dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Jul 01, 2005 at 01:52:08PM -0400, Faheem Mitha wrote:
> 
> 
> On Fri, 1 Jul 2005, Anders Logg wrote:
> 
> >It shouldn't be a problem getting PETSc + DOLFIN running on
> >Debian. I'm using Debian myself.
> 
> [snip]
> >It can be a little tricky to get PETSc installed, since they don't 
> >follow the standard configure + make. (On the other hand, they are using 
> >Matt Knepley's new BuildSystem which is pretty powerful.)
> 
> I wonder why they aren't using scons.

Because Matthew Knepley at Argonne (PETSc maintainer) wrote his own
system in Python which is presumably better. (I have not compared it
with scons, but I trust Matt.)

> >in the DOLFIN directory to compile DOLFIN. (And then an optional make
> >install to install it on your system.) DOLFIN assumes that PETSc is in
> >/usr/local/lib/petsc/ so either make a symlink to where you compiled
> >PETSc or give the full path name with --with-petsc-dir=<path>. DOLFIN
> >also assumes that you want PETSc BOPT=g_c++, so if you want to compile
> >with PETSc BOPT=O_c++, use the flag --with-petsc-bopt=O_c++.
> 
> Yes, but does PETSc require the sources in order to compile, or just the 
> static/shared libraries + headers?

You mean if DOLFIN needs the PETSc sources? No, it shouldn't. But
you will need the sources for PETSc to compile PETSc.

> [snip]
> >It's getting ready. There will be a new version later today (or early
> >next week, as soon as a new version of FIAT comes out) with many
> >improvements and fixes.
> 
> What is FIAT?

The finite element backend of DOLFIN developed by Robert C. Kirby, see
http://www.fenics.org/fiat/.

> >If you want to try out this version, download the latest CVS snapshot of 
> >DOLFIN from
> 
> >   http://www.fenics.org/pub/software/dolfin/cvs/
> 
> Ok. Have you considered moving to Subversion (or something else) at some 
> point?

Yes, I'm waiting for the SCM situation to stabilize and then pick
something. CVS will do fine until then. I've been looking at Mercurial
which I think looks really promising. It's Python which shows proof of
good taste... :-)

> >Take a look in the subdirectory src/demo/poisson/ of DOLFIN, which
> >contains a simple demo program with some instructions.
> >
> >Things are starting to stabilize enough now for us to write a manual,
> >but I can't promise that we'll have it ready until after the summer.
> >
> >What does Johan + Johan say? Are you willing to contribute to the
> >manual?
> 
> Are you asking me?

I was asking Johan Hoffman and Johan Jansson (DOLFIN maintainers), but
anyone can contribute.

> >Yes, we will switch to 2.3 as soon as possible. I'll gladly accept any
> >patches for this.
> 
> You mean patches in order for DOLFIN to work against PETSc 2.3?

Yes.

> >There are no Debian packages yet, but I would like to see DOLFIN (and
> >the rest of the FEniCS projects) added to Debian at some point.
> 
> Well, I usually want things that I am using packaged.
> 
> I'm going to try using the preexisting packaging for Debian 2.2.0 and 
> convert it to 2.2.1 (if I can). There are tools like svn-buildpackage that 
> can help with this.

I don't think that will work. Things have changed in PETSc since 2.2.0
(like the number of arguments to functions).

> It seems best to stay with the current Debian packaging system, since it 
> is presumably policy-compliant etc.

Yes, but the PETSc Debian package has not been updated in a year.

> The question then, is whether the DOLFIN developers are willing to help 
> get DOLFIN to compile against this version out of the box, which I think 
> would be nice.

It's not a big deal to get DOLFIN to compile with PETSc 2.2.0. I have
done it before and for a while we had an option to compile with 2.2.0
or 2.2.1. The problem is that someone using PETSc 2.2.0 has to
maintain it and keep checking DOLFIN against PETSc 2.2.0. I don't have
the time, but if someone else is willing to do the job it's no problem.

> You said that DOLFIN would not work with the current Debian package. Is 
> this just because it is out of date, or did it never work with it?
> 
> I'm tempted to try to package 2.3, but this may be harder, plus DOLFIN is 
> not actually using it yet.

I suggest you download and compile 2.2.1 and then move to 2.3.0 as
soon as we get around to updating DOLFIN.

> Packaging DOLFIN itself should be comparatively easy, since it uses a 
> standard build system.
> 
> I've heard rumors that the PETSc developers are planning to integrate 
> DOLFIN into PETSc for version 3. Do you know anything about this?

They probably won't be integrating DOLFIN as far as I know. But they
are planning to use FFC and FIAT. DOLFIN will be one front-end for
FEniCS (the C++ interface) and PETSc 3 will supply another (Python or
language-independent). This is the short-term plan, but things might
change. The good thing is that with our component-based approach, it
won't really matter which front-end you use.

/Anders