Remove reduntant regex from calls to StringUtils::isUtf8()

I've cautiously moved the regex out of the most used code path.
There is no string that will match that regex check that will not also be
passed by mb_check_encoding.  I think the regex was intended as a shortcut
evaluation, but it is no faster than mb_check_encoding which will often
need to be run anyway.

I think it could just be deleted, but I have limited motivation to
risk introducing a bug to improve performance on old PHP vesions and
unusual configurations, so I've moved it to the fallback code path.

Change-Id: Ie9425cc23ba032e5aff42beeb44cbb1146050452

diff --git a/includes/StringUtils.php b/includes/StringUtils.php
index 9e21d03..c1545e6 100644 (file)
--- a/includes/StringUtils.php
+++ b/includes/StringUtils.php
@@ -51,10 +51,6 @@ class StringUtils {
         */
        static function isUtf8( $value, $disableMbstring = false ) {
                $value = (string)$value;
-               if ( preg_match( "/[\x80-\xff]/S", $value ) === 0 ) {
-                       // String contains only ASCII characters, has to be valid
-                       return true;
-               }
 
                // If the mbstring extension is loaded, use it. However, before PHP 5.4, values above
                // U+10FFFF are incorrectly allowed, so we have to check for them separately.
@@ -68,6 +64,11 @@ class StringUtils {
                                ( $newPHP || preg_match( "/\xf4[\x90-\xbf]|[\xf5-\xff]/S", $value ) === 0 );
                }
 
+               if ( preg_match( "/[\x80-\xff]/S", $value ) === 0 ) {
+                       // String contains only ASCII characters, has to be valid
+                       return true;
+               }
+
                // PCRE implements repetition using recursion; to avoid a stack overflow (and segfault)
                // for large input, we check for invalid sequences (<= 5 bytes) rather than valid
                // sequences, which can be as long as the input string is. Multiple short regexes are