[cap-talk] Is "Authority" Subjective?

[Toby Murray]

On Sat, 2007-06-23 at 15:06 -0400, Jonathan S. Shapiro wrote:
> On Sat, 2007-06-23 at 13:03 +0200, Pierre THIERRY wrote:
> > Scribit Marc Stiegler dies 22/06/2007 hora 11:33:
> > > We generally consider the ability to invoke the object that creates a
> > > new List object to be non-authority-bearing.
> > 
> > Wouldn't it be more consistent if we consider it bearing some authority
> > that we want each subject to have? Moreover, this is not authority to
> > modify some state, it is merely authority to computation, as you say.
> 
> >From a formal perspective, "create" operations are a bit of a mess. They
> often lead to models that do not converge. A fairly common trick for
> dealing with this is to have a model in which there is a pre-existing
> finite pool of objects of each type. The "create" operation actually
> changes the state of an existing object from "unallocated" to "in-use"
> and the "destroy" operation changes the state of the object from
> "in-use" to "destroyed".
> 
> While you can choose to model the allocator authority explicitly (and
> this is perfectly sensible), create and destroy operations are
> intrinsically a bit wierd.
> 
The other approach, that was taken for example by Fred Spiessens in his
work, is to have a finite set of objects, say O. each o in O represents
one or more actual subjects in your model. This allows a set of actual
subjects to be aggregated together and represented by a single object in
your model. A child subject can then be represented in the model by the
same object that represents the parent who crated it, for example.

This allows the possibility of creating infinite children while keeping
the model tractable. It works because we often assume that all children
have the same behaviour, for example.