[MOAB-dev] moab/ParallelComm.hpp include order

[Nico Schlömer]

Well, perhaps removing just the workaround will do. I'll test it...

--Nico

On Tue, Nov 3, 2015 at 8:55 AM Nico Schlömer <nico.schloemer at gmail.com>
wrote:

> > MPI_INCLUDED is defined in mpi.h (easy to check)
>
> Not in my installation. This
> ```
> #include <iostream>
> #include <mpi.h>
>
> int main() {
> #ifdef MPI_INCLUDED
>   std::cout << "defined" << std::endl;
> #else
>   std::cout << "NOT defined" << std::endl;
> #endif
>   return 0;
> }
> ```
> Returns `NOT defined` on my machine (OpenMPI 1.6.3); check out [1].
>
> I checked the history of `./src/parallel/moab_mpi.h` and found that it's
> had it's first and last changes in 2010. Perhaps the issues it tried to
> work around don't exist anymore nowadays.
>
> I would suggest removing the file `./src/parallel/moab_mpi.h` and have
> some people test it. Alright?
>
> --Nico
>
> [1] https://github.com/open-mpi/ompi/blob/master/ompi/include/mpi.h.in
>
>
> On Tue, Nov 3, 2015 at 12:34 AM Vijay S. Mahadevan <vijay.m at gmail.com>
> wrote:
>
>> I cant seem to replicate the issue locally. Nevertheless, that file
>> definitely looks quite strange since I don't think we hit many of the
>> paths in moab_mpi.h as MB_SEEK*/MB_RESTORE* are not defined in
>> configure (autotools) anymore. Certainly, not in CMake.
>>
>> MPI_INCLUDED is defined in mpi.h (easy to check) and this just checks
>> if you have previously included the header already (note that it is
>> ifndef and not ifdef). So I'm not worried about the undefs really
>> unless your MPI version does not define MPI_INCLUDED
>> (openmpi/intel-mpi ??).
>>
>> Now in your case, if all of the MPI variables are getting undefined,
>> we need to track why that is the case. Looking at the history for that
>> file didn't reveal much but let us know your configuration command for
>> MOAB so that we can create a build that replicates the failure.
>>
>> Vijay
>>
>>
>> On Mon, Nov 2, 2015 at 5:09 PM, Nico Schlömer <nico.schloemer at gmail.com>
>> wrote:
>> > @Iulian, oh yes, you're right. But the error remains the same:
>> > ```
>> > MPI_Comm my = MPI_COMM_WORLD;
>> > ```
>> > The reason lies `moab_mpi.h`. There we find
>> > ```
>> > /* Work around brain-dead netcdf.h */
>> > #ifndef MPI_INCLUDED
>> > #  undef MPI_Comm
>> > #  undef MPI_Info
>> > #  undef MPI_COMM_WORLD
>> > #  undef MPI_INFO_NULL
>> > #endif
>> > ```
>> > Whoops! All of MPI undefined. No idea what's going on here and why.
>> >
>> > Any insight?
>> >
>> > --Nico
>> >
>> > On Tue, Nov 3, 2015 at 12:02 AM Grindeanu, Iulian R. <
>> iulian at mcs.anl.gov>
>> > wrote:
>> >>
>> >> this error has nothing to do with moab/ParallelComm.hpp
>> >>
>> >> MPI_COMM_WORLD is just a constant, not a type
>> >>
>> >> this should give compile errors too.
>> >>
>> >> Or I don't get what are you trying to do
>> >>
>> >> #include <mpi.h>
>> >>
>> >> int main() {
>> >>   MPI_COMM_WORLD my;
>> >> }
>> >>
>> >>
>> >> ________________________________
>> >> From: moab-dev-bounces at mcs.anl.gov [moab-dev-bounces at mcs.anl.gov] on
>> >> behalf of Nico Schlömer [nico.schloemer at gmail.com]
>> >> Sent: Monday, November 02, 2015 4:49 PM
>> >> To: Vijay S. Mahadevan
>> >> Cc: moab-dev at mcs.anl.gov
>> >> Subject: Re: [MOAB-dev] moab/ParallelComm.hpp include order
>> >>
>> >> Alright. This
>> >> ```
>> >> #include <mpi.h>
>> >> #include <moab/ParallelComm.hpp>
>> >>
>> >> int main() {
>> >>   MPI_COMM_WORLD my;
>> >> }
>> >> ```
>> >> creates the same
>> >> ```
>> >>  error: ‘MPI_COMM_WORLD’ was not declared in this scope
>> >> ```
>> >> Can you reproduce?
>> >>
>> >> Cheers,
>> >> Nico
>> >>
>> >> On Mon, Nov 2, 2015 at 11:38 PM Nico Schlömer <
>> nico.schloemer at gmail.com>
>> >> wrote:
>> >>>
>> >>> Including <moab/MOABConfig.h> doesn't help. I'll see about creating a
>> >>> minimal example...
>> >>>
>> >>> --Nico
>> >>>
>> >>> On Mon, Nov 2, 2015 at 10:05 PM Vijay S. Mahadevan <vijay.m at gmail.com
>> >
>> >>> wrote:
>> >>>>
>> >>>> Can you try including moab/MOABConfig.h directly in your test before
>> >>>> ParallelComm and see if that resolves issues ?
>> >>>>
>> >>>> ParallelComm includes Interface.hpp which in turn includes
>> >>>> MOABConfig.h. If you are including some other MOAB header before
>> >>>> these, MOAB_HAVE_MPI might not be defined.
>> >>>>
>> >>>> Vijay
>> >>>>
>> >>>> On Mon, Nov 2, 2015 at 3:00 PM, Nico Schlömer <
>> nico.schloemer at gmail.com>
>> >>>> wrote:
>> >>>> > It's a little convoluted with Trilinos includes right now, I'm
>> trying
>> >>>> > to
>> >>>> > generate a minimal example.
>> >>>> >
>> >>>> > --Nico
>> >>>> >
>> >>>> > On Mon, Nov 2, 2015 at 10:00 PM Vijay S. Mahadevan <
>> vijay.m at gmail.com>
>> >>>> > wrote:
>> >>>> >>
>> >>>> >> Nico, can you give an example ? What else did you include before
>> >>>> >> ParallelComm to get such an error ? I am guessing this has to do
>> with
>> >>>> >> how moab/MOABConfig.h gets included since that's where
>> MOAB_HAVE_MPI
>> >>>> >> is defined.
>> >>>> >>
>> >>>> >> Vijay
>> >>>> >>
>> >>>> >> On Mon, Nov 2, 2015 at 2:47 PM, Nico Schlömer
>> >>>> >> <nico.schloemer at gmail.com>
>> >>>> >> wrote:
>> >>>> >> > When playing with MOAB, I found that if I include the header
>> >>>> >> > ```
>> >>>> >> > #include <moab/ParallelComm.hpp>
>> >>>> >> > ```
>> >>>> >> > after something that's also using MPI, I'm getting strange
>> >>>> >> > compilation
>> >>>> >> > errors of the kind
>> >>>> >> > ```
>> >>>> >> > error: ‘MPI_COMM_WORLD’ was not declared in this scope
>> >>>> >> > ```
>> >>>> >> > Has anyone else experienced this?
>> >>>> >> >
>> >>>> >> > Cheers,
>> >>>> >> > Nico
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/moab-dev/attachments/20151103/688a5e83/attachment.html>