#242 Performance problem in SegmentFinder

[Christian Ohr]

From the Mailing list:

I was profiling some of our code which uses HL7, and I noticed that we had a large number of calls to java.util.Pattern#compile.

I tracked a number of them down to SegmentFinder#matches method. I would therefore suggest the patch below, which cuts down the number of Pattern#compile calls from 4 to 1.

The unit tests passes. I have written the code myself and would like to submit it under the same GPL/MPL license as the rest of the code.

Regards,

Niels Harremoës

Index: SegmentFinder.java
===================================================================
--- SegmentFinder.java   (revision 983)
+++ SegmentFinder.java (working copy)
@@ -157,8 +157,13 @@
         }
         return matches;
     }*/
-   
+
     /**
+     * Patterns used for matching can only contain these characters
+     */
+    private static final Pattern VALID_PATTERN_PATTERN = Pattern.compile("[\\w\\*\\?]*");
+
+    /**
      * Tests whether the given name matches the given pattern.
      */
     private boolean matches(String pattern, String candidate) {
@@ -166,12 +171,12 @@
         if (pattern.equals(candidate)) {
             return true;
         }
-       
-        if (!Pattern.matches("[\\w\\*\\?]*", pattern))
+
+        if (!VALID_PATTERN_PATTERN.matcher(pattern).matches())
             throw new IllegalArgumentException("The pattern " + pattern + " is not valid.  Only [\\w\\*\\?]* allowed.");

-        pattern = Pattern.compile("\\*").matcher(pattern).replaceAll(".*");
-        pattern = Pattern.compile("\\?").matcher(pattern).replaceAll(".");
+        pattern = pattern.replace("*",".*");
+        pattern = pattern.replace('?','.');        

         return Pattern.matches(pattern, candidate);
     }