Re: [Gnash-dev] ACT module; white box testing and aspect templates

[Eric Hughes]

Here's the explanatory reply.

At 08:17 AM 7/18/2007, strk wrote:
> I don't get the template<class,class> syntax on line 122.

It's complicated.  The prior sentence may be an understatement.

A basic problem with getting an aspect class defined is the following tension:
- The owner class may not know the definition of the aspect at the point of 
definition, because the aspect must be able to change.
- An owner class must have a full aspect class definition at its point of 
instantiation, because an instance of the aspect class is a member of the 
owner class.
- The aspect class must have a full owner class definition at its point of 
instantiation, because it must be able to refer to the innards of the owner 
class.

This all boils down to a definition cycle of length two.  (Self-reference 
is a cycle of length one.)    There's an issue of infinite regress in 
definition that must be addressed.  The solution, in this case, is to make 
the aspect parameter a template, where the final argument of said template 
must be instantiated with the owner class itself.  Observe the 
specialization definition (around line 167 in my current copy):
        template< class T >
        class Null_Aspect_1< T, ACT::Handle_Registry_Leader< T, 
Null_Aspect_1 > >
This class defines the null aspect for class Handle_Registry_Leader (which, 
without an aspect, would be a template class with parameter T).  The second 
parameter is a class, which class is the owner, instantiated with its 
template arguments, which include the null aspect.

Here's the magic.  The use of Null_Aspect_1 in the specialization 
definition refers to the specialized class itself, not the non-specialized 
definition (which is in Aspect.hpp).  Such a specialization creates a 
reference cycle of length two, just as required.

Eric