Re: [PATCH 1/4] gnu: openmpi: Add memchecker support

[ludo@gnu.org (Ludovic Courtès)]

Eric Bavier <ericbavier@gmail.com> skribis:

> Ludovic Courtès writes:
>
>> Eric Bavier <ericbavier@gmail.com> skribis:
>>
>>> From 3aaab98e3b9a0b1b893dd5bc64e098136364e717 Mon Sep 17 00:00:00 2001
>>> From: Eric Bavier <bavier@member.fsf.org>
>>> Date: Sat, 17 May 2014 00:02:44 -0500
>>> Subject: [PATCH 1/4] gnu: openmpi: Add memchecker support
>>>
>>> * gnu/packages/mpi.scm (openmpi) [inputs]: Add valgrind.
>>>   [arguments]: Add configure flags for memchecker.
>>
>> [...]
>>
>>>      (native-inputs
>>>       `(("pkg-config" ,pkg-config)))
>>>      (propagated-inputs
>>> -     `(("gfortran" ,gfortran-4.8)))
>>> +     `(("gfortran" ,gfortran-4.8)
>>> +       ("valgrind" ,valgrind)))
>>
>> Why is Valgrind propagated?  I would expect it to be used only in
>> OpenMPI’s test suite, no?
>
> The intention of OpenMPI's memchecker option is to allow one to more
> easily use valgrind when executing mpi programs.  That being said, it is
> an optional feature, so it may not need to be propagated.  I'll look at
> that again.

But that still makes Valgrind optional when using Open MPI, no?  I would
not propagate it if it’s optional.

>> (I also wonder why GFortran is propagated, probably overlooked it
>> before.)
>
> GFortran needs to be propagated for the OpenMPI compiler wrappers.

OK.  I would not propagate it either because, suppose you just want to
use the library in a C program, then it doesn’t make sense to pull the
Fortran compiler.

WDYT?

Ludo’.