use of study() with utf8 enabled breaks regexps