Re: [eigen] RFC - how to avoid (impossible) recursive declarations

[Helmut Jarausch <jarausch@xxxxxxxxxxxxxxxxxxx>]

On 09/21/2012 10:56:29 AM, Christoph Hertzberg wrote:
> On 21.09.2012 10:46, Helmut Jarausch wrote:
> > template <typename Derived, typename OtherDerived>
> > VectorXd Block_Solve(MatrixBase<Derived>& A, const
> > MatrixBase<OtherDerived>& b )
> > { int   N = A.rows(),  m = Teil_Dim, mrest= N-m;
> >    typedef typename Eigen::Block<Derived,Dynamic,Dynamic>  
> > PartialMatrix;
> >    typedef typename OtherDerived::SegmentReturnType PartialVector;
> >
> > ....
> >
> > seems to get the compiler into endless recursion since I get endless
> > error messages.
> > It probably tries recursive definitions which have to fail.
>
> That's essentially the same problem Norman Goldstein faces at the  
> moment (see the thread "Recursion and block matrices").
> The easiest solution would be to use the Ref-class from the  
> dev-branch (see Gael's mails in that thread),

Many thanks for that pointer, I'm going to look into it.

> otherwise you need to handcraft some kind of Map object.

This evokes some old, ugly memories on FORTRAN EQUIVALANCE, which is  
almost a guarantee for bugs.

> The other alternative is to avoid recursion at all. Essentially you  
> can simulate both "walking up" and "walking down" the stack by one  
> for-loop each.

Yes, thanks. Currently, I don't "need" a solution. It's some sort of  
sport. I'm trying to convert an example using my own, old, non-template
matrix library. And recursion is so much more elegant while loops are  
sometimes hard to understand.

Many thanks,
Helmut.

P.S.  I'm using the dev-branch all the time - I just didn't know about  
the Ref-class.