[DebianGIS-dev] How to use libhdf5-serial-dev and -mpi/-lam friends correctly?

[Thomas Weber]

On Mon, Oct 18, 2010 at 11:03:00PM +0200, Francesco P. Lovergine wrote:
> On Sun, Oct 03, 2010 at 11:40:41AM +0200, Thomas Weber wrote:
> > Hi, 
> > 
> > [Please CC me, I'm not subscribed].
> > 
> > I'm currently faced with #598227, 
> > http://bugs.debian.org/598227
> > which concerns the way Octave uses hdf5 and its header files (or rather,
> > how hdf5 includes mpi.h). Please read the bug report for information,
> > it's presented there far better than I could formulate it.
> > 
> > Background information:
> > The preferred way to build binary 'extensions' or 'plugins' for Octave
> > is using the mkoctfile shell script. I'm not aware of any software in
> > Debian that does not use mkoctfile for building such extension.
> > 
> > Now, I could fix the above bug by hacking mkoctfile, but somehow this
> > looks like the wrong solution to me. If I include hdf.h, do I really
> > have to care about mpi which may or may not be used?
> > 
> > So, are there better solutions than adding /usr/include/mpi/ to
> > mkoctfile's include flags?
> > 
> 
> 
> I see no reasons why mpi.h has to be included but for the case you
> are using one of the -mpi flavors. In that case you should depend
> on libhdf5-mpi-dev. I don't undestand why you are trying to use
> both -serial and -parallel flavors.

Both -serial and parallel flavors are used due to users' demand:
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=231114
It has been like this for at least 6 years (I don't use any of the hdf5
features of Octave). 
It used to work all the time without problems; or maybe there were
problems all the time and nobody noticed, who knows.

Anyway, I don't mind dropping the parallel hdf5 support if that's the
easiest way forward. Do you see another way?

	Thomas