Francesco Rizzi

> > condensing the exespace and team impl into one is not feasible because of [#5150 (comment)](https://github.com/kokkos/kokkos/pull/5150#issuecomment-1199934499) > > The main problem seems to be warnings or errors for executing...

> We have > > https://github.com/kokkos/kokkos/blob/56ef02c0cd45b62c39b5a789488467331822499d/core/src/impl/Kokkos_Error.hpp#L201-L205 > > for example where > https://github.com/kokkos/kokkos/blob/56ef02c0cd45b62c39b5a789488467331822499d/core/src/impl/Kokkos_Error.hpp#L36 > > This is similar to what `replace_impl` would be doing, right? You can call both from...

> Hmm... would you indulge me and discuss this on a godbolt example? My understanding is that https://godbolt.org/z/MfnzPn6oq would do what we want. What am I missing? yeah that is...

@masterleinad unless you have major objections, i think we should try to merge this PR in its current form and then i will work on a follow up PR to...

decided during meeting to break this apart, so leaving it draft for now

replace by smaller ones

closing for now since sort API was revised

@rwgk i added a test too, which I am building locally now - will push once i have it

Agree on fixing things like having mutable_at call this etc. But wanted to first discuss the choice of overloading `operator()` below > Therefore I'd lean toward avoiding operator() for this...

so you are suggesting doing this? ```cpp template T &operator()(Ix... index) { #ifndef NDEBUG bounds check here #endif return *(static_cast(array::mutable_data()) + byte_offset(ssize_t(index)...) / itemsize()); } ``` if so, I agree...