On 20/05/17 15:10 +0800, Xi Ruoyao wrote:
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.
https://gcc.gnu.org/onlinedocs/gcc/Integers-implementation.html
"GCC only supports two's complement integer types"
I doubt that's ever going to change, but keeping the negation also
doesn't do any harm. I'll test and commit this, thanks.
