What is the fastest way to find out how many non-empty lines are in a file, using Java?

Question

What is the fastest way to find out how many non-empty lines are in a file, using Java?

Answer 1

The easiest would be with a scanner (yes I like verbose code... you can make it physically shorter). Scanner() also takes File, Reader, etc... so you can pass it whatever you have.

import java.util.Scanner;


public class Main
{
    public static void main(final String[] argv)
    {
        final Scanner scanner;
        final int     lines;

        scanner = new Scanner("Hello\n\n\nEvil\n\nWorld");
        lines   = countLines(scanner);
        System.out.println("lines = "  + lines);
    }

    private static int countLines(final Scanner scanner)
    {
        int lines;

        lines = 0;

        while(scanner.hasNextLine())
        {
            final String line;

            line = scanner.nextLine();

            if(line.length() > 0)
            {
                lines++;
            }
        }

        return lines;
    }
}

Answer 2

The easiest way would be to use BufferedReader, and check which lines are empty. However, this is a relatively slow way, because it needs to create a String object for every line in the file. A faster way would be to read the file into arrays using read(), and then iterate through the arrays to count for line breaks.

Here's the code for the two options; the second one took about 50% of the time on my machine.

public static void timeBufferedReader () throws IOException
{
    long bef = System.currentTimeMillis ();

    // The reader buffer size is the same as the array size I use in the other function
    BufferedReader reader = new BufferedReader(new FileReader("test.txt"), 1024 * 10);
    int counter = 0;
    while (reader.ready())
    {
        if (reader.readLine().length() > 0)
            counter++;
    }

    long after = System.currentTimeMillis() - bef;

    System.out.println("Time: " + after + " Result: " + counter);

}

public static void timeFileReader () throws IOException
{
    long bef = System.currentTimeMillis();

    FileReader reader = new FileReader("test.txt");
    char[] buf = new char[1024 * 10];
    boolean emptyLine = true;
    int     counter = 0;
    while (reader.ready())
    {
        int len = reader.read(buf,0,buf.length);
        for (int i = 0; i < len; i++)
        {
            if (buf[i] == '\r' || buf[i] == '\n')
            {
                if (!emptyLine)
                {
                    counter += 1;
                    emptyLine = true;
                }
            }
            else emptyLine = false;
        }
    }

    long after = System.currentTimeMillis() - bef;

    System.out.println("Time: " + after + " Result: " + counter);

}

Answer 3

If it really must be the fastest possible, you should look into NIO. And then, test your code on your target platform to see if it's really and truly better using NIO. I was able to get an order of magnitude improvement in some code I was playing with for the Netflix Prize. It involved parsing thousands of files into a more compact, quick-loading binary format. NIO was a big help on my (slow) development laptop.

Answer 4

I am with Limbic System on the NIO recommendation. I've added a NIO method to Daphna's test code and bench marked it against his two methods:

public static void timeNioReader () throws IOException {
    long bef = System.currentTimeMillis();

    File file = new File("/Users/stu/test.txt");
    FileChannel fc = (new FileInputStream(file)).getChannel(); 
    MappedByteBuffer buf = fc.map(MapMode.READ_ONLY, 0, file.length());
    boolean emptyLine = true;
    int     counter = 0;

    while (buf.hasRemaining())
    {
        byte element = buf.get();

        if (element == '\r' || element == '\n') {
            if (!emptyLine) {
                counter += 1;
                emptyLine = true;
            }
        } else 
            emptyLine = false;

    }

    long after = System.currentTimeMillis() - bef;

    System.out.println("timeNioReader      Time: " + after + " Result: " + counter);

}

Here are the warmed up results for a 89MB file:

timeBufferedReader Time: 947 Result: 747656
timeFileReader     Time: 670 Result: 747656
timeNioReader      Time: 251 Result: 747656

NIO is 2.5x faster than FileReader and 4x fastser than the BufferedReader!

With a 6.4MB file the results are even better, although the warm up time is much longer.

//jvm start, warming up
timeBufferedReader Time: 121 Result: 53404
timeFileReader     Time: 65 Result: 53404
timeNioReader      Time: 40 Result: 53404

//still warming up
timeBufferedReader Time: 107 Result: 53404
timeFileReader     Time: 60 Result: 53404
timeNioReader      Time: 20 Result: 53404

//ripping along
timeBufferedReader Time: 79 Result: 53404
timeFileReader     Time: 56 Result: 53404
timeNioReader      Time: 16 Result: 53404

Make of it what you will.