constexpr fails with clang but not with g++

[Florian Lindner]

Hello,

I have this piece of code: (not written by me)


#include "mpi.h"

// template <size_t>
// struct MPI_Select_unsigned_integer_datatype;

// template <>
// struct MPI_Select_unsigned_integer_datatype<1> {
// static constexpr MPI_Datatype const datatype = MPI_UNSIGNED_CHAR;
// };


int main(int argc, char *argv[])
{
constexpr MPI_Datatype const datatype = MPI_UNSIGNED_CHAR;
return 0;
}


The actual code is the commented out, but the constexpr line produces the
same error.

It compiles fine with g++ 5.1 but fails with clang++ 3.6.1:


% OMPI_CXX="clang++" mpic++ -std=c++11 test.cpp
test.cpp:14:32: error: constexpr variable 'datatype' must be initialized by
a constant expression
constexpr MPI_Datatype const datatype = MPI_UNSIGNED_CHAR;
^ ~~~~~~~~~~~~~~~~~
test.cpp:14:43: note: cast from 'void *' is not allowed in a constant
expression
constexpr MPI_Datatype const datatype = MPI_UNSIGNED_CHAR;
^
/usr/include/mpi.h:1049:27: note: expanded from macro 'MPI_UNSIGNED_CHAR'
#define MPI_UNSIGNED_CHAR OMPI_PREDEFINED_GLOBAL(MPI_Datatype,
ompi_mpi_unsigned_char)
^
/usr/include/mpi.h:308:47: note: expanded from macro
'OMPI_PREDEFINED_GLOBAL'
#define OMPI_PREDEFINED_GLOBAL(type, global) ((type) ((void *) &(global)))
^
1 error generated.


First question: Is the const statement superfluous?

I don't understand why clang fires the error here. MPI_UNSIGNED_CHAR is just
a #define in mpi.h

#define MPI_UNSIGNED_CHAR OMPI_PREDEFINED_GLOBAL(MPI_Datatype,
ompi_mpi_unsigned_char)

which is

#if !OMPI_BUILDING
#define OMPI_PREDEFINED_GLOBAL(type, global) ((type) ((void *) &(global)))
#else
#define OMPI_PREDEFINED_GLOBAL(type, global) ((type) &(global))
#endif


And a bit beyond my understanding of C and C++.

What is the best way to fix compilation with clang?

Is that the same problem?
http://stackoverflow.com/questions/24398102/constexpr-and-initialization-of-a-static-const-void-pointer-with-reinterpret-cas

Thanks!

Florian

[Jens Maurer]

Yes. "constexpr" implies "const" for variables:

7.1.5p9
"A constexpr specifier used in an object declaration declares the object as const."

> I don't understand why clang fires the error here. MPI_UNSIGNED_CHAR is just
> a #define in mpi.h
>
> #define MPI_UNSIGNED_CHAR OMPI_PREDEFINED_GLOBAL(MPI_Datatype,
> ompi_mpi_unsigned_char)
>
> which is
>
> #if !OMPI_BUILDING
> #define OMPI_PREDEFINED_GLOBAL(type, global) ((type) ((void *) &(global)))
> #else
> #define OMPI_PREDEFINED_GLOBAL(type, global) ((type) &(global))
> #endif

5.20p2.11:

"-- a conversion from type cv void * to a pointer-to-object type;"

(This is intended to prevent type punning in constant expressions.)

> And a bit beyond my understanding of C and C++.
>
> What is the best way to fix compilation with clang?

Don't use constexpr in that case. It's very unlikely you need it here.

> Is that the same problem?
> http://stackoverflow.com/questions/24398102/constexpr-and-initialization-of-a-static-const-void-pointer-with-reinterpret-cas

No, that's 5.20p2.13.

Jens

[Faisal Vali]

I don't think MPI_Datatype is a pointer - i believe this falls afoul
of 5.20p2.13 too.

>
>> And a bit beyond my understanding of C and C++.
>>
>> What is the best way to fix compilation with clang?
>
> Don't use constexpr in that case. It's very unlikely you need it here.
>
>> Is that the same problem?
>> http://stackoverflow.com/questions/24398102/constexpr-and-initialization-of-a-static-const-void-pointer-with-reinterpret-cas
>
> No, that's 5.20p2.13.
>
> Jens
>

> --
>
> ---
> You received this message because you are subscribed to the Google Groups "ISO C++ Standard - Discussion" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to std-discussio...@isocpp.org.
> To post to this group, send email to std-dis...@isocpp.org.
> Visit this group at http://groups.google.com/a/isocpp.org/group/std-discussion/.