[e-lang] EProxyResolver and sameness-breaking

[Kevin Reid]

The current protocol of EProxyResolver provides a number of ways to  
break sameness by creating multiple proxies that don't behave the  
same despite being ==. I would like to eliminate this, so that user  
comm systems (and other uses for EProxyResolver) do not have this  
unnecessary authority.

Here is an idea:

Remove the identity argument to makeEProxyResolver, and the resolve  
and smash methods from the resolver.

Instead, makeEProxyResolver has an optional second argument (run/1  
and run/2).

For run/1, the proxy is resolved.

For run/2, the proxy is a promise, and the second argument is a  
promise which will resolve to the resolution of the proxy.

Either way, the proxy's identity is the same as that of all other  
calls to makeEProxyResolver with the same arguments.

Since the handler is the same, two == proxies cannot respond  
differently to messages, and since the resolution is specified by a  
promise, two == proxies cannot resolve to different references.

Usage:

In the case of far refs, a comm system can maintain proper identities  
by using a weak-value hash table to store handlers, or by using  
Selfless handlers.

In the case of remote promises, a comm system must use a hash table  
to store the promise-and-handler pairs. This is an additional cost  
over the current system, and I haven't yet thought of a fix.


What do you think of this idea?

-- 
Kevin Reid                            <http://homepage.mac.com/kpreid/>