PatternSyntaxException when calling String.replaceAll in multithreaded app

Question

Dear,
Someone ever encountered or knows of problems when invoking String.replaceAll in a multithreaded application (JDK 1.4.2)?
I have an application that invokes replaceAll on a String and that runs just fine when only a single thread is active, but once multiple threads execute the same code, each on its own String instance, the following stack is produced (more often than not).
java.util.regex.PatternSyntaxException: Unknown character category {Digit} near index 9
^\p{Digit}
^
at java.util.regex.Pattern.error(Unknown Source)
at java.util.regex.Pattern.familyError(Unknown Source)
at java.util.regex.Pattern.retrieveCategoryNode(Unknown Source)
at java.util.regex.Pattern.family(Unknown Source)
at java.util.regex.Pattern.sequence(Unknown Source)
at java.util.regex.Pattern.expr(Unknown Source)
at java.util.regex.Pattern.compile(Unknown Source)
at java.util.regex.Pattern.<init>(Unknown Source)
at java.util.regex.Pattern.compile(Unknown Source)
at java.lang.String.replaceAll(Unknown Source)
The 1.4.2 source code seems to use a static HashMap (java.util.regex.Pattern#retrieveCategoryNode) to store some read only data, but it looks like the initialization of that HashMap in an multi threaded environment is not 'locked'. Could that be the/an issue? Anybody any thoughts?
Thanks,
Peter

Answer 1

I ran peter's program on a hyperthreaded Intel P4 box and XP Pro. XP reports that there are 2 processors.
I used 1.4.2_06 and 1.5.0_01, with both -client and -server options.
The program was run with numThreads 10 and loadPattern 0 and 1.
The program was run 50 times for each of the configurations. Only one resulted in failures, as follows:
Java version     cli/svr          Pgm Parms   # Failures      
1.5.0_01     -client          10, 0          0
                    10, 1          0
          -server          10, 0          0
                    10, 1          0
1.4.2_06     -client          10, 0          10
                    10, 1          0
          -server          10, 0          0
                    10, 1          0
Of the 10 failures, 1 reported 3 errors, 2 reported 2 errors, and 7 reported 1 error.
The triple-error report is below:
"C:\Program Files\Java\jdk1.4.2_06\bin\java.exe" -client PatternProblem 10 0
Thread-7:Unknown character category {Digit} near index 9
^\p{Digit}
         ^
Thread-5:Unknown character category {Digit} near index 9
^\p{Digit}
         ^
Thread-3:Unknown character category {Digit} near index 9
^\p{Digit}
         ^
Note that this wording is not the same as peter encountered. The thread number varied,
apparently randomly, from 0 to 8