[cap-talk] kernel object knowledge - object destruction

[Jed Donnelley]

Jonathan S. Shapiro wrote:
> Jed:
>
> Did NLTSS have a mechanism that notified the object implementer (server)
> when the last capability to a given object was dropped?
>   
No.
> If not, how do servers learn when objects are no longer in use if the
> clients fail to advise the servers of this?
>   
They don't.  NLTSS had an independent accounting mechanism.  The NLTSS 
philosophy
was that the existence of capabilities to an object was not a defining 
indicator of the
value of the object.  Rather the defining indicator was whether anybody 
was willing
to 'pay' for the object.  NLTSS could both recreate a capability to an 
object that
had no outstanding capabilities (through the accounting system - if I'm 
paying for
an object, I should be able to get access to it), and - more commonly - 
destroy
objects for which there are still outstanding capabilities (I no longer 
want to pay
for this object).  At that point of course any outstanding capabilities 
became
dangling references - something that wasn't considered a problem in NLTSS,
except perhaps to clean up.

--Jed
http://www.webstart.com/jed/