[PATCH] [PATCH] PR9804 - __is_signed conflicts with libstdc++

[Richard Smith]

On Mon, May 11, 2015 at 12:27 PM, Charlie Turner <charlie.turner at arm.com>
wrote:

> In http://reviews.llvm.org/D9157#159383, @rsmith wrote:
>
> > What is the motivation for this? Every version of libstdc++ I can find
> uses the keyword `bool` and not a typedef.
>
>
> Sorry for the late reply Richard... I've also lost the previous thread I
> had with. I originally suggested the motivation was for cases like,
>
>   typedef bool BOOL;
>
>   class Foo {
>       static const BOOL __is_signed;
>     };
>
>   #include <map>
>
>   int main() {}
>
> But as you pointed out, this is ill-formed. I found a smaller example,
> which I'm confident is well-formed yet rejected by Clang,
>
>   typedef bool Bool;
>   #define bool Bool
>
>   #include <algorithm>
>
>   int main(void) {}
>
> I see the following errors
>
>   In file included from blentest.c:4:
>   In file included from
> /usr/lib/gcc/x86_64-linux-gnu/4.9/../../../../include/c++/4.9/algorithm:61:
>   In file included from
> /usr/lib/gcc/x86_64-linux-gnu/4.9/../../../../include/c++/4.9/bits/stl_algobase.h:63:
>
> /usr/lib/gcc/x86_64-linux-gnu/4.9/../../../../include/c++/4.9/ext/numeric_traits.h:63:25:
> error: expected member name or ';' after declaration specifiers
>         static const bool __is_signed = __glibcxx_signed(_Value);
>         ~~~~~~~~~~~~~~~~~ ^
>
> /usr/lib/gcc/x86_64-linux-gnu/4.9/../../../../include/c++/4.9/ext/numeric_traits.h:74:50:
> error: expected unqualified-id
>       const bool __numeric_traits_integer<_Value>::__is_signed;
>                                                    ^
>
> /usr/lib/gcc/x86_64-linux-gnu/4.9/../../../../include/c++/4.9/ext/numeric_traits.h:106:25:
> error: expected member name or ';' after declaration specifiers
>         static const bool __is_signed = true;
>         ~~~~~~~~~~~~~~~~~ ^
>
> /usr/lib/gcc/x86_64-linux-gnu/4.9/../../../../include/c++/4.9/ext/numeric_traits.h:115:51:
> error: expected unqualified-id
>       const bool __numeric_traits_floating<_Value>::__is_signed;
>                                                     ^
>   In file included from blentest.c:4:
>   In file included from
> /usr/lib/gcc/x86_64-linux-gnu/4.9/../../../../include/c++/4.9/algorithm:61:
>
> /usr/lib/gcc/x86_64-linux-gnu/4.9/../../../../include/c++/4.9/bits/stl_algobase.h:941:49:
> error: expected unqualified-id
>            && !__gnu_cxx::__numeric_traits<_ValueType1>::__is_signed
>                                                          ^
>
> /usr/lib/gcc/x86_64-linux-gnu/4.9/../../../../include/c++/4.9/bits/stl_algobase.h:946:45:
> error: non-type template argument of type 'Bool' (aka 'bool') is not an
> integral constant expression
>         return std::__lexicographical_compare<__simple>::__lc(__first1,
> __last1,
>                                               ^~~~~~~~
>   6 errors generated.
>
> The only restrictions I can see stated in the C++ standard (N4296,
> 2014-11-19) are in section 17.6.4.3.1, and they only cover that you're not
> allowed to `#define` or `#undef` names declared in a standard library
> header, or `override`, `final` and some other bits about attributes.


It also says "names lexically identical to keywords", which covers 'bool'.


> `bool` is a fundamental type, so I think my example above should be
> accepted.
>
> I think reopening PR9804 for this was incorrect, it looks quite similar,
> but perhaps I should open a new PR for this issue. Do you agree?
>
> Thank you for your time. :)
>
>
> REPOSITORY
>   rL LLVM
>
> http://reviews.llvm.org/D9157
>
> EMAIL PREFERENCES
>   http://reviews.llvm.org/settings/panel/emailpreferences/
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20150511/30004008/attachment.html>