Taxonomy of Facets & Composites

Mark S. Miller markm@caplet.com
Mon, 31 Jul 2000 11:02:30 -0700


I wrote:
>a) at least the local naming perspective of the lambda 
>calculus, which is shared by the treatment of clist indexes in capability 
>OSes http://www.mediacity.com/~norm/CapTheory/NaP.html 

At 08:51 AM 7/31/00 , Tyler Close wrote:
>E the language is just a way of expressing compositions of the
>Granovetter operator. Is it reasonable to expect that if we can
>understand all capability systems in terms of the Granovetter operator
>that E the language could be used to script them? If so, then the
>style of the E language (lexical scoping) does not belong in a
>taxonomy of capability systems, since it could be applied to all
>capability systems. Perhaps a second taxonomy of 'Granovetter
>composers'.

By "local naming perspective of the lambda calculus", I mean something less 
than full lexical scoping, as I hoped my inclusion of clist indexes would 
indicate.  I would also claim that my Pet Name Markup Language proposal 
http://www.erights.org/elib/capability/pnml.html shares this "perspective", 
even though its particulars are very different from the first two.  What is 
this perspective?  http://www.mediacity.com/~norm/CapTheory/NaP.html isn't a 
bad start at stating it.


>I don't get b)? Has it escaped from your head yet?

No.  But in the meantime, we've got three very clear examples represented 
clearly by at least seven systems.


Sorry to be so vague, using perspectives and examples rather than formal 
models.  Such is the stuff of the conversations that eventually lead to 
formal models.  Once again, perhaps we should simply be using Hewitt's 
Actors work rather than trying to reinvent it.  Perhaps all capability 
systems are security-equivalent to Actors.  Actors have the Granovetter 
nature, the lambda naming perspective, and a local side effects semantics 
adequate to model all the others in a security-equivalent way.  CLPs are 
also a plausible choice of formal foundation, but they seem more complicated 
and less "foundational" to me.  Perhaps this is a matter of taste.

(Note: As a foundational formal model of computation, I lump Kernel-Joule in 
with Actors, though one may argue that it's yet a third model.)

Unfortunately, neither Actors nor CLPs are widely known formal models even 
on this list.  This is another reason to try informal restatements of the 
important properties of capability systems in general.


         Cheers,
         --MarkM