Re: ITP: aster -- Finite Element Analysis (FEA) software for engineering simulations
Il giorno dom, 29/05/2011 alle 17.34 +0200, Christophe Trophime ha
scritto:
> >
> > A little update:
> > - I've updated the copyright with the upstream approval to post the
> > sources and binaries of metis on the debian mirrors
>
> Great news so we could upload metis-edf, right?
I think so...
> > - code aster is no longer linked to metis-edf but to scotchmetis; now
> > metis-edf is only a suggested package
>
> Just a "stupid" question.
> aster is build with libmetis from scotch
>
> If we use metis-edf there is no libmetis library but a libmetis-edf library
> to avoid conflict when trying to install both metis-edf and scotchmetis.
> So I had the feeling that we have to build either with scotchmetis or metis-edf.
>
> If we choose scotchmetis, then we cannot use onmetis...
If I get it right, things are like this:
- in CA metis is not called directly, but only as a "second order"
library; e.g. is called by mumps.
- when the solver wants to use metis directly, it calls it through the
executables onmetis.exe etc.
So, we can link code aster against scotchmetis instead of metis and live
happy. Moreover, if ca were to call sone metis call not present in
scothmetis, the linker would have complained; but it did not, so I think
it's ok like this.
On the other hand, onmetis is built from the metis-edf source, and is
linked against libmetis-edf; and since CA calls onmetis like an
extrernal executable, we can link onmetis on whatever we want - in this
case, we link it to libmetis-edf.
> Do you have any idea if it will be (possible and) long to rewrite onmetis with scotch?
I have absolutely no idea... moreover I have absolutely no knowledge
neither of the scotch or metis library nor of partitioning.
> > About the last change, I made only a quick test and it seemed to work;
> > anyway, a deepest testing should be needed: in the next days I'll run
> > the test suite coming with code aster.
> >
> > The package has still some issues:
> > - the file mpirun_template, needed to complete the mpi build by as_run
> > is missing; I've opened a bug for code-aster-run[1], is someone taking
> > care of this?
>
> I could take care of this point. But I wish to wait for the new version of astk.
> Upstream has made some changes to close the open bugs on code-aster.
Uhm.. this is a missing file for as_run, not for astk... anyway I think
this is a little change in the package and is orthogonal to other
changes... can't you take care of this sooner?
> > - some sed commands in debian/rules does not seem to work; the affected
> > file is astout.export, used to run the test suite
>
> As far as I remember they were introduced by Andre to limit the testsuite
> to a given set of tests not including metis.
The problem is also that those commands are intended to change some
variables like ASTER_ROOT et al. to their actual value; however the sed
lines don't work in the makefile - but they work on the terminal... I
haven't yet had time do dig into this... suggestions anyone?
> > - the package is still not lintian clean, since there are some .pyc
> > files around
>
> I think we can safely remove the .pyc
Yes... but they should not be there in the first place, since there's a
find+rm *.pyc somewhere. I'll also look into this in the next days...
> > What now? I'll continue to work on the packages... what should be fixed
> > first to make the packages enter the distribution?
>
> I think we have first to decide what to choose with metis-edf / scotchmetis?
If I'm right, there's no need to choose; anyway, in the next week I'll
run the test suite and see the results.
Bye
Andrea
Reply to: