the GNU libstdc++ offers a flexible character encoding convertion primitive (essentially a wrapper for the iconv C API), but in a non-standard partial specialization of codecvt. for example, if I want to use this codecvt, to convert between internal encoding UCS-4LE and external encoding GB18030 (which is China's national standard for Chinese character encoding), I will write // define the codecvt type typedef std::codecvt<wchar_t, char, __gnu_cxx::encoding_state> w_codecvt; // define a std::locale instance only for codecvt purpose std::locale loc(std::locale::classic(), new w_codecvt); const w_codecvt &cvt = std::use_facet<w_codecvt>(loc); // the internal/external character encoding is specified as the // argument to the constructor of __gnu_cxx::encoding_state w_codecvt::state_type state("UCS-4LE", "GB18030"); // use cvt and state cvt.in(state, ...); the problem is, if I want to use this codecvt<.., .., __gnu_cxx::encoding_state> with code_converter, I can't find a way to tell code_convert the encoding names (UCS-4LE, GB18030), seems that code_converter will always use the default constructor of Codecvt::state_type. And if I specify the Codecvt template argument of code_converter to codecvt<.., .., encoding_state>, imbue() will become a no-op, which makes things even worse. BTW: the documentation for GNU libstdc++ codecvt is available here http://gcc.gnu.org/onlinedocs/libstdc++/22_locale/codecvt.html