AFAIK, the problems discussed here are non-issues for Windows and Mac OS X, so you can ignore this post if you only care about those two operating systems. http://svn.boost.org/trac/boost/changeset/72855 fixes the problem of misconfigured POSIX system throwing an exception before main() has started. See: http://svn.boost.org/trac/boost/ticket/4688 http://svn.boost.org/trac/boost/ticket/5100 http://svn.boost.org/trac/boost/ticket/5289 There are at least two remaining problems: * There may be other BSD based operating systems that like Mac OS X should not even try to use std::locale("") to get the codecvt facet for conversion between wide and narrow strings. If you know of such a system (FreeBSD? Solaris? Others?), please let me know (1) the system, (2) the predefined macro to use to identify the system, and (3) what codecvt facet should be used instead. * For systems such as Linux, the code is currently still assuming that if std::locale("") throws, the best action is to let that exception propagate up the call stack. One of the tickets [5289] supplies a patch that tries various alternatives, and eventually falls back on the Boost UTF-8 codecvt. See below. It strikes me that such fallbacks are better left to user code. User code can always try {path(L"foo");} and then imbue their preferred fallback facet. What do others think? --Beman + // Get default OS encoding + char const *plang = getenv("LC_CTYPE"); + if(!plang) + plang = getenv("LC_ALL"); + if(!plang) + plang = getenv("LANG"); + if(!plang) + return global_loc; + + std::string lang = plang; + size_t charset_start = lang.find('.'); + if(charset_start == std::string::npos) + return global_loc; + charset_start ++; + size_t end_of_charset = lang.find(charset_start,'@'); + std::string encoding = lang.substr(charset_start,end_of_charset - charset_start); + for(size_t i=0;i<encoding.size();i++) { + if('a' <= encoding[i] && encoding[i]<='z') { + encoding[i]=encoding[i]-'a' + 'A'; + } + } + // + // Support at least the most popular and widely used encoding + // + if(encoding!="UTF-8" && encoding!="UTF8") + return global_loc; + std::locale loc(global_loc, new boost::filesystem::detail::utf8_codecvt_facet); + return loc;