[[nodiscard]] not understood by the intel compiler

[angainor]

I get a lot of warnings like this when compiling with ICC:

/cluster/projects/nn9999k/marcink/ghex/angainor/GHEX/include/ghex/communication_object_2.hpp(309): warning #1292: unknown attribute "nodiscard"
              [[nodiscard]] std::enable_if_t<std::is_same<Arch,cpu>::value, handle_type>
                ^

[angainor]

There are also these warnings:

In file included from /cluster/projects/nn9999k/marcink/ghex/angainor/GHEX/include/ghex/structured/cubed_sphere/../../arch_traits.hpp(14),
                 from /cluster/projects/nn9999k/marcink/ghex/angainor/GHEX/include/ghex/structured/cubed_sphere/field_descriptor.hpp(19),
                 from /cluster/projects/nn9999k/marcink/ghex/angainor/GHEX/bindings/cubed_sphere_bind.cpp(5):
/cluster/projects/nn9999k/marcink/ghex/angainor/GHEX/include/ghex/structured/cubed_sphere/../.././allocator/pool_allocator_adaptor.hpp(133): warning #488: template parameter "U" is not used in declaring the parameter types of function template "gridtools::ghex::allocator::pool_allocator_adaptor<Allocator>::pool_allocator_adaptor<U>(gridtools::ghex::allocator::pool_allocator_adaptor<Allocator>::rebind<U>::other &&)"
                  template<typename U>
                                    ^

[mbianco]

We need to discuss the support of ICC. In GridTools we had to stop supporting it, since ICC is having lot of issues with recent C++ features. Make sure the latest standards are selected, but it may be too difficult to support it decently.

[angainor]

@mbianco There are two potential complications I see with the bindings. I'm not quite sure if they are a problem yet, but maybe.

At this point it seems to me that if I want to build the Fortran bindings for use with ifort (which is a rather usual choice), then I need to compile at least (part of) the bindings with ifort. So the question is will all work if I use a ifort/g++ hybrid compilation.

Even if it is possible to build the C++ side with a different compiler and use ifort for the fortran side, then there is the question of the MPI library. We can't compile against g++ with OpenMPI if the user is using Intel MPI in his fortran code. Then we need to use the g++ with Intel MPI. Haven't used that, so not sure if there is a standard way to do this.

[mbianco]

Unfortunately this is going to be problematic... Projects we worked on switched to other compilers (without relevant consequences for performance, by the way), especially if they use OpenMP