Japanese character problem while using “JISAutoDetect” encoding with Java1.4.2

Some Japanese characters have a problem while using the “JISAutoDetect” encoding with Java1.4.2_26. The issue does not fail in JDK 5.0_22 or 6.0_19.
 

Steps to run the test case

1. Set LANG to 'ja_JP.PCK'

2. Javac and run.
java moji2 | od -x

 Here is the test result.

JDK 1.4.2_26

  /opt/java14224/bin/java moji2 | od -x

  0000000 815b 3f8a 8740 3f54 3f41 3f55 0a00

               ==

 So the first byte of 0x878a is converted to 0x3f.

JDK 6u19

  /opt/java16006/bin/java moji2 | od -x

  0000000 815b 878a 8740 8754 8741 8755 0a00

               ====

 The value 0x878a, is the correct result.

Testcase is attached, and appended below:
import java.io.*;
import java.net.*;

public class moji2 {

    public static void main(String[] args) {

        try{
        String str = "¿[¿¿¿@¿T¿A¿U";

        byte[] bytes = str.getBytes("MS932");
        String str2 = new String(bytes, "JISAutoDetect");
// String str2 = new String(bytes, "MS932");
        System.out.println(str2);
        }catch(UnsupportedEncodingException e){
            e.printStackTrace();
        }

  }
}