dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Tue, Jun 03, 2008 at 11:52:30PM +0200, Jed Brown wrote:
> Perhaps I have a strange usage pattern, but I decided to build Dolfin against
> with a different PETSC_ARCH than I had used in the previous build.  Of course,
> scons does not check on PETSC_ARCH if there are pkgconfig files hanging around.
> I remembered this oddity and reset dolfin/scons so that the pkgconfig files
> would be gone.  However, prefix/lib/pkgconfig is also checked so even though
> there was nothing in the Dolfin directory or in my environment which pointed to
> my old PETSC_ARCH, Dolfin was still trying to build against it.  I'm not aware
> of any fix except to manually remove prefix/lib/pkgconfig/{petsc,slepc}.pc in
> order to build against the new one.  I think there should be an option in the
> scons build to ignore/regenerate the pkgconfig and a warning if the environment
> variables don't match the pkgconfig variables.  It may be handy to be able to do
> hg pull -u && scons even with a dirty environment, but there ought to an easier
> way to make scons pay attention to the environment.
> 
> I'm also not sure that putting the current pkgconfig in prefix is necessarily a
> friendly thing to do since there may be many different installs around.  For
> instance, I have several of the combinations from
> petsc-{2.3.3-p12,dev}{openmpi,mpich2}{c,c++}{debug,opt} and the one I am
> currently building Dolfin with may not be the one I use for something else.  I
> don't know of any other software that looks at petsc.pc since other software
> generally just uses the current environment, but this isn't really an excuse for
> forcefully writing config information for a different package into a potentially
> global location.
> 
> 
> If you think my usage is odd and are happy with the current build system, feel
> free to ignore me.  I just feel like this is liable to trip people up.
> 
> Jed

I also think we should not install pkgconfig files for other packages
than the one being installed. It should be enough to cache them
locally (as we do now in scons/pkgconfig).

On the other hand, it would be *very* nice if PETSc could generate
pkg-config files and not rely on environment variables and having to
include Makefiles.

-- 
Anders

Attachment: signature.asc
Description: Digital signature