From: Andy Heninger <andy.heninger@gmail.com>
Date: Fri, 27 Apr 2012 21:29:34 +0000 (+0000)
Subject: ICU-9283 fix for look-behind assertions w/ case insensitive matching.
X-Git-Tag: milestone-59-0-1~3842
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=b916fcb50b04ee230f6fe47711d7581af6a31e5c;p=icu

ICU-9283 fix for look-behind assertions w/ case insensitive matching.

X-SVN-Rev: 31782
---

diff --git a/icu4c/source/i18n/regexcmp.cpp b/icu4c/source/i18n/regexcmp.cpp
index 5a15ab6b9dd..146d73cfafe 100644
--- a/icu4c/source/i18n/regexcmp.cpp
+++ b/icu4c/source/i18n/regexcmp.cpp
@@ -3306,10 +3306,31 @@ int32_t   RegexCompile::maxMatchLength(int32_t start, int32_t end) {
             }
 
         case URX_STRING_I:
-            // TODO:  Is the case-folded string the longest?
-            //        If so we can optimize this the same as URX_STRING.
-            loc++;
-            currentLen = INT32_MAX;
+            // TODO:  This code assumes that any user string that matches will be no longer
+            //        than our compiled string, with case insensitive matching.
+            //        Our compiled string has been case-folded already.
+            //
+            //        Any matching user string will have no more code points than our
+            //        compiled (folded) string.  Folding may add code points, but
+            //        not remove them.
+            //
+            //        There is a potential problem if a supplemental code point 
+            //        case-folds to a BMP code point.  In this case our compiled string
+            //        could be shorter (in code units) than a matching user string.
+            //
+            //        At this time (Unicode 6.1) there are no such characters, and this case
+            //        is not being handled.  A test, intltest regex/Bug9283, will fail if
+            //        any problematic characters are added to Unicode.
+            //
+            //        If this happens, we can make a set of the BMP chars that the
+            //        troublesome supplementals fold to, scan our string, and bump the
+            //        currentLen one extra for each that is found.
+            //
+            {
+                loc++;
+                int32_t stringLenOp = (int32_t)fRXPat->fCompiledPat->elementAti(loc);
+                currentLen = safeIncrement(currentLen, URX_VAL(stringLenOp));
+            }
             break;
 
         case URX_CTR_INIT:
diff --git a/icu4c/source/test/intltest/regextst.cpp b/icu4c/source/test/intltest/regextst.cpp
index 09f205aad99..1e197a90e45 100644
--- a/icu4c/source/test/intltest/regextst.cpp
+++ b/icu4c/source/test/intltest/regextst.cpp
@@ -26,6 +26,7 @@
 #include "unicode/regex.h"
 #include "unicode/uchar.h"
 #include "unicode/ucnv.h"
+#include "unicode/uniset.h"
 #include "unicode/ustring.h"
 #include "regextst.h"
 #include "uvector.h"
@@ -127,6 +128,9 @@ void RegexTest::runIndexedTest( int32_t index, UBool exec, const char* &name, ch
         case 20: name = "CheckInvBufSize";
             if (exec) CheckInvBufSize();
             break;
+        case 21: name = "Bug 9283";
+            if (exec) Bug9283();
+            break;
 
         default: name = "";
             break; //needed to end loop
@@ -5184,6 +5188,34 @@ void RegexTest::Bug7029() {
     delete pMatcher;
 }
 
+// Bug 9283
+//   This test is checking for the existance of any supplemental characters that case-fold
+//   to a bmp character.  
+//
+//   At the time of this writing there are none. If any should appear in a subsequent release 
+//   of Unicode, the code in regular expressions compilation that determines the longest 
+//   posssible match for a literal string  will need to be enhanced.  
+//
+//   See file regexcmp.cpp, case URX_STRING_I in RegexCompile::maxMatchLength()
+//   for details on what to do in case of a failure of this test.
+//
+void RegexTest::Bug9283() {
+    UErrorCode status = U_ZERO_ERROR;
+    UnicodeSet supplementalsWithCaseFolding("[[:CWCF:]&[\\U00010000-\\U0010FFFF]]", status);
+    REGEX_CHECK_STATUS;
+    int32_t index;
+    UChar32 c;
+    for (index=0; ; index++) {
+        c = supplementalsWithCaseFolding.charAt(index);
+        if (c == -1) {
+            break;
+        }
+        UnicodeString cf = UnicodeString(c).foldCase();
+        REGEX_ASSERT(cf.length() >= 2);
+    }
+}
+
+
 void RegexTest::CheckInvBufSize() {
   if(inv_next>=INV_BUFSIZ) {
     errln("%s: increase #define of INV_BUFSIZ ( is %d but needs to be at least %d )\n",
diff --git a/icu4c/source/test/intltest/regextst.h b/icu4c/source/test/intltest/regextst.h
index 77742fb6b15..22a77eaa2e3 100644
--- a/icu4c/source/test/intltest/regextst.h
+++ b/icu4c/source/test/intltest/regextst.h
@@ -1,6 +1,6 @@
 /********************************************************************
  * COPYRIGHT:
- * Copyright (c) 2002-2011, International Business Machines Corporation and
+ * Copyright (c) 2002-2012, International Business Machines Corporation and
  * others. All Rights Reserved.
  ********************************************************************/
 
@@ -45,6 +45,7 @@ public:
     virtual void Bug7740();
     virtual void Bug8479();
     virtual void Bug7029();
+    virtual void Bug9283();
     virtual void CheckInvBufSize();
     
     // The following functions are internal to the regexp tests.
diff --git a/icu4c/source/test/testdata/regextst.txt b/icu4c/source/test/testdata/regextst.txt
index 21ecae9377d..53bd73a7ef3 100644
--- a/icu4c/source/test/testdata/regextst.txt
+++ b/icu4c/source/test/testdata/regextst.txt
@@ -1141,6 +1141,11 @@
 "[\w]+"                           "  <0>abc\u200cdef\u200dghi</0>   "
 "[\w]+"                         i "  <0>abc\u200cdef\u200dghi</0>   "
 
+# Bug 9283
+#  uregex_open fails for look-behind assertion + case-insensitive
+
+"(ab)?(?<=ab)cd|ef"             i  "<0><1>ab</1>cd</0>"
+
 #  Random debugging, Temporary
 #
 #"^(?:a?b?)*$"	                  "a--"