了结果。但是,尝试将此代码编译为 jar 并将其作为 jar 运行是一个问题。由于异常,其余代码将不会在终端中执行

有人知道如何解决这个问题吗?

我知道

这是一个老问题,但是mabye我的解决方案对其他人也很有用:在这个基准框架的StringUtil类的源代码中,记录了为什么要进行此检查。基于此,如果您有一个 JDK,其中 substring 创建了底层 char 数组相关部分的副本,您可以简单地注释 stringContructorTrimsGarbage 标志,而在 newString 方法中,您只需返回 s。