Re: [eigen] CRTP in eigen: using .derived()

[Dan Čermák <dan.cermak@xxxxxxxxxxxxxxxxxxx>]

Hi Benoit,

thanks for making that clear, I was afraid that some things might break 
otherwise.


Cheers,

Dan

On Thursday, May 05, 2016 02:47:49 PM Benoit Jacob wrote:
> Hi Dan,
> 
> Yes, the derived() in mat.derived().determinant() should not be needed in
> code like this.
> 
> There *might* be places in Eigen code itself where it's needed (inside the
> definition of a class inheriting such a method from its curious base)..
> 
> Also, the only thing that can happen, is that something fails to build if a
> method is not defined. But if it does build, then for sure it makes no
> difference. So just check if it compiles.
> 
> Cheers,
> Benoit
> 
> 2016-04-20 5:31 GMT-04:00 Dan Čermák <dan.cermak@xxxxxxxxxxxxxxxxxxx>:
> > Hi folks,
> > 
> > I have a question concerning the CRTP base classes in eigen, for instance
> > Eigen::MatrixBase. In most examples of generic functions in the
> > documentation,
> > I see something like this:
> > 
> > template <class Derived>
> > void do_smth(const Eigen::MatrixBase<Derived>& mat) {
> > 
> >         const double det = mat.derived().determinant();
> > 
> > }
> > 
> > My question is about the mat.derived(), is that necessary?
> > 
> > If I understood crtp correctly, than mat.determinant() should do exactly
> > the
> > same (provided the base class implements this function).
> > 
> > Are there advantages of using .derived() when constructing expressions?
> > 
> > I could imagine that it makes a difference whether one does e.g.:
> >         return mat.block(...);
> > 
> > or:
> >         return mat.derived().block(...);
> > 
> > as the types will be different, but I have no idea what the actual impact
> > is.
> > 
> > 
> > Thanks in advance,
> > 
> > Dan