Description:
------------
Could you please explain why json_encode() takes care about the encoding at all? Why not to treat all the string data as a binary flow? This is very inconvenient and disallows the usage of json_encode() in non-UTF8 sites! :-(

I have written a small substitution for json_encode(), but note that it of course works much more slow than json_encode() with big data arrays..

    /**
     * Convert PHP scalar, array or hash to JS scalar/array/hash.
     */
    function php2js($a)
    {
        if (is_null($a)) return 'null';
        if ($a === false) return 'false';
        if ($a === true) return 'true';
        if (is_scalar($a)) {
            $a = addslashes($a);
            $a = str_replace("\n", '\n', $a);
            $a = str_replace("\r", '\r', $a);
            $a = preg_replace('{(</)(script)}i', "$1'+'$2", $a);
            return "'$a'";
        }
        $isList = true;
        for ($i=0, reset($a); $i<count($a); $i++, next($a))
            if (key($a) !== $i) { $isList = false; break; }
        $result = array();
        if ($isList) {
            foreach ($a as $v) $result[] = php2js($v);
            return '[ ' . join(', ', $result) . ' ]';
        } else {
            foreach ($a as $k=>$v) 
                $result[] = php2js($k) . ': ' . php2js($v);
            return '{ ' . join(', ', $result) . ' }';
        }
    }

So, my suggestion is remove all string analyzation from json_encode() code. It also make this function to work faster.

Reproduce code:
---------------
<?php
$a = array('a' => '&#1087;&#1088;&#1086;&#1074;&#1077;&#1088;&#1082;&#1072;', 'b' => array('&#1089;&#1083;&#1091;&#1093;&#1072;', '&#1075;&#1083;&#1091;&#1093;&#1086;&#1075;&#1086;'));
echo json_encode($a);
?>

Expected result:
----------------
Correctly encoded string in the source 1-byte encoding.

Actual result:
--------------
Empty strings everywhere (and sometimes - notices that a string contains non-UTF8 characters).