When running my converter test program I got a large number of errors in
JDK 1.4 due to a bug in the String constructor. When malformed byte
sequence is given, the code in java.lang.StringCoding.java thows an Error
with the text "Converter malfunction". The new class Error is designed for
cases of internal JVM errors and should not be thrown for cases of bad
input which the app may well want to check for. In all previous versions
of Java, and in the String.getBytes() method malformed sequences are
ignored and there is no exception (or Error) thrown.

The code at line 123 of StringCoding.java needs to set the count based on
nextCharIndex() as is done in String.java in JDK 1.3 (about line 862). A
much better solution would be to do what is done in the String.getBytes()
and restart the convert after ignoring the malformed input. This matches
what is done when using the nio converters.


public class StringErrorTest {
     public static void main (String[] args) {
         byte [] b = {(byte)0x82};
         String s;
         try {
             s = new String(b, "Cp950");
         } catch (Exception e) {
             e.printStackTrace(System.err);
             s = "<exception>";
         }
         System.out.println("Convert = " + s);
     }
}

>java StringErrorTest
Exception in thread "main" java.lang.Error: Converter malfunction:
sun.io.ByteToCharCp950
        at java.lang.StringCoding$ConverterSD.decode(StringCoding.java:123)
        at java.lang.StringCoding.decode(StringCoding.java:189)
        at java.lang.String.<init>(String.java:325)
        at java.lang.String.<init>(String.java:346)
        at StringErrorTest.main(StringErrorTest.java:6)
Caused by: sun.io.MalformedInputException
        at sun.io.ByteToCharDBCS_ASCII.flush(ByteToCharDBCS_ASCII.java:38)
        at java.lang.StringCoding$ConverterSD.decode(StringCoding.java:120)
        ... 4 more