dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, Oct 27, 2006 at 12:23:56AM +0200, Johan Jansson wrote:
> On Thu, Oct 26, 2006 at 10:47:39PM +0200, logg@xxxxxxxxx wrote:
> > > On Thu, Oct 26, 2006 at 05:06:21PM +0200, Anders Logg wrote:
> > >> I have just two concerns:
> > >>
> > >>   1. I would like to see more people helping out with general
> > >>   development and maintenance.
> > >
> > > Yes, I think we all do.
> > >
> > >>   2. I would like changes to be propagated to the main tree in
> > >>   small pieces (if they are not independent) so I know what is
> > >>   happening and get a chance to influence the development.
> > >
> > > Do you mean in the kernel or in all modules/applications as well? I
> > > think it would be counter-productive to demand that application
> > > developers have to submit to a certain check-in schedule (piece by
> > > piece).
> > 
> > Just the kernel. I certainly won't mind any big updates to the elasticity
> > or Navier-Stokes modules.
> 
> Ok, that's good.
> 
> > 
> > > Otherwise a simple solution where I think everyone would be happy
> > > would be to have one branch where the kernel is stable which is used
> > > to develop modules/applications (since they won't sabotage eachother),
> > > and one branch where the kernel is unstable, but where very little
> > > application development is done (perhaps only a few simple test
> > > cases). The two branches could then be merged at regular intervals (at
> > > releases for example).
> > 
> > Do you mean that we maintain a separate branch that only gets bug fixes
> > between releases? This would of course be good to have, but then we need
> > someone to maintain it. Do you want to maintain such a branch? We could
> > put it in a separate repository (which could replace dolfin-kth) and be
> > named something more official, like dolfin-stable.
> > 
> 
> Yes, only bugfixes are allowed in the kernel. But the idea is also
> that modules/applications are developed against dolfin-stable (because
> that development requires a stable kernel). This is assuming that
> modules/applications are supposed to exist in DOLFIN at all, and not
> be separate. But I think we had consensus that that was a good idea,
> since we want to move away from having separate "solvers" for every
> equation.

Yes, since that will force us to develop something generic, rather than
specialized fixes for each equation.

I don't remember if I said this in our previous discussion, but I
think separate *interfaces* (like Ko) for specialized applications
could be a good idea, although in many cases one can provide such an
interface through a DOLFIN module.

> I can maintain dolfin-stable, no problem. It will result in less total
> work for me compared to not having a dolfin-stable.
> 
>   Johan

Excellent!

When you have this set up (presumably after the next release), make an
announcement to this list, on the web page (download page) and in the
manual. I'm sure this will be appreciated at Simula as well. The main
application PyCC depends on dolfin-dev and depending on dolfin-stable
would make life easier for some people here as well.

We should discuss a working practice for how/when to push or pull bug
fixes from dolfin-stable to dolfin-dev.

I'd prefer if you always pushed bug fixes directly to dolfin-dev from
dolfin-stable.

/Anders