Why is adding an else case solving a compiler warning? I don't get that.

NVCC was complaining that the trailing return statements were unreachable. I guess this is because nvcc is a little over zealous for analyzing const expressions? I agree that it doesn't make much sense, it doesn't hurt either, though.

Huh, any idea why CircleCI would fail? This particular issue (docker login failing) doesn't seem to be related to my change.

@gentryx it decided to run on your circleci account (no idea why), so it can't access our (CI) environment variables (like the docker credentials).

Hmm, I remember similar issues from the past.

Anyway, is there any blocker to merge this?

I don't mind merging this, but it does undo some clang-tidy warnings that were fixed in #3154, which will come back next time someone decides to run clang-tidy. Being able to successfully compile is more important than being clang-tidy though.

What version of NVCC were you using? How much of HPX did you try to compile?

Necessary for compilation is only the change to hpx/runtime/components/server/managed_component_base.hpp.

The other changes are just to make nvcc compile without issuing tons of warnings. So, I should be good if someone ran clang-tidy in the future. :-)

I'm using CUDA 9.1.85, I was compiling from HPX only what's required by LibGeoDecomp. Not sure which parts that are exactly.

PTAL

@gentry: @msimberg is the master of ceremonies for the upcoming release - he will decide whether this should go in or not.

Thanks! Will add this to the release.

Thanks! :-)