This is the mail archive of the
libstdc++@gcc.gnu.org
mailing list for the libstdc++ project.
Re: Incorrect char extraction code
- To: "Rosen, Hyman" <Hyman dot Rosen at kbcfp dot com>
- Subject: Re: Incorrect char extraction code
- From: Jason Merrill <jason at redhat dot com>
- Date: 30 Jan 2001 09:22:48 +0000
- Cc: "'libstdc++ at gcc dot gnu dot org'" <libstdc++ at gcc dot gnu dot org>
- References: <415C10091CFFD21193A50060083606C601AE3526@msny2.nyc.kbcfp.com>
>>>>> "Rosen," == Rosen, Hyman <Hyman.Rosen@kbcfp.com> writes:
> std_istream.h has the following pair of templates:
> template<class _Traits>
> basic_istream<char, _Traits>&
operator> (basic_istream<char, _Traits>& __in, unsigned char& __c)
> { return (__in >> static_cast<char>(__c)); }
> template<class _Traits>
> basic_istream<char, _Traits>&
operator> (basic_istream<char, _Traits>& __in, signed char& __c)
> { return (__in >> static_cast<char>(__c)); }
> These are clearly wrong; the result of a static_cast<char> is not an
> lvalue, so code that tries an extraction into a signed or unsigned char
> shouldn't even compile. Those casts need to be reinterpret_cast<char &>,
Yes.
> and then you need to verify that you won't run afoul of gcc's alias
> detection.
If it did, that would be a bug in the compiler; anything can be aliased by
a char*.
Jason