What's your rationale for this fix? What sort of regression happened?
OK, so let me explain:
preg_replace($regex, $replace, $str, $limit);
Short answer: It was failling if $limit < $number_of_matches.
How this bug was introduced? This started with your commit 3 weeks ago that
introduced the preg_last_error(). Later when Tony fixed a mem leak in the
same place, the problem became even worst (although the commit was correct).
Now looking at the bug report (http://bugs.php.net/37800), example #1. After
your commit, it would return ''One ". After Tony's commit it returned NULL.
The correct answer is "One 111111111"
The problematic code change:
@@ -1011,7 +1088,7 @@ PHPAPI char *php_pcre_replace(char *rege
if (limit != -1)
limit--;
- } else { /* Failed to match */
+ } else if (count == PCRE_ERROR_NOMATCH) {
/* If we previously set PCRE_NOTEMPTY after a null match,
this is not necessarily the end. We need to advance
the start offset, and continue. Fudge the offset values
@@ -1036,6 +1113,9 @@ PHPAPI char *php_pcre_replace(char *rege
result[*result_len] = '\0';
break;
}
+ } else {
+ pcre_handle_exec_error(count TSRMLS_CC);
+ break;
}
Previously it entered in the first else when limit == 0 (and when there was
an error too, but thats another story). My fix just restores the old
behaviour, that is enter in that piece of code.
I hope it is more clear now.
Nuno
On Jun 15, 2006, at 11:33 AM, Nuno Lopes wrote:
nlopess Thu Jun 15 15:33:25 2006 UTC
Added files: (Branch: PHP_5_2)
/php-src/ext/pcre/tests bug37800.phpt
Modified files:
/php-src/ext/pcre php_pcre.c
Log:
fix bug #37800: preg_replace() limit parameter odd behaviour
#this is a regression in PHP_5_2 and HEAD branches only
http://cvs.php.net/viewcvs.cgi/php-src/ext/pcre/php_pcre.c?
r1=1.168.2.9.2.5&r2=1.168.2.9.2.6&diff_format=u
Index: php-src/ext/pcre/php_pcre.c
diff -u php-src/ext/pcre/php_pcre.c:1.168.2.9.2.5 php-src/ext/pcre/
php_pcre.c:1.168.2.9.2.6
--- php-src/ext/pcre/php_pcre.c:1.168.2.9.2.5 Wed Jun 14 17:52:56 2006
+++ php-src/ext/pcre/php_pcre.c Thu Jun 15 15:33:25 2006
@@ -16,7 +16,7 @@
+---------------------------------------------------------------------
-+
*/
-/* $Id: php_pcre.c,v 1.168.2.9.2.5 2006/06/14 17:52:56 nlopess Exp $ */
+/* $Id: php_pcre.c,v 1.168.2.9.2.6 2006/06/15 15:33:25 nlopess Exp $ */
#ifdef HAVE_CONFIG_H
#include "config.h"
@@ -1091,7 +1091,7 @@
if (limit != -1)
limit--;
- } else if (count == PCRE_ERROR_NOMATCH) {
+ } else if (count == PCRE_ERROR_NOMATCH || limit == 0) {
/* If we previously set PCRE_NOTEMPTY after a null match,
this is not necessarily the end. We need to advance
the start offset, and continue. Fudge the offset values
http://cvs.php.net/viewcvs.cgi/php-src/ext/pcre/tests/bug37800.phpt?
view=markup&rev=1.1
Index: php-src/ext/pcre/tests/bug37800.phpt
+++ php-src/ext/pcre/tests/bug37800.phpt
--
PHP CVS Mailing List (http://www.php.net/)
To unsubscribe, visit: http://www.php.net/unsub.php
