'possible loss of data' warning when using wide-character wtime_input_facet

[trueqbit]

Compilers emit a 'possible loss of data' warning when using wtime_input_facet::get(). E.g. Visual C++ emits "_warning C4244: 'argument': conversion from 'const wchar_t' to 'const Elem', possible loss of data" with the following code:

wstring s;
// in.imbue({ locale::classic(), new wtime_input_facet{L"%m-%e-%Y %H:%M:%S"} });
wistringstream in{ s };
ptime pt;
in >> pt;

This is because wtime_input_facet::get() aka time_input_facet<ptime,wchar_t,istreambuf_iterator<wchar_t>>::get() checks for errors in regard to special formatting characters, and the error handling code path effectively converts the wide-character input sequence to a narrow-character std::string, date_time::convert_string_type<wchar_t, char> being the culprit in date_time library land.

The call and warning message stack look like this:

>   26: date_time::convert_string_type<wchar_t,char>(const wstring& inp_str)
|-> 1260: date_time::wtime_input_face::check_special_value<ptime>(istreambuf_iterator<wchar_t>& sitr, istreambuf_iterator<wchar_t>& stream_end, ptime& tt, wchar_t c)
  |-> 1125: date_time::wtime_input_face::get(istreambuf_iterator<wchar_t>& sitr, istreambuf_iterator<wchar_t>& stream_end, ios_base& ios_arg, ptime& t, wstring& tz_str, bool time_is_local)
    |->  936: date_time::wtime_input_face::get(istreambuf_iterator<wchar_t>& sitr, istreambuf_iterator<wchar_t>& stream_end, ios_base& ios_arg, ptime& t)
      |->   80: posix_time::operator>><wchar_t>(wistream& is, ptime& pt)

I am currently uncertain how to fix this properly, but it is definitely a PITA because of the lengthy warning message stack.

[hajokirchhoff]

At the minimum, the convert_string_type should use narrow or widen to properly convert char to wchar_t and vice versa, instead of passing a wchar_t* iterator to a string. This issue is even more bothersome, because the only place I find where it is used, is to throw an exception std::ios_base::failure, when the parsing of special chars failed. All it does there is to include the problematic string.

IMHO, convert_to_string could/should be completely removed, since the conversion will fail for most non-english languages anyway and throwing std::ios_base::failure with the text "parse failed" should be enough info.

And the warning is more than just "annoying" to us as we have a strict "warnings as errors" policy.