[e-lang] multiway when-catch

[Kevin Reid]

On Oct 29, 2007, at 22:36, Rob Withers wrote:

>> Okay, this looks reasonable, though it's more mutable than is good
>> style. However, whenResolved is not a primitive operation on
>> references. Look at jsrc/org/erights/e/elib/ref/Ref.java for the
>> implementation of whenResolved.
>
> I see that, now that you point it out.  I am sure there is a good  
> reason for
> it.  Is it because you want to limit the MirandaMethods defined, and
> __whenMoreResolved handles the behavior of what you want, so  
> whenResolved
> and whenBroken are specified in terms of __whenMoreResolved?  Or  
> are there
> other reasons?

Design principle: Operations should be as simple and un-duplicative  
as possible; having overlapping operations allows misbehaving  
implementations to implement them differently.

Also, since there are many things which must be aware of the entire  
Miranda protocol (forwarders, membranes, other-object-system  
bindings, etc.) it is good to minimize the complexity each must  
reimplement/understand.

Also, Ref.whenResolved has stronger guarantees than  
__whenMoreResolved which derive exactly from that it is a separate  
implementation. For examples, for any Ref.whenResolved(x, f),

   1. f will not be invoked until x *really is* completely resolved  
(as defined by Ref.isResolved), rather than when it is more-resolved  
or pretending to be resolved.
   2. f will be invoked at most once.
   3. f will be passed a reference which is the same as x.

There are probably other reasons I haven't thought of (MarkM?).

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