Description:
------------
A user contributed note[1] mentions an "unexpected behavior" of
preg_replace() with regard to matching only an empty subpattern.
In my opinion, this is a bug, because UTF-8 encoded code points
must never be split when the u modifier is set. Matching for an
empty subpattern only might be regarded as a bug per se, but then
at least a warning should be raised or an error should be reported
by preg_last_error().

According to <http://3v4l.org/AMq5N> the current behavior has been
introduced with PHP 5.2.9, and is not directly related to the PCRE
library.

[1] <http://php.net/manual/en/function.preg-replace.php#117314>


Test script:
---------------
<?php

var_dump(preg_replace('/()/u', '*', "a\xc3\x84b"));

?>


Expected result:
----------------
string(8) "*a*Ä*b*"


Actual result:
--------------
string(9) "*a*�*�*b*"