On Wed, Feb 22, 2012 at 09:34:20AM +0100, Julien Cristau wrote:
> On Tue, Feb 21, 2012 at 20:10:17 -0600, Steve M. Robbins wrote:
> > On Tue, Feb 21, 2012 at 03:01:12PM +0100, Julien Cristau wrote:
> > 
> > > I guess the best would be to reintroduce them but as part of a separate
> > > libhdf5-cpp-7 binary package (the reason I dropped them was that they
> > > were in libhdf5-7 but not in the mpi variants, which meant the mpi
> > > variants didn't really "Provide" libhdf5-7).
> > 
> > Maybe.  However, there are already four variants of C bindings:
> > serial, mpich2, openmpi, and default "mpi".  Splitting C++ libraries
> > would mean doubling this to 8.
> > 
> > Why is that better than simply reactiving C++ (for all variants)?
> > 
> First because having more than one shared lib per binary package is a
> recipe for trouble down the road.

Not always.  It's fine as long as they all increment SONAME together.
I don't know if that's the case here, but if it is, I'd say the
benefit of splitting is outweighed by the nuisance of increasing
package count.

> Second, the c++ libs were only built
> with the serial libhdf5 variant, not with any of the mpi ones; 

Yes, that's why I specified to produce C++ libs "for all variants".

> I don't know why, and am not familiar enough with hdf5 to know if
> that makes any difference.

I don't know why either, but I do know that it makes a difference.


Attachment: signature.asc
Description: Digital signature

Pkg-grass-devel mailing list

Reply via email to