https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114244
--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> --- Yup, the seconds part "00.002" is parsed using std::numpunct (in order to handle the locale's decimal point) and then converted to milliseconds using duration_cast: auto& __ng = use_facet<num_get<_CharT>>(__loc); long double __val; ios_base::iostate __err2{}; __ng.get(__buf, {}, __buf, __err2, __val); if (__is_failed(__err2)) [[unlikely]] __err |= __err2; else { duration<long double> __fs(__val); __s = duration_cast<_Duration>(__fs); } We also use duration_cast when parsing %c and %X using std::time_get, but that's an integer conversion there (but the duration_cast should have been qualified to prevent ADL): __h = hours(__tm.tm_hour); __min = minutes(__tm.tm_min); __s = duration_cast<_Duration>(seconds(__tm.tm_sec)); and another duration_cast in chrono::from_stream for durations. That one could be used with either integral or floating-point reps. Do we want to parse "00:00:31" as minutes(1)? I don't think we do, so only the first case where converting long double to milliseconds should be rounded?