Daniel Arndt
Would you mind rebasing on `develop` so we get a clearer picture from CI?
For reference, I am seeing ``` Time 0.001588 s Time 0.003619 s Time 0.001705 s Time 0.002141 s ``` on my MacBook with a `Release` build.
Retest this please.
> do we need to re-review this? The only thing that really changed here is moving fences from `finalize` to the `destructor` and changing the error message accordingly.
It might be good to make sure we don't copy the callbacks but just use a reference, like ```diff diff --git a/core/src/impl/Kokkos_Profiling.cpp b/core/src/impl/Kokkos_Profiling.cpp index 53d1fbfb2..c0a025394 100644 --- a/core/src/impl/Kokkos_Profiling.cpp +++ b/core/src/impl/Kokkos_Profiling.cpp...
That configuration command doesn't enable tests. `tpls/mdspan/include/experimental/__p0009_bits/config.hpp` has ``` #if !defined(MDSPAN_IMPL_USE_ATTRIBUTE_NO_UNIQUE_ADDRESS) # if ((MDSPAN_IMPL_HAS_CPP_ATTRIBUTE(no_unique_address) >= 201803L) && \ (!defined(__NVCC__) || MDSPAN_HAS_CXX_20) && \ (!defined(MDSPAN_IMPL_COMPILER_MSVC) || MDSPAN_HAS_CXX_20)) # define MDSPAN_IMPL_USE_ATTRIBUTE_NO_UNIQUE_ADDRESS 1...
> I am proposing to bundle a copy because I expect that has more chances to get consensus. I'd prefer to use FetchContent_Declare like we do for benchmarks (and switch...
Looks like we need the patch in https://github.com/google/googletest/issues/3554#issuecomment-2636883816 for making Windows with shared libraries work.
> I see no reason why not to do the same on the macos workflows. I also don't see a reason to change it. Or is there? 🙂
> I would suggest building on Mac with C++20? Do we aim to target mandatory C++20 support in the near future anyway? Probably for deal.II 10. 😉