On 2017-05-19 15:38 +0100, Jonathan Wakely wrote: > On 18/05/17 19:10 +0800, Xi Ruoyao wrote: > > This UB has been hiding so long... > > Indeed! Thanks for the patch. > > > 2017-03-11 Xi Ruoyao <r...@stu.xidian.edu.cn> > > > > PR libstdc++/67214 > > * include/bits/locale_facets.tcc (_M_extract_int): > > Add explicit conversion to avoid signed overflow. > > --- > > libstdc++-v3/include/bits/locale_facets.tcc | 3 ++- > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > diff --git a/libstdc++-v3/include/bits/locale_facets.tcc > > b/libstdc++-v3/include/bits/locale_facets.tcc > > index 351190c..5f85d15 100644 > > --- a/libstdc++-v3/include/bits/locale_facets.tcc > > +++ b/libstdc++-v3/include/bits/locale_facets.tcc > > @@ -470,7 +470,8 @@ _GLIBCXX_BEGIN_NAMESPACE_LDBL > > bool __testoverflow = false; > > const __unsigned_type __max = > > (__negative && __gnu_cxx::__numeric_traits<_ValueT>::__is_signed) > > - ? -__gnu_cxx::__numeric_traits<_ValueT>::__min > > + ? -static_cast<__unsigned_type>(__gnu_cxx:: > > + __numeric_traits<_ValueT>::__min) > > Do we need to keep the negation, or can we just cast to > __unsigned_type?
For 2's complement we can just cast to __unsigned_type. But for clarity and other strange architectures I think we should keep the negation. -- Xi Ruoyao <r...@stu.xidian.edu.cn> School of Aerospace Science and Technology, Xidian University
