[Insight-developers] Removing SpatialOrientation stuff from itk::ImageBase

[Kent Williams]

I knew that partial specialization isn't really feasible, but it would 
have been optimal.

I'm not clear about how making the Orientation a trait of the class 
would work. The problem with templating just over image dimension, is 
that what we want here is a conversion from Direction Cosines to a data 
type, but with the same base function signature:

template <class T, unsigned int Dimension = 3>
class OrientationAdapter
{
public:
    T FromDirectionCosines(const typename 
ImageBase<Dimension>::DirectionType &dir);
    ImageBase<Dimension>::DirectionType ToDirectionCosines 
ToDirectionCosines(const T &Orienatation);
};

Perhaps a better solution in this context would be to use inheritance 
from a template base to approximate the 'ideal' template solution:

template <class T, unsigned int Dimension = 3>
class OrientationAdapterBase
{
public:
    virtual T FromDirectionCosines(const typename 
ImageBase<Dimension>::DirectionType &dir) = 0
    virtual ImageBase<Dimension>::DirectionType ToDirectionCosines 
ToDirectionCosines(const T &Orienatation) = 0
};

template <int Dimension>
class SpatialOrientationAdapter : public 
OrientationAdapterBase<SpatialOrientation::ValidCoordinateOrientationFlags, 
Dimension>
{
public:
    typedef typename SpatialOrientation::ValidCoordinateOrientationFlags 
OrientationType;
    typedef typename ImageBase<Dimension>::DirectionType DirectionType;
    virtual OrientationType FromDirectionCosines(DirectionType &dir);
    virtual DirectionType ToDirectionCosines(const OrientationType &Or);
}

Luis Ibanez wrote:

>
> Hi Kent,
>
> As long as we support VC++ 6.0 we cannot do any partial specialization.
>
> This fact is unfortunate, and it has prevented us from doing a lot of
> cool MetaProgramming... so we look forward to the day when we can drop
> the support for Visual C++ 6.0.
>
> In the meantime,... looking at your code, it seems that we don't
> really need to template the class OrientationAdapter over:
>
>             1) OrientationRepresentation
>             2) ImageDimension
>
> Instead, I would suggest to only template it over ImageDimension
> and to make OrientationRepresentation to be a trait of the
> specialization of OrientationAdapter.
>
> The drawback of this approach is that it will only allow one
> OrientationRepresentation per image dimension.
>
> From the developers point of view this may not be too bad,
> since in their applications they could create a variation of
> the OrientationAdapter, that has in its traits a different
> definition of the OrientationRepresentation.
>
> I would guess that in the typical uses of the orientation
> representation we don't need to switch from one representation
> to another.
>
> Do you see other drawbacks in moving the OrientationRepresentaiton
> from being a template argument to being an internal trait ?
>
>
>
>    My 2 cents,
>
>
>      Luis
>
>
>
> ------------------------
> Kent Williams wrote:
>
>> My proposal for removing SpatialOrientation stuff from itk::ImageBase 
>> (scroll to the bottom):
>>
>> http://www.itk.org/Wiki/Proposals:Orientation#Cosine_Direction_Vectors.2C_and_itk::SpatialOrientation_.28Kent_Williams.29 
>>
>>
>> In summary, the conversion betwee SpatialOrientation and Direction 
>> Cosines is moved to a template 'helper' class.  I'm open to 
>> discussions/suggestions on whether this is the way to go -- the 
>> drawback to the scheme presented is that it requires partial template 
>> specialization to handle general N-dimensional orientation 
>> representations elegantly.
>>
>> _______________________________________________
>> Insight-developers mailing list
>> Insight-developers at itk.org
>> http://www.itk.org/mailman/listinfo/insight-developers
>>
>>
>