Hi,
I have rewritten large part of the source code of gnu.regexp package.
The imortant points of this change are:
(1) A new method REToken#matchThis. This method tries to match
the input string against the REToken itself and does not
try to match the next RETokens chained to it. The currently
used REToken#match method should be defined using REToken#matchThis.
This is useful for (3).
(2) A new method REToken#findMatch. This is almost the same as
the current REToken#match but returns a resulting REMatch
instead of a boolean value. This is useful for the depth-first
search with backtracking.
(3) New methods REToken#returnsFixedLengthMatches and
REToken#findFixedLengthMatches. These will fasten the
search for repeated matches if the matched string is
supposed to have a fixed length.
(4) RETokenOneOf and RETokenRepeated perform a depth-first
search with backtracking.
After this change, the test attached below shows 400% improved
performance compared with the current CVS version. The improved
performance comes mainly from the change (3). To my regret,
The change (4) had a negative effect on performance.
ChangeLog
2006-03-01 Ito Kazumitsu <[EMAIL PROTECTED]>
* gnu/regexp/BacktrackStack.java: New file.
* gnu/regexp/RE.java(findMatch): New method.
* gnu/regexp/REMatch.java(next,matchFlags,MF_FIND_ALL,
REMatchList): Removed. (backtrackStack): New field.
* gnu/regexp/REToken.java(match): Changed from an abstract
method to an ordinary method defined with the new method
matchThis. (matchThis, getNext, findMatch, returnsFixedLengthMatches,
findFixedLengthMatches, backtrack, toString): New methods.
* gnu/regexp/RETokenAny.java: Inplemented new methods of REToken.
* gnu/regexp/RETokenBackRef.java: Likewise.
* gnu/regexp/RETokenChar.java: Likewise.
* gnu/regexp/RETokenEnd.java: Likewise.
* gnu/regexp/RETokenEndSub.java: Likewise.
* gnu/regexp/RETokenIndependent.java: Likewise.
* gnu/regexp/RETokenLookAhead.java: Likewise.
* gnu/regexp/RETokenLookBehind.java: Likewise.
* gnu/regexp/RETokenNamedProperty.java: Likewise.
* gnu/regexp/RETokenPOSIX.java: Likewise.
* gnu/regexp/RETokenRange.java: Likewise.
* gnu/regexp/RETokenStart.java: Likewise.
* gnu/regexp/RETokenWordBoundary.java: Likewise
* gnu/regexp/RETokenOneOf.java: Rewriten.
* gnu/regexp/RETokenRepeated.java: Rewriten.
The performance test follows:
import java.io.*;
import java.util.*;
import java.util.regex.*;
import java.io.*;
import java.util.*;
import java.util.regex.*;
public class RegExTestCase {
static final Pattern dynPatString =
Pattern.compile("(.*?)(<dynstr\\s+)(property)\\s*=\\s*\"(.*?)\"\\s*(/>)"
);
// | 1 ||<- 2 ->||<- 3 ->| |<4>| |5 |
long time = 0;
public RegExTestCase () {
for (int i = 0; i < 10000; i++) {
String s = replaceDynamicStringAll(testString, "Foo");
}
System.out.println("Elapsed time = " + time);
}
public String replaceDynamicStringAll(String inStr, String
replaceStr) {
StringBuffer sb = new StringBuffer();
Matcher m = dynPatString.matcher(inStr);
time -= System.currentTimeMillis();
boolean b = m.find();
time += System.currentTimeMillis();
while (b) {
sb.append(m.group(1));
m.appendReplacement(sb, replaceStr);
b = m.find();
}
m.appendTail(sb);
return sb.toString();
}
private final static String testString = "ABC<dynstr property =
\"X\"/>";
public final static void main(String[] args) {
new RegExTestCase();
}
}
--- gnu/java/nio/charset/iconv/IconvProvider.java.orig Sat Jul 16 00:12:48 2005
+++ gnu/java/nio/charset/iconv/IconvProvider.java Thu Oct 20 23:41:57 2005
@@ -62,7 +62,11 @@
}
}
- private IconvProvider()
+ // Declaring the construtor public may violate the use of singleton.
+ // But it must be public so that an instance of this class can be
+ // created by Class.newInstance(), which is the case when this provider is
+ // defined in META-INF/services/java.nio.charset.spi.CharsetProvider.
+ public IconvProvider()
{
IconvMetaData.setup();
}
