Daniel Arndt
> Please update the title and make sure the description says that it does not actually guarantee proper enqueuing. How would you define "enqueueing properly"? The definition I'm following is...
I amended the description some for the expected behavior.
> The serial tests are timing out Reverting https://github.com/kokkos/kokkos/pull/6151/commits/ac408ddfe4c3ae327f014d2c6dad31b71b3519c0 fixed that.
> Please update the PR title (see [#6151 (review)](https://github.com/kokkos/kokkos/pull/6151#pullrequestreview-2164014790)) done
> Did you measure improvement in build time? https://github.com/dealii/dealii/issues/16065#issuecomment-1745015746 includes some other fixes but none of them changed build time significantly but the respective compile items vanished from the report....
> Work in progress: using unit tests for reproduction ([b99aaa2](https://github.com/kokkos/kokkos/commit/b99aaa2cfe083626e4fc028f5af75e033f91d604)) didn't work for me, so I have moved the reproducer to `examples`. You were missing ```diff diff --git a/core/unit_test/CMakeLists.txt b/core/unit_test/CMakeLists.txt...
It should be enough to set the device before creating the singleton stream which we should do now anyway.
I would be very interested to see if the test added here just works out-of-the-box now.
Retest this please.
Locally this seems to work fine without extra changes on `Cuda` but we still have similar complaints when using `MDRangePolicy`, see https://kokkosteam.slack.com/archives/C5BGU5NDQ/p1706727845918379. It might be necessary to do ```C++ diff...