Potentially add a conflict related to the error with gcc@9 [email protected] and fmt@10

[aurianer]

Error:
https://gitlab.com/cscs-ci/ci-testing/webhook-ci/mirrors/479009878135925/5304355110917878/-/jobs/5574910212

/opt/spack-495252f7f6f1e51842ccf37a44b8bf2c8ddb8950/opt/spack/linux-ubuntu22.04-broadwell/gcc-9.3.0/fmt-10.1.1-tbu42poqymdpq54qldfaskewebo7lw5z/include/fmt/core.h(1576): error: incomplete type is not allowed
          detected during:
            instantiation of "auto fmt::v10::detail::make_arg<PACKED,Context,T,<unnamed>>(T &)->fmt::v10::detail::value<Context> [with PACKED=true, Context=fmt::v10::format_context, T=const pika::cuda::experimental::cusolver_handle, <unnamed>=0]" 
(1808): here
            instantiation of "fmt::v10::format_arg_store<Context, Args...>::format_arg_store(T &...) [with Context=fmt::v10::format_context, Args=<const char *, int, const char *, const char *, pika::cuda::experimental::cusolver_handle, pika::cuda::experimental::cusolver_handle>, T=<const char *, int, const char *, const char *, const pika::cuda::experimental::cusolver_handle, const pika::cuda::experimental::cusolver_handle>]" 
(1826): here
            instantiation of "auto fmt::v10::make_format_args(T &...)->fmt::v10::format_arg_store<Context, fmt::v10::remove_cvref_t<T>...> [with Context=fmt::v10::format_context, T=<const char *, int, const char *, const char *, const pika::cuda::experimental::cusolver_handle, const pika::cuda::experimental::cusolver_handle>]" 
/opt/spack-495252f7f6f1e51842ccf37a44b8bf2c8ddb8950/opt/spack/linux-ubuntu22.04-broadwell/gcc-9.3.0/fmt-10.1.1-tbu42poqymdpq54qldfaskewebo7lw5z/include/fmt/ostream.h(179): here
            instantiation of "void fmt::v10::print(std::ostream &, fmt::v10::format_string<T...>, T &&...) [with T=<const char *&, int &, const char *&, const char *&, const pika::cuda::experimental::cusolver_handle &, const pika::cuda::experimental::cusolver_handle &>]" 
/pika/source/testing/testing/include/pika/testing.hpp(135): here
            instantiation of "__nv_bool pika::detail::fixture::check_not_equal(const char *, int, const char *, pika::detail::counter_type, const T &, const U &, const char *) [with T=pika::cuda::experimental::cusolver_handle, U=pika::cuda::experimental::cusolver_handle]" 
/pika/source/libs/pika/async_cuda/tests/unit/cusolver_handle.cu(78): here
/opt/spack-495252f7f6f1e51842ccf37a44b8bf2c8ddb8950/opt/spack/linux-ubuntu22.04-broadwell/gcc-9.3.0/fmt-10.1.1-tbu42poqymdpq54qldfaskewebo7lw5z/include/fmt/core.h(1579): error: static assertion failed with "Cannot format an argument. To make type T formattable provide a formatter<T> specialization: https://fmt.dev/latest/api.html#udt"
          detected during:
            instantiation of "auto fmt::v10::detail::make_arg<PACKED,Context,T,<unnamed>>(T &)->fmt::v10::detail::value<Context> [with PACKED=true, Context=fmt::v10::format_context, T=const pika::cuda::experimental::cusolver_handle, <unnamed>=0]"

cuda 11.8 config passes, this is the one currently used in #863

[msimberg]

For more context, this only happens with GCC 9, and it starts working again with CUDA 11.3 and newer. I will be adding a conflict to DLA-Future since it actually triggers the problem, whereas just building the core pika library does not trigger the problem. It's only triggered in device code which we only have for tests.

In CI I would simply enforce a version of fmt that works.

[msimberg]

We can keep this open until I've added the conflict to spack.