[Shootout-list] Re: Re: [Shootout-list] Binary trees

[Simon Geard]

Jon Harrop wrote:

>The difference is that the C++ has separate structures for spheres and groups, 
>i.e. a node in the scene tree is either a "Sphere" or a "Group":
>
>struct Scene {
>  ...
>};
>struct Sphere : public Scene {
>  Vec center;
>  double radius;
>  ...
>};
>typedef list<Scene *> Scenes;
>struct Group : public Scene {
>  Vec center;
>  double radius;
>  Scenes child;
>  ...
>};
>
>In contrast, the Fortran uses the same data structure for both spheres and 
>groups, i.e. a node in the scene tree is always of type "Tree":
>
>  type Sphere
>     real*8, dimension(3) :: centre = (/ 0.0d0,0.0d0,0.0d0 /)
>     real*8               :: radius = 0.0d0
>  end type Sphere
>  type Tree
>     type(Sphere)        :: data
>     type(Tree), pointer :: children(:) => null()
>     integer             :: num_children = 0
>  end type Tree
>
>On Monday 20 June 2005 11:34, simon@whiteowl.co.uk wrote:
>  
>
>>If the current implementation really is deficient as you describe it
>>would be useful to have a specification so that it can be implemented
>>as you intend.
>>    
>>
>
>I am perfectly happy with the current implementations provided they work. The 
>deficiency would only show up if the ray tracer were extended, for example to 
>allow colored spheres, because the most of the implementations allow extra 
>data like color to be associated only with visible spheres and not with 
>bounding spheres whereas the Fortran implementation would have to carry an 
>unused color with every bounding sphere.
>
>  
>
Ok, I understand the difference now. When f2003 compilers are released
(later this year?) this type of functionality will be supported.

Simon