dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Fri, May 09, 2008 at 12:36:42AM +0200, Martin Sandve Alnæs wrote:
> 2008/5/9 Andy Ray Terrel <aterrel@xxxxxxxxxxxx>:
> > I hit a bug where my mesh generation script was giving me a zero volume
> > cell.  I believe in an older version of dolfin this threw an error in the
> > assembly process. For my code it just output nan's in the matrix and then
> > the solver told me the matrix was singular.  I was wondering how hard would
> > it be to throw the error in assembly, the current flow makes things
> > difficult to debug.
> > -- Andy
> 
> An alternative for this particular problem could be assembling the vector
>   e = FiniteElement("DG", polygon, 0)
>   v = TestFunction(e)
>   a = v*dx
> which should yield 0 only for cells with zero volume. Right?
> 
> But in general, I think DOLFIN would really benefit from more extensive
> error checking a lot of places. Relying on pretty syntax is not enough,
> providing good error messages is much more important for usability.

I agree. Assembler::check() does a few checks already but feel free to
add more.

-- 
Anders