dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Tue, Dec 05, 2006 at 04:11:40PM +0100, Johan Hoffman wrote:
> > Johan Hoffman wrote:
> >>> I think so.
> >>>
> >>> I don't think a special --disable-modules will be necessary.
> >>> But we will need the following targets:
> >>>
> >>>     make modules
> >>>     make modules_install
> >>>
> >>> Then if one likes, one may enter into src/modules/demo and do
> >>>
> >>>     make
> >>
> >> Sounds good.
> >>
> >
> > Sounds good to me too.
> >
> >>> One problem I have is that just moving demo, bench, test for modules
> >>> into src/modules makes a very flat structure, with navierstokes at the
> >>> same level as the demo for navierstokes. Perhaps we could have
> >>>
> >>>     src/modules/
> >>>     src/modules/solvers
> >>>     src/modules/solvers/navierstokes
> >>>     src/modules/solvers/elasticity
> >>>     src/modules/solvers/plasticity
> >>>     src/modules/demo
> >>>     src/modules/bench
> >>>     src/modules/test
> >>>
> >>> This could open up for creating other kinds of modules that are not
> >>> necessarily solvers.
> >>
> >> The modules seem not that specified right now, and naming a module after
> >> a
> >> certain equation may not be that good, for example since a typical
> >> module
> >> may be multi-physics based, or several methods may be used for solving
> >> the
> >> same equation. Also, I do not think the /modules/solvers level is
> >> needed,
> >> it should be enough with /modules. And I'm not sure that we need to have
> >> a
> >> separate level for classifying the type of modules either (flow, solid,
> >> etc.)
> >>
> >> I think we should have src/modules, where we today can put modules that
> >> are maintained (navierstokes, elasticity,...), in a flat structure, and
> >> the naming could be quite free: "fluid-structure-interaction-solver",
> >> "fsi-flash", "heart", or whatever reasonable (where maybe not *-falsh
> >> would qualify...).
> >>
> >
> > I would like to introduce some structure just to keeps things looking
> > tidy. I don't mind categorising modules (flow, solid, etc). Rearranging
> > is only a question of "mv". To me it's just important that the modules
> > remain bundled together.
> 
> I just worry that there may be as many categories as modules. Maybe we
> could add this categories layer later on when structures emerge, today we
> do not have that many modules.

The question is: should we say "this is the DOLFIN Navier-Stokes
solver module" or should we say "here's a module that solves
Navier-Stokes and here's another one".

If we want to provide just one module for each equation, then we
should try to enforce some structure.

/Anders