Re: [eigen] Do we need geometry refactoring?

[Benoit Jacob <jacob.benoit.1@xxxxxxxxx>]

> On Mon, Aug 16, 2010 at 5:14 PM, Benoit Jacob <jacob.benoit.1@xxxxxxxxx> wrote:
>> ...except that the tests are still failing at runtime,
>> geo_transformations_1:
>> /home/bjacob/eigen/Eigen/src/Geometry/Transform.h:246:
>> Eigen::Transform<Scalar, Dim, Mode>::Transform(const
>> Eigen::Transform<_Scalar, Dim, OtherMode>&) [with int OtherMode = 32,
>> _Scalar = double, int _Dim = 3, int _Mode = 2]: Assertion
>> `OtherMode!=Projective && "You cannot directly assign a projective
>> transform to an affine one."' failed.
>
> Ok, I can't finish this today, since I have to leave.
>
> One of the problems seems to come from the definition of
> ei_transform_transform_product_impl in line 1299 (approx.) in
> Transform.h. The template looks like this
>
>  template<typename Scalar, int Dim, int LhsMode, int RhsMode>
>  struct ei_transform_transform_product_impl { ... }
>
> and the ResultType is solely depending on LhsMode where I assume that
> this is wrong. I did not yet quite look into all details such as
> ei_transform_right_product_impl but I think we need something like
> this
>
>  template<int ModeLhs, int ModeRhs> struct
> ei_transform_product_result<TMode,TMode> { enum { Mode = ... }; };
>
> I am wondering whether there is a simpler way but to specialize for
> all combinations. Maybe someone of you has immediately an idea how to
> simplify this. Benoit wanted to prevent relying on the Mode, otherwise
> I would have suggested
>
>  Mode = ModeLhs > ModeRhs ? ModeLhs : ModeRhs;
>
> but that is not really nice.

Yep, so, indeed I don't think it's a good idea to rely 
on the particular numerical values of the modes.

But it doesn't 
have to make a big difference, you can always write and use:

template<
  int LhsMode,
  int RhsMode>
struct 
ei_transform_product_result_mode
{
  int ret = (LhsMode == 
Projective || RhsMode == Projective) ? int(Projective)
          : (LhsMode
 == Affine || RhsMode == Affine) ? int(Affine)
         
 : int(AffineCompact)
};


Benoit