Re: [Boost-mpi] [EXTERNAL] regression on develop

Subject: Re: [Boost-mpi] [EXTERNAL] regression on develop
From: Alain Miniussi (Alain.Miniussi_at_[hidden])
Date: 2017-07-24 07:56:07

On 24/07/2017 05:29, Belcourt, Kenneth wrote:
>
>> On Jul 23, 2017, at 2:24 AM, Alain Miniussi <alain.miniussi_at_oca.eu 
>> <mailto:alain.miniussi_at_oca.eu>> wrote:
>>
>> On 23/07/2017 08:35, Alain Miniussi wrote:
>>
>>> On 23/07/2017 05:09, Belcourt, Kenneth wrote:
>>>>> On Jul 22, 2017, at 5:12 PM, Alain Miniussi <Alain.Miniussi_at_oca.eu 
>>>>> <mailto:Alain.Miniussi_at_oca.eu>> wrote:
>>>>> On 23/07/2017 00:41, Alain Miniussi via Boost-mpi wrote:
>>>>>> On 22/07/2017 22:11, Belcourt, Kenneth wrote: I see that your 
>>>>>> recent changes have now broken develop for clang-linux-3.8.1, and 
>>>>>> gcc-4.7.2. So I guess Boost is dropping support for gcc-4.7.x, or 
>>>>>> was that unintentional on your part Alain?  Can you look at that 
>>>>>> and see if we can either revert the broken develop change, or fix 
>>>>>> it so it’s still buildable with gcc-4.7?
>>>>>> Unintentional, and weird, the change were already on develop, was 
>>>>>> just a resync to make them unconflictig with master.
>>>>>
>>>>> Work fine with 4.4.7 and intel mpi 5.0.3.
>>>>> Can you point me to the error messages ?
>>>>
>>>> They were all green, but perhaps I’m not patient enough and tests 
>>>> need to cycle again.  Look at the Sandia-gcc-4.7.2, 
>>>> Sandia-gcc-4.7.2-c++11, and Sandia-clang-3.8, all available from 
>>>> this link.
>>>>
>>>> http://www.boost.org/development/tests/develop/developer/mpi.html
>>>>
>>>> Thanks for looking into this Alain.
>>>
>>> There is an issue specific to the 2 cartesian test which is probably 
>>> https://github.com/boostorg/mpi/pull/47 waiting for merge (I'll 
>>> check and merge if relevant). These are the ones failing with 
>>> clang3.8 and warning with g++5.4.0/no c++11. I'll check with 4.4.7/no11.
>> Checked with 4.3 to be safe. I merged 47, those should be ok.
>
> Looks like different issues now:
>
> gcc.compile.c++ 
> ../../../bin.v2/libs/mpi/build/gcc-4.7.2/debug/threading-multi/point_to_point.o
>
> ../../../libs/mpi/src/point_to_point.cpp: In function ‘void 
> boost::mpi::detail::packed_archive_send(MPI_Comm, int, int, const 
> boost::mpi::packed_oarchive&)’:
> ../../../libs/mpi/src/point_to_point.cpp:32:3: error: invalid 
> conversion from ‘const void*’ to ‘void*’ [-fpermissive]
> In file included from ../../../boost/mpi/config.hpp:20:0,
>                  from ../../../boost/mpi/detail/point_to_point.hpp:12,
>                  from ../../../libs/mpi/src/point_to_point.cpp:20:
> /SDK/mpi/openmpi/1.6.4-gcc-4.7.2-RHEL6/include/mpi.h:1384:20: error:   
> initializing argument 1 of ‘int MPI_Send(void*, int, MPI_Datatype, 
> int, int, MPI_Comm)’ [-fpermissive]
>
Looks like a non-conforming MPI implementation (according to 
http://mpi-forum.org/docs/mpi-3.0/mpi30-report.pdf, the first argument 
of MPI_Send *is* a "void const*").
Weird, I tested with open MPI...
I'm looking into it.

Alain