Re: [Boost-mpi] Boost 1.62.0 with MS-MPI v7

Subject: Re: [Boost-mpi] Boost 1.62.0 with MS-MPI v7
From: alainm (alain.miniussi_at_[hidden])
Date: 2016-11-09 08:44:31

There is a chance that I won't be involved in Boos::mpi anymore, so I 
won't be able to help on that one. But I can still talk about it :-)
Now, as far as MS-MPI is concerned, I think that the main problem is 
that most HPC platforms are not using MS in the first place. As a result 
the platform is not very well (read, at all) tested.
I do not have access to a windows platform, and all the HPC facilities I 
use are running Linux, so validating a patch would be tricky. I'm afraid 
that most peoples involved in boost::mpi are in that situation.

And then there is the boost build system. Even on Linux, the support 
for, say, Intel's MPI was not easy to provide. The Bjam build system is 
tailored for MPI implementations based on Open-MPI and Mpich (it runs 
"mpicc -show<xxx> to find out the compilation options, so if those 
options are not available or do not work as expected, you're in 
trouble), and I don't really understand how the jam thing works (but 
then, who does?) in order to make it more flexible. I could try reading 
the doc.. (just kidding).

I do not think it's going to improve soon, as the only solution that 
boost could come up with to fix the bjam issues is to develop a new bjam...

As a side note, I'm starting a small project as a fork of boost::mpi and 
moved the build system to CMake, and it's such a comfort !
For example, if you need to add a test, you can use the add_test 
command, for which there is a documentation, in the cmake documentation 
(provided with cmake).

Thruth be told, I am not a big fan of CMake, except when I have to work 
with bjam.

Now, maybe someone with access to MS-MPI can try submitting a 
documentation  PR through github.

Cheers

Alain

On 09/11/2016 12:21, csiga.biga_at_[hidden] wrote:
>
> Hi fellow boost::mpi users!
>
> I came across this stackoverflow post which solves building boost::mpi 
> using MS-MPI, but my question is: could someone already involved in 
> boost::mpi development mainline this patch?
>
> Truth be told, I am no big fan of Boost.Build, mostly due to rough 
> edges such as this one. I was out of my mind when I found the project 
> that aimed on using CMake as a primary build system generator for 
> Boost, but was also deeply saddened by the fact that the project was 
> recieved with lethal resistance. FindMPI.cmake is in much better shape 
> than mpi.jam in this regard.
>
> Or is MS-MPI not supported at all? The project page would suggest so, 
> but the build scripts do otherwise. Or are these the remnants of 
> legacy support? MS-MPI has recieved updates in the past few years, so 
> it is far from being discontinued and is in farily good shape. So why 
> is it not on the project webpage?
>
> Thanks in advance,
>
> Máté
>
>
>
> _______________________________________________
> Boost-mpi mailing list
> Boost-mpi_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-mpi

-- 
Alain Miniussi
Pole Génie Logiciel Scientifique
Observatoire de la Côte d'Azur
Blv de l'Observatoire, Nice