C++17 Woes

[kab163]

when we go to c++17, we get a bunch of camp warnings like:

/g/g0/belcher6/Umpire/src/tpl/umpire/camp/include/camp/resource/hip.hpp:181:11: warning: ignoring return value of function declared with 'nodiscard' attribute [-Wunused-result]
  181 |           campHipErrchk(
      |           ^~~~~~~~~~~~~~
  182 |               hipStreamWaitEvent(get_stream(), hip_event->getHipEvent_t(), 0));
      |               ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/g/g0/belcher6/Umpire/src/tpl/umpire/camp/include/camp/defines.hpp:195:28: note: expanded from macro 'campHipErrchk'
  195 | #define campHipErrchk(ans) ::camp::hipAssert((ans), #ans, __FILE__, __LINE__)
      |                            ^~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

and I can go in to all the function calls of campHipErrchk and add a (void) to the beginning of the line and that will resolve it, but then it continues on to

/g/g0/belcher6/Umpire/src/tpl/umpire/camp/test/resource.cpp:204:3: warning: ignoring return value of function declared with 'nodiscard' attribute [-Wunused-result]
  204 |   hipStreamCreate(&stream1);
      |   ^~~~~~~~~~~~~~~ ~~~~~~~~

(And so on...)

So unless we use a -Wno-unused-result flag when building (or some other similar workaround), then it looks like C++17 wants every function return to be used.

Umpire would like to use C++17 in the next release, but this is creating a long list of warnings in addition to our own set of warnings (same type). How should we proceed?

[MrBurmark]

I think starting with adding the equivalent of casting to void or RAJA_UNUSED_VAR into campHipErrchk. The next one you mentioned looks like a bug as it is not error checked, so we should fix that and fix others.

[rhornung67]

I recently added a macro to camp analogous to RAJA_UNUSED_VAR : https://github.com/LLNL/camp/blob/main/include/camp/defines.hpp#L28

I used it to quiet some compiler warnings.

[kab163]

Oh I see, yea that might be useful to add for us too. We do use a UMPIRE_UNUSED_VAR for similar things...

[kab163]

The difficulty is that sometimes the campHipErrchk return value is used (see here) and sometimes it is not (as in the errors above)..

I will start a branch to work on this. Either making it 2 separate macro's (maybe NoReturnCampHipErrchk and campHipErrchk) or perhaps some other method...

[trws]

Is this all resolved by #171?

[kab163]

Is this all resolved by #171?

Yes!