[Gmsh] gmsh segfaulting on simple file

Crni Gorac cgorac at gmail.com
Wed Jun 2 10:18:58 CEST 2010


On Wed, Jun 2, 2010 at 7:55 AM, David Colignon <David.Colignon at ulg.ac.be> wrote:
> Hi,
>
> it works perfectly over here (Gmsh 2.4.2, MacOSX), both with the GUI and on
> the command line.
>
>   [ ... ]
>

Thanks Eric and David for trying meshing the file on your machines.
However, it is still segfaulting on my machine, and later I realized
that same happens with number of tutorial files, for example with
t2.geo.  Don't know what could be the issue here, I'm just compiling
the latest stable version of the source code, through simple sequence
of "cmake /path/to/gmsh/source && make -j4".  The only indicator for
some kind of problem is that I have these "Error   : Singular value
decomposition requires LAPACK" messages printed out by gmsh before 1D
meshing started.  I can see this issue is already mentioned on the
mailing list, and I do have ATLAS and LAPACK properly installed on my
machine, so I'm not sure why gmsh build script is not picking them up,
or at least reporting an error if these are needed for gmsh to run...

In any case, I've built debug version of gmsh too ("cmake
-DCMAKE_BUILD_TYPE=Debug ..."), and when I run gmsh within gdb, I can
see it is segfaulting in DocRecord::Successor() method, because the
first argument ("a") of this method gets value -1 on call from
DocRecord::CountPointsOnHull() - then, the pointer
"points[a].adjacent" of course gets some crazy value, and gmsh
segfaults when trying to de-reference this pointer.  So, still seems
to me like some kind of bug in the code: for example, I don't
understand why DocRecord::CountPointsOnHull() is not detecting that
DocRecord::Successor() actually returned -1 on previous call... Any
suggestion on how to proceed to debug this?

Regards.