boost.log issue creating a file with a UTF8 file name

Question

I use boost.log to create multi log file according to string value. But when the string is UTF8 coded, the file created has incorrect name (like this: è°.æ­¦ç).

BOOST_LOG_SCOPED_LOGGER_ATTR(Logger::motion_checker, "RoleName", boost::log::attributes::constant< std::string >(name))

typedef boost::log::sinks::asynchronous_sink<boost::log::sinks::text_multifile_backend> multifile_sink;
boost::shared_ptr<multifile_sink> sink(new multifile_sink);
sink->locked_backend()->set_file_name_composer(boost::log::sinks::file::as_file_name_composer(
    boost::log::expressions::stream << "./log/MotionCheck/" << boost::log::expressions::attr< std::string >("RoleName") << ".log"));
sink->set_formatter
    (
        boost::log::expressions::format("[%1%] - %2%")
        % boost::log::expressions::attr< boost::posix_time::ptime >("TimeStamp")
        % boost::log::expressions::smessage
        );
sink->set_filter(channel == motion_check_channel);
core->add_sink(sink);

How to let boost.log handle UTF8 file name?

Answer 1

Boost.Log composes the file name in the encoding that is native for the underlying operating system. On Windows the file name is a UTF-16 string (the character type is wchar_t), on most POSIX systems it is typically UTF-8 (the character type is char).

In order to produce the file name in the native encoding the as_file_name_composer adapter creates a stream that performs character code conversion as needed when the adapted formatter is invoked. This basically lets you use both narrow and wide strings in the formatter as long as the encoding can be converted to the native one. You have to know though that the same-typed strings are assumed to have the same encoding, so if the native multibyte encoding is UTF-8 then all your narrow strings must also be UTF-8.

When character code conversion happens, the stream uses a locale that you can provide as the second argument for as_file_name_composer. By default the locale is default-constructed. If your default-constructed locale is not UTF-8 then the conversion will produce incorrect result, which I think is what's happening. You have to either set up your global locale to be UTF-8 or create a UTF-8 locale and pass it to the as_file_name_composer adapter. You can use Boost.Locale to generate a UTF-8 locale easily.