GetFem++ 4.1.1

[logari81]

On Tue, 2010-11-16 at 09:33 +0100, Konstantinos Poulios wrote:
> On Mon, Nov 15, 2010 at 7:23 PM, Sylvestre Ledru <sylvestre at debian.org> wrote:
> > Le lundi 15 novembre 2010 à 19:16 +0100, logari81 a écrit :
> >> On Thu, 2010-11-04 at 15:33 +0100, Sylvestre Ledru wrote:
> >> > Le vendredi 29 octobre 2010 à 14:47 +0200, logari81 a écrit :
> >> > > Hi Sylvestre,
> >> > >
> >> > > I have uploaded some changes for the new version 4.1.1 of GetFem++.
> >> > > Maybe this is a good time to add the Scilab support as well. So you can
> >> > > enable it before committing my changes.
> >> > I had a look and despite some minor problems, the Scilab support seems to be OK.
> >> > Thanks for updating.
> >> >
> >> > > I had to add libatlas-base-dev to the build dependencies, otherwise the
> >> > > current liblapack3gf wouldn't link.
> >> > Hmm, that is unexpected. Can you give me the build log and the error  ?
> >> >
> >>
> >> the configure script fails to run the test for the smumps_c function in
> >> MUMPS. The linking of the test case needs libatlas-base-dev otherwise
> >> fails with:
> >>
> >>
> >> configure:20005: checking for library containing smumps_c
> >> configure:20044: g++ -o conftest -g -O2 -O3 -Wall -W -fmessage-length=0
> >> -ftemplate-depth-40 -pedantic -Wshadow -Wpointer-arith -Wcast-qual
> >> -Wwrite-strings -Wconversio$
> >> /tmp/ccO84gIo.o: In function `main':
> >> /tmp/buildd/getfem++-4.1.1/conftest.cpp:55: undefined reference to
> >> `smumps_c'
> > Sorry but I don't see how it is related to atlas. In atlas source tree:
> >
> > $ grep -r smumps *|wc -l
> > 0
> >
> > I think a library in missing in the check (probably -lmumps)
> 
> no, I think this is not the case
> 
> >
> > Sylvestre
> >
> 
> hmm, the fact is that just adding libatlas-base-dev to the build deps
> list makes the configure script work. The explanation is a bit more
> complicated and I had to drop in a shell in the pbuilder in order to
> understand it. As far as I remember I tried to compile myself in the
> pbuilder shell the same conftest which fails in the configure script
> in order to realize that the problem was related to the linking of
> liblapack3gf. Anyway I will try later to create a minimal test case
> free of any mumps stuff demonstrating the problem.
> 
> BR
> 
> Kostas

Hi Sylvestre,

I think I have finally fixed the problem more correctly. The configure
checks that failed before where actually redundant so I removed them and
I am going to push this change upstream as well.

If you are interested in the problem which occurred during this tests,
try the following:
g++ helloworld.cpp -lsmumps -lblas
this will fail if you don't have libatlas-base-dev with errors in the
form:
/usr/lib/atlas/liblapack.so.3gf: undefined reference to `ATL_*'
but this is probably normal because the parallel version of mumps uses
atlas.

Best regards,

Kostas