dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Dec 08, 2006 at 10:43:21AM +0100, Johan Jansson wrote:
> On Thu, Dec 07, 2006 at 10:34:16PM +0100, Anders Logg wrote:
> 
> ...
> 
> > Agree. I think dolfin-stable should be stable or we should call it
> > something else. It does not seem natural to have a branch named
> > 
> >     dolfin-kernel-stable-and-modules-unstable
> > 
> > We really need something like
> > 
> >     dolfin (dolfin-dev)
> >     dolfin-stable
> >     dolfin-modules (dolfin-modules-dev)
> >     dolfin-modules-stable
> > 
> > This would require us to move the modules to a separate tree.
> > 
> > If we can't do that (for lack of maintainers), I think we should keep
> > dolfin-stable stable and make big changes only in dolfin.
> 
> I thought me and Johan H made it quite clear though that we wanted a
> system where the kernel and module development is separated, so that
> the kernel doesn't suddenly break (or subtly changes) between two
> checkins. The system we have today is such a system. The system where
> all changes happen in the same repository is not such a system.

But if this is the purpose, then the repository should not be called
dolfin-stable since the name is misleading.

Another option is that you have a separate branch dolfin-kth (on a KTH
server) that you can organize in any way that works for you.

> Since the kernel interface changes all the time, I think it's natural
> to keep the kernel and modules in the same tree.

The problem with this is that it makes me and Garth responsible for
updating the Navier-Stokes module when we work on the kernel.

> If the trees would be separate it would be a nightmare to sync the
> modules tree with the kernel tree, since the kernel development
> would be totally disconnected from the modules, i.e. then you cannot
> create a changeset which spans both trees. Once the kernel interface
> stabilizes, we can consider splitting the trees.

Why would this be a nightmare? The only difference would be that Hoffman
would be responsible for keeping the Navier-Stokes module in sync with
the kernel and Jansson would be responsible for keeping the elasticity
module in sync. Someone must do it and it seems more reasonable to put
this responsibility on the module developers than on the kernel
developers.

> If the trees would have been separate the change would have been
> exactly the same, it would have happened in the modules tree, and not
> in the kernel tree. I'm not sure what the big resistance is here, if
> it's just a matter of having to keep compiling the modules until the
> next merge, then that should be a non-issue, or?

We are discussing a number of different issues at the same time
here: separate module tree or not and the role of dolfin-stable.

> > Yes, it requires a specialized solver. The specialized solver can be
> > very short and it can be implemented using the TimeDependentPDE
> > interface.
> 
> Wherein do you mean the specialization lies? Do you mean that you need
> to define coefficients which consist of h and norms?

Specialized means that the solver defines the equation and maps a
string "poisson" or "convection-diffusion" to an equation.

> If these are the only outstanding issues, perhaps it would be
> worthwile to solve them so we can leave this behind us.

Agree.

/Anders