Re: [boost] Re: cyclic ptr (was: Re: Re: Re: Re: Re: Re: Re: Proposals: p2m offsets, prebuild<>, squad_ptr<> (of course))

From: Larry Evans (jcampbell3_at_[hidden])
Date: 2002-09-07 10:56:15

Greg Colvin wrote:

>At 08:45 AM 9/7/2002, Larry Evans wrote:
>  
>
>Tell me more about how you are using Lin's method.
>
>
>  
>
Pretty much the same way I used it in my earlier
GarbageCollectionByTemplateSpecialization folder in files.
The main difference is that the internal pointers were pushed
onto a stack and then processed by the various phases in
Lin's method.  Comparison with your cyclic_ptr showed this
was too slow, so I tried speeding it up.  I did this using iterators.
The trick making an static member template where the template
argument was the iterator type.  That way, it could handle stl
containers of shared_ptrs as well as individual shared_ptr.  However,
vector, or any other stl container, cannot be used without modification.
This modification is implemented in the scoped_cyclic_container.  For
the ip_offset_iterator internal pointer method, this is specialized as:

template
  < template<typename>class Container
  , template<typename>class Element
    , gc_methods GcMethod
  >
  class
scoped_cyclic_container
  < Container
  , Element
  , cyclic_count_gc<GcMethod,ip_offset_iterator>
  >
  : public obj_id
  , public 
Container<shared_cyclic_ptr<Element,cyclic_count_gc<GcMethod,ip_offset_iterator> 
 > >
  , private detail::mk_proxiters_descriptor_static::proxiter_record
  {...};

Container can be a stl container template.  A similar specialization is 
needed, I think,
for ip_assign_op_switch because it needs to assign to each element of 
the container
in order to work.  At least that was my experience.

Anyway, the shared_ptr and shared_count also were modified to allow 
this.  I could
post these if you want, or I could give further explanation.  I realize 
this is kinda sketchy,
but let me know and I try and be clearer.