为什么java DirectoryStream执行得这么慢？

Question

我用nio包的DirectoryStreams专用的Streams做了一些测试。我只是尝试获取一个目录中所有文件的列表，该目录按上次修改的日期和大小排序。

JavaDoc of old File.listFiles()向文件的中的方法声明了一个备注：

注意，Files类定义了打开目录的newDirectoryStream方法，并遍历目录中文件的名称。当使用非常大的目录时，这可能会使用较少的资源。

我在下面多次运行代码(前三次)：

第一轮：

Run time of Arrays.sort: 1516
Run time of Stream.sorted as Array: 2912
Run time of Stream.sorted as List: 2875

第二轮：

Run time of Arrays.sort: 1557
Run time of Stream.sorted as Array: 2978
Run time of Stream.sorted as List: 2937

第三轮：

Run time of Arrays.sort: 1563
Run time of Stream.sorted as Array: 2919
Run time of Stream.sorted as List: 2896

我的问题是：为什么流表现这么差？

import java.io.File;
import java.io.IOException;
import java.io.UncheckedIOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.nio.file.attribute.FileTime;
import java.util.Arrays;
import java.util.Comparator;
import java.util.List;
import java.util.stream.Collectors;
import java.util.stream.Stream;

public class FileSorter {

  // This sorts from old to young and from big to small
  Comparator<Path> timeSizeComparator = (Path o1, Path o2) -> {
    int sorter = 0;
    try {
      FileTime lm1 = Files.getLastModifiedTime(o1);
      FileTime lm2 = Files.getLastModifiedTime(o2);
      if (lm2.compareTo(lm1) == 0) {
        Long s1 = Files.size(o1);
        Long s2 = Files.size(o2);
        sorter = s2.compareTo(s1);
      } else {
        sorter = lm1.compareTo(lm2);
      }
    } catch (IOException ex) {
      throw new UncheckedIOException(ex);
    }
    return sorter;
  };

  public String[] getSortedFileListAsArray(Path dir) throws IOException {
    Stream<Path> stream = Files.list(dir);
    return stream.sorted(timeSizeComparator).
            map(Path::getFileName).map(Path::toString).toArray(String[]::new);
  }

  public List<String> getSortedFileListAsList(Path dir) throws IOException {
    Stream<Path> stream = Files.list(dir);
    return stream.sorted(timeSizeComparator).
            map(Path::getFileName).map(Path::toString).collect(Collectors.
            toList());
  }

  public String[] sortByDateAndSize(File[] fileList) {
    Arrays.sort(fileList, (File o1, File o2) -> {
      int r = Long.compare(o1.lastModified(), o2.lastModified());
      if (r != 0) {
        return r;
      }
      return Long.compare(o1.length(), o2.length());
    });
    String[] fileNames = new String[fileList.length];
    for (int i = 0; i < fileNames.length; i++) {
      fileNames[i] = fileList[i].getName();
    }
    return fileNames;
  }

  public static void main(String[] args) throws IOException {
    // File (io package)
    File f = new File("C:\\Windows\\system32");
    // Path (nio package)
    Path dir = Paths.get("C:\\Windows\\system32");

    FileSorter fs = new FileSorter();

    long before = System.currentTimeMillis();
    String[] names = fs.sortByDateAndSize(f.listFiles());
    long after = System.currentTimeMillis();
    System.out.println("Run time of Arrays.sort: " + ((after - before)));

    long before2 = System.currentTimeMillis();
    String[] names2 = fs.getSortedFileListAsArray(dir);
    long after2 = System.currentTimeMillis();
    System.out.
            println("Run time of Stream.sorted as Array: " + ((after2 - before2)));

    long before3 = System.currentTimeMillis();
    List<String> names3 = fs.getSortedFileListAsList(dir);
    long after3 = System.currentTimeMillis();
    System.out.
            println("Run time of Stream.sorted as List: " + ((after3 - before3)));
  }
}

更新

在应用Peter的代码之后，我得到了以下结果：

Run time of Arrays.sort: 1615
Run time of Stream.sorted as Array: 3116
Run time of Stream.sorted as List: 3059
Run time of Stream.sorted as List with caching: 378

更新2

在对Peter的解决方案做了一些研究之后，我可以说，用for来读取文件属性。Files.getLastModified肯定是一个沉重的危机。只将比较国的这一部分改为：

  Comparator<Path> timeSizeComparator = (Path o1, Path o2) -> {
    File f1 = o1.toFile();
    File f2 = o2.toFile();
    long lm1 = f1.lastModified();
    long lm2 = f2.lastModified();
    int cmp = Long.compare(lm2, lm1);
    if (cmp == 0) {
      cmp = Long.compare(f2.length(), f1.length());
    }
    return cmp;
  };

在我的计算机上获取更好的结果：

Run time of Arrays.sort: 1968
Run time of Stream.sorted as Array: 1999
Run time of Stream.sorted as List: 1975
Run time of Stream.sorted as List with caching: 488

但是正如您所看到的，缓存对象是最好的方法。正如jtahlborn所提到的，它是一种稳定的类型。

更新3(我找到了最好的解决方案)

经过进一步的研究，我已经看到，方法Files.lastModified和Files.size都在同一件事情上做了大量的工作:属性。因此，我制作了三个版本的PathInfo类来测试：

1. 彼得斯版本，如下所述
2. 旧式文件版本，其中我在构造函数中执行一个Path.toFile()操作，并使用f.lastModified和f.length从该文件中获取所有值
3. Peters解决方案的一个版本，但现在我用Files.readAttributes(path，BasicFileAttributes.class)读取了一个属性对象，并完成了相关工作。

把所有这些都放在一个循环中，每个循环完成100次，我得到了以下结果：

After doing all hundred times
Mean performance of Peters solution: 432.26
Mean performance of old File solution: 343.11
Mean performance of read attribute object once solution: 255.66

PathInfo构造函数中用于最佳解决方案的代码：

public PathInfo(Path path) {
  try {
    // read the whole attributes once
    BasicFileAttributes bfa = Files.readAttributes(path, BasicFileAttributes.class);
    fileName = path.getFileName().toString();
    modified = bfa.lastModifiedTime().toMillis();
    size = bfa.size();
  } catch (IOException ex) {
    throw new UncheckedIOException(ex);
  }
}

，我的结果是：从来不会读取属性两次，而对象中的缓存则是非常好的性能。

Answer 1

Files.list()是O(N)操作，而排序是O(N log N)。更有可能的是，排序中的操作哪些重要。考虑到比较并不是一回事，这是最有可能的解释。C:/Windows/System32 32下有许多文件具有相同的修改日期，这意味着将经常检查大小。

为了说明大部分时间不是在FIles.list(dir)流中度过的，我优化了比较，以便每个文件只获得一次有关文件的数据。

import java.io.File;
import java.io.IOException;
import java.io.UncheckedIOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.nio.file.attribute.FileTime;
import java.util.Arrays;
import java.util.Comparator;
import java.util.List;
import java.util.stream.Collectors;
import java.util.stream.Stream;

public class FileSorter {

    // This sorts from old to young and from big to small
    Comparator<Path> timeSizeComparator = (Path o1, Path o2) -> {
        int sorter = 0;
        try {
            FileTime lm1 = Files.getLastModifiedTime(o1);
            FileTime lm2 = Files.getLastModifiedTime(o2);
            if (lm2.compareTo(lm1) == 0) {
                Long s1 = Files.size(o1);
                Long s2 = Files.size(o2);
                sorter = s2.compareTo(s1);
            } else {
                sorter = lm1.compareTo(lm2);
            }
        } catch (IOException ex) {
            throw new UncheckedIOException(ex);
        }
        return sorter;
    };

    public String[] getSortedFileListAsArray(Path dir) throws IOException {
        Stream<Path> stream = Files.list(dir);
        return stream.sorted(timeSizeComparator).
                map(Path::getFileName).map(Path::toString).toArray(String[]::new);
    }

    public List<String> getSortedFileListAsList(Path dir) throws IOException {
        Stream<Path> stream = Files.list(dir);
        return stream.sorted(timeSizeComparator).
                map(Path::getFileName).map(Path::toString).collect(Collectors.
                toList());
    }

    public String[] sortByDateAndSize(File[] fileList) {
        Arrays.sort(fileList, (File o1, File o2) -> {
            int r = Long.compare(o1.lastModified(), o2.lastModified());
            if (r != 0) {
                return r;
            }
            return Long.compare(o1.length(), o2.length());
        });
        String[] fileNames = new String[fileList.length];
        for (int i = 0; i < fileNames.length; i++) {
            fileNames[i] = fileList[i].getName();
        }
        return fileNames;
    }

    public List<String> getSortedFile(Path dir) throws IOException {
        return Files.list(dir).map(PathInfo::new).sorted().map(p -> p.getFileName()).collect(Collectors.toList());
    }

    static class PathInfo implements Comparable<PathInfo> {
        private final String fileName;
        private final long modified;
        private final long size;

        public PathInfo(Path path) {
            try {
                fileName = path.getFileName().toString();
                modified = Files.getLastModifiedTime(path).toMillis();
                size = Files.size(path);
            } catch (IOException ex) {
                throw new UncheckedIOException(ex);
            }
        }

        @Override
        public int compareTo(PathInfo o) {
            int cmp = Long.compare(modified, o.modified);
            if (cmp == 0)
                cmp = Long.compare(size, o.size);
            return cmp;
        }

        public String getFileName() {
            return fileName;
        }
    }

    public static void main(String[] args) throws IOException {
        // File (io package)
        File f = new File("C:\\Windows\\system32");
        // Path (nio package)
        Path dir = Paths.get("C:\\Windows\\system32");

        FileSorter fs = new FileSorter();

        long before = System.currentTimeMillis();
        String[] names = fs.sortByDateAndSize(f.listFiles());
        long after = System.currentTimeMillis();
        System.out.println("Run time of Arrays.sort: " + ((after - before)));

        long before2 = System.currentTimeMillis();
        String[] names2 = fs.getSortedFileListAsArray(dir);
        long after2 = System.currentTimeMillis();
        System.out.println("Run time of Stream.sorted as Array: " + ((after2 - before2)));

        long before3 = System.currentTimeMillis();
        List<String> names3 = fs.getSortedFileListAsList(dir);
        long after3 = System.currentTimeMillis();
        System.out.println("Run time of Stream.sorted as List: " + ((after3 - before3)));
        long before4 = System.currentTimeMillis();
        List<String> names4 = fs.getSortedFile(dir);
        long after4 = System.currentTimeMillis();
        System.out.println("Run time of Stream.sorted as List with caching: " + ((after4 - before4)));
    }
}

这是我笔记本上的指纹。

Run time of Arrays.sort: 1980
Run time of Stream.sorted as Array: 1295
Run time of Stream.sorted as List: 1228
Run time of Stream.sorted as List with caching: 185

如您所见，大约85%的时间用于反复获取文件的修改日期和大小。