Re: New (experimental) PETSc Debian package
Adam,
Please feel free to contact us (petsc-maint@mcs.anl.gov) with questions,
suggestions on PETSc installation. We don't know anything about debian but are
interested in helping (a little :-)).
If you like eventually we can keep a debian package version of PETSc at
our web site once it has evolved a bit.
Unfortunately the configure stuff in PETSc doesn't work (it seems pretty
complicated to get it working on all machines since we use a lot of system
stuff.)
Barry
On Thu, 3 Aug 2000, Adam C Powell IV wrote:
> Hooray! My first Debian package! Still preliminary, but I think I'll
> celebrate anyway. :-)
>
> Get the source and i386 binary debs at http://lyre.mit.edu/~powell/debs
> (or follow the links buried in http://lyre.mit.edu/~powell/debian.html).
>
> Issues/questions:
>
> * It's currently in section devel (shlibs in libs, docs in docs),
> maybe it belongs in math?
> * I put the header files in /usr/include/petsc, should they just go
> in /usr/include? There are quite a few of them...
> * Once installed, you can build against it using
> PETSC_DIR=/usr/lib/petsc (bmake is there, include and lib are
> symlinked somewhat appropriately, see below) and PETSC_ARCH=linux.
> * It currently depends on mpich, because there are a couple of things
> I can't find in the lam packages (like MPE, and the machines list).
> * It currently depends on atlas, because it's so much faster! I'll
> change this if anyone complains.
> * Shared libs are built from static ones, which may break on
> platforms requiring -fPIC.
> * There doesn't seem to be a "make install" target, so all of that is
> done in debian/rules, which seems inelegant.
>
> To do (at some point in the future :-):
>
> * Fix the link in /usr/share/doc/petsc-doc/readme.html to point to
> /usr/include/petsc/petscversion.h .
> * Make the binary library dependencies right. (Currently, petsc1
> only depends on libc6, it should depend on mpich, atlas2 and
> xlib6g; petsc-dev should depend on mpich-dev, atlas-dev and
> xlib6g-dev.)
> * Build the C++ wrappers using BOPT=O/c++ and include in packages as
> appropriate.
> * Build both sets of libs with BOPT=g and BOPT=O (and g/c++, O/c++)
> with switching between them in /etc/alternatives. This may be a
> bit messy with my current placement of libs in /usr/lib...
> * Add symlinks from /usr/lib/petsc to readme.html and docs, add the
> makefile, and redo the symlink to /usr/lib so BOPT works right,
> that way PETSC_DIR will have everything people are looking for.
> * Try to put all (well, perhaps most, I don't know what issues may
> arise) of the examples in petsc-dev.examples, or maybe a separate
> petsc-test package, so one can cd
> /usr/share/doc/petsc-[dev|test]/examples and type "make
> BOPT=whatever testexamples testfortran" and it should all just
> work...
> * Oops, just discovered a native configure script in bin! Using this
> would make the patch-debian file a lot smaller; it seems configure
> would be more "elegant" than the current approach, and easier to
> reconcile with future upgrades.
> * Package BlockSolve, ParMETIS, etc., make the PETSc makefiles use
> them, and make the packages depend on them. :-)
>
> So, please try it, bang on it, bugreport and patch, share and enjoy it!
>
> -Adam P.
>
>
>
Reply to: