Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

From: Ralf Wildenhues (Ralf.Wildenhues_at_[hidden])
Date: 2005-09-01 02:01:38


Hi George,

* George Bosilca wrote on Thu, Sep 01, 2005 at 06:49:48AM CEST:
>
> Now I see the reason behind this change. Anyway, few month ago we decide
> to switch the compilation process, and to modify all the files in order to
> start all the #include directives with the full path of the include files
> (starting the main components top directories).

I can see the reason behind this. For external packages, esp. ones that
are replaced every time you run autogen.sh, this is a bit dangerous,
however.

> Personally, I prefer to keep this rule for all things inside, libltdl
> included. Later, when the libtool-2.x will became available we can add
> a define in the ompi_config.h that will trigger the correct include.

It's not my decision which way OpenMPI will go, but: I'd volunteer to
provide a patch which fixes all the include paths if that can be agreed
upon.

My patch is motivated by the fact that I am trying to use OpenMPI as
testing ground for Libtool-2.x. OpenMPI is a good candidate package for
several reasons: it exercises libtool fairly thoroughly, it uses
libltdl, it is big (and as such not a model project), and I happen to
have used its predecessor anyway. Advantage for both sides: hopefully
the transition will be as smooth as possible.

If you choose not to fix the include paths, you will have to modify
ltdl.h in autogen.sh after each `libtoolize --ltdl' for 2.x. I would
provide a patch to this end, too, but not guarantee that this would be
forward-compatible.

You get to decide which way to go. :)

> If we provide a ltdl.h file and finally the compilation use the one from
> /usr/include that's really confusing.

If you configure with --enable-ltdl-convenience (which is the default in
OpenMPI), then yes, that would be a bug.

Cheers,
Ralf