https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80137
--- Comment #3 from John Salmon <john.salmon at deshaw dot com> ---
It's easy to overthink this. 0.0 is perfectly acceptable, as is any other
_RealType in the range [0, 1.). But since rounding was, presumably, to-nearest
or up, it's slightly disconcerting that 0.0 is neither near nor up from the
"exact" value.
How about:
if (__builtin_expect(__ret >= _RealType(1), 0))
{
#if _GLIBCXX_USE_C99_MATH_TR1
__ret = std::nextafter(_RealType(1), _RealType(0));
#else
__ret = _RealType(1) - std::numeric_limits<_RealType>::epsilon()/2.;
#endif
}
I.e., if there's no nextafter, then use numeric_limits::epsilon() to find the
value just below 1.0.
John Salmon
