Description:
------------
When "mb_ereg_replace" is used under "UTF-8" source environment, it does not seem to recognize ?gmulti-byte space?h (zenkaku-space in Japanese).   The same code runs correctly under ?gEUC-JP?h source environment, so I supposed this was something buggy rather than mis-usage of the function.


[mbstring]
mbstring.language = Japanese
mbstring.internal_encoding = UTF-8
mbstring.http_input = auto
mbstring.http_output = UTF-8
;mbstring.encoding_translation = Off
mbstring.detect_order = auto
mbstring.substitute_character = none;
;mbstring.func_overload = 0


Reproduce code:
---------------
<?php
    /*
    the all three var_dump()s "UTF-8", and this verifies 
    the source is actually run under "UTF-8"
    */
    var_dump(mb_regex_encoding());
    var_dump(mb_internal_encoding());

    mb_regex_encoding(mb_internal_encoding());
    var_dump(mb_regex_encoding());

    /*
    Although this var_dump should give 
    
    "**********?@*****" 
    
    by recognising "?@"(multi-byte space), 
    when it is run under "UTF-8", it gives 
    
    "****************".  
    
    This assures that the "multi-byte space" is not 
    recognised as it should.
    */
    $string = "multi-byte?@space";
    var_dump(mb_ereg_replace('[^?@]', '*', $string));
?>

Expected result:
----------------
multi-byte space should not be converted to "*" in the reproduce code since the regular expression is "[^?@]".

Actual result:
--------------
multi-byte space is ignored and converted to "*".