Sharing instance of NumberFormat can cause diffrent results

[capussopy]

When parallel requests are sent via filter, the sharing NumberFormat instance parse randomly different numbers. Sometimes it fails , sometime its correct and sometimes it parses a double out.

The issue is caused by your StringConverter.class

An simple example that causes the issue:

public class Scratch extends Thread {

    public static final NumberFormat NUMBER_FORMAT ;

    public static void main(String args[]) {
        for (int i = 0; i < 3; i++) {
            Scratch thread = new Scratch();
            thread.start();
        }
    }

    public void run() {
        ParsePosition position = new ParsePosition(0);
        Number number = NUMBER_FORMAT.parse("1011", position);
        System.out.println(number);
    }

    static {
        NUMBER_FORMAT = NumberFormat.getInstance(Locale.US);
        NUMBER_FORMAT.setGroupingUsed(false);
    }
}

The issue disapers, if we use a new instance of Numberformat in every Thread:

public class Scratch extends Thread {


    public static void main(String args[]) {
        for (int i = 0; i < 5; i++) {
            Scratch thread = new Scratch();
            thread.start();
        }
    }

    public void run() {
        ParsePosition position = new ParsePosition(0);
        NumberFormat format = NumberFormat.getInstance(Locale.US);
        format.setGroupingUsed(false);
        Number number = format.parse("1011", position);
        System.out.println(number);
    }
  
}

[torshid]

@capussopy Thanks for noticing about this. I think that we have a similar issue with SimpleDateFormat, which is not thread safe too. It should be possible to wrap these fields with ThreadLocal, do you have other solutions?

[capussopy]

@torshid This is true. The Javadoc for SimpleDateFormat says:

Date formats are not synchronized. It is recommended to create separate format instances for each thread. If multiple threads access a format concurrently, it must be synchronized externally.

yes, you can use Threadlocal or create a new instance in each thread. I think the solution with ThreadLocal should have the better performance. But im not 100% sure.

[capussopy]

@torshid any updates on this?

[FelipeBerrios]

any updates?

[torshid]

Fixed in https://github.com/turkraft/spring-filter/releases/tag/2.1.2.

[FlorianCassayre]

@skamv