[
Lists Home |
Date Index |
Thread Index
]
At 23:23 11/02/2002 +0000, Alaric Snell wrote:
>Thinking on different scales. An object-oriented distributed hypermedia
>system would have getName() et al depending on the precise class of object
>involved - and a standard interface, getUserView(), that returns HTML or
>whatever.
>
>With the Web, we got the HTML-retrieval part and now we have to do all this
>XML mess to get back to getName() and friends (computer-accessible
>interfaces...)
Not so - this is, I suggest "RPC think". Don't think RPC - think
"conversation". A conversation
is a collection of discrete "messages"[1], exchanged over time according to
some
agreed protocol. There is a big difference between "here is a purchase order"
and "SendPurchaseOrder()".
The former is a message (not a method call), classified in some taxonomy. They
can be sent to and received from URIs with HTTP.
Armed with these, I can build a distributed system on the Web. I do not
need detailed
and fixed knowledge of an object model with associated method signatures to
build a
distributed system.
The beauty of the URI-centric approach is that it allows a greater level of
de-coupling
between the various actors in a distributed system. I do not need to (and
the failure
of Corba shows it is a bad idea to) inter-twingle systems via shared object
models.
Sean
[1] The world of distributed computing conversations is not aided by the
fact that the
word "message" means different things to different people. In particular the OO
interpretation as "method invocation" carries layers of semantic baggage not
shared by other uses of the term in distributed computing.
|