dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Thu, Sep 10, 2009 at 10:16:11PM +0200, Ola Skavhaug wrote:
> On Thu, Sep 10, 2009 at 10:04 PM, Anders Logg<logg@xxxxxxxxx> wrote:
> > On Thu, Sep 10, 2009 at 09:38:55PM +0200, Ola Skavhaug wrote:
> >> On Thu, Sep 10, 2009 at 9:13 PM, Anders Logg<logg@xxxxxxxxx> wrote:
> >> > On Thu, Sep 10, 2009 at 08:54:32PM +0200, Ola Skavhaug wrote:
> >> >> On Thu, Sep 10, 2009 at 5:32 PM, Anders Logg<logg@xxxxxxxxx> wrote:
> >> >> > There seem to be only a few issues remaining before we can make a
> >> >> > release:
> >> >> >
> >> >> >  https://launchpad.net/dolfin/+milestone/0.93
> >> >> >
> >> >> > In summary:
> >> >> >
> >> >> >  1. Parallel parsing of MeshFunctions
> >> >>
> >> >> What is the status on the solution to this?
> >> >>
> >> >> The right fix is to extend MeshFunction storage to contain the mapping
> >> >> from the mesh entity to vertex numbers. The quick fix (that would fix
> >> >> vertex and cell MeshFunctions) would be to store the global cell
> >> >> numbers read from file. The latter solution is simple, and could be
> >> >> done with little effort, whereas the first is a bit more involved.
> >> >> Your calls.
> >> >
> >> > I think we should make a release as soon as we can and then a new
> >> > release in about a month. I have some things planned for that release.
> >> >
> >> > My suggestion is that we aim for the correct solution and postpone
> >> > that to 0.94. Then it would be enough to make a call to
> >> > not_working_in_parallel() in a suitable place (reading MeshFunctions).
> >> >
> >> > We also need to solve (or disable) 3. plotting from C++ in
> >> > parallel. What's your opinion on that one?
> >>
> >> From a debugging point of view, having runtime plotting in parallel is
> >> nice but not needed. The most important aspect of Viper is to provide
> >> direct feedback when setting up a problem. Knowing that the scalar
> >> version of dolfin works should be sufficient in this respect.
> >
> > It works fine from Python (I think) but not from C++. We could leave
> > it as it is for now and issue a sensible error message instead.
> >
> >
>
> I agree. Lets reconsider this issue sometime before the stable release.
>
> Ola

ok, so it just remains to

1. Check parallel speedup (I don't think anyone has dared to try this yet).

2. Make all demos run or exit gracefully in parallel.

--
Anders

Attachment: signature.asc
Description: Digital signature