ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

ok, I understand. Let's create a new class Constant in DOLFIN as you
suggest, which is good since then Constant and Function work the same
way in FFC/DOLFIN: you have to declare a pair of corresponding objects
in FFC and DOLFIN.

Even if we do this, you could run into things like

init()
{
  Constant c(13.7);
  L = new LinearForm(c)
}

which would give you the same problem.

What if we overload a constructor LinearForm(real c) in addition to
LinearForm(const real& c)? Will that work?

/Anders

On Sun, Sep 11, 2005 at 03:11:09PM +0200, Johan Jansson wrote:
> On Sun, Sep 11, 2005 at 07:49:35AM -0500, Anders Logg wrote:
> > The reason it's a reference is that in a time-stepping loop, you want
> > to be able to give the time step k or time t as an argument to the
> > form and have it automatically get updated as you change k and t.
> > 
> > Maybe one should require that one uses references for the constants,
> > that is, the constant must exist in a variable outside of the form, in
> > the same way as a Function must exist outside of the form?
> > 
> > /Anders
> > 
> 
> Ah right, I'll change it back.
> 
> I think the only reason it's a problem is because basic types are
> usually passed as copies, not references, and it's very common to
> construct them on the stack. Maybe it's enough to just make a class
> Constant? Then people would treat it just as a Function, and it could
> also be generalized to being tensor-valued etc.
> 
> I don't think we can keep it this way. As soon as you construct a form
> in a separate function from where you use it, you will run into this
> issue. Consider:
> 
> LinearForm* L;
> 
> init()
> {
>   real c = 13.7;
>   L = new LinearForm(c)
> }
> 
> solve()
> {
>   L.assemble(); // Here L->c0 is undefined
> }
> 
> If it wasn't for valgrind telling me exactly what the problem was, I
> would have spent a long time scratching my head on this.
> 
>   Johan
> 
> _______________________________________________
> FFC-dev mailing list
> FFC-dev@xxxxxxxxxx
> http://www.fenics.org/cgi-bin/mailman/listinfo/ffc-dev
> 

-- 
Anders Logg
Research Assistant Professor
Toyota Technological Institute at Chicago
http://www.tti-c.org/logg/