Carl Pearson

Can close https://github.com/kokkos/kokkos-kernels/issues/989 when this is done

Hi @schulzchristian, just wanted to let you know I haven't forgotten about this. I am a few days from defending my Ph.D thesis and then I will sort this out.

@crtrott is this the way to disable fencing? https://github.com/cwpearson/kokkos-tools/blob/ad8848843e6fed0c0c3d76a8689b648eba44f3aa/profiling/nvtx-connector/kp_nvtx_connector.cpp#L60-L62 If so, I think we're good :D

Based on discussion with @crtrott as well, a better plan is to just implement `operator==`, which will make Spaces more well-behaved value types but not address this container question. Perhaps...

After further discussion with @crtrott and @PhilMiller we can try for `operator==` and `std::hash`, so as to avoid implying an ordering of execution spaces but we still get `std::unordered_map`

> Is there testing covering this code? There really ought to be. Ideally, you would run it for many repetitions to see that it never fails, to try to detect...

Am I understanding correctly that * we like that `single` is shown in the e.g. [TeamThreadRange](https://kokkos.github.io/kokkos-core-wiki/API/core/policies/TeamThreadRange.html) docs * we'd like a standalone reference for it, like the many listed on...

It's not obvious to me what is wrong with the currently-failing test.

Here's an issue to track the `Kokkos::single` documentation: https://github.com/kokkos/kokkos/issues/5459

Looks like there was an infrastructure failure of some kind in the test - can this be marked for retest once that is resolved?