- From: Amelia A Lewis <alewis@tibco.com>
- Date: Thu, 01 Jul 2004 13:34:47 -0400
- To: Mark Nottingham <mark.nottingham@bea.com>
- Cc: www-ws-desc@w3.org
On Thu, 01 Jul 2004 10:12:55 -0700 Mark Nottingham <mark.nottingham@bea.com> wrote: > Because it may be desirable to model applications as state transfer, > even if this isn't advertised in a way that HTTP implementations (e.g., > caches) can take advantage of (i.e., the HTTP method). State transfer > is a higher-level and more useful abstraction than simple messaging. Umm. Perhaps a different model needs to be used, then? I don't think WSDL does, or was designed to, describe REST-oriented interactions. It describes methods, which are named, and one of our recently hot issues was specifically related to the question of identifying the method invoked. Note that I am not claiming that these are RPC methods (they may be pub/sub methods, as an extreme counterexample). Although this may cause the mob to rise up and lynch me, I also fail to be convinced that REST is applicable outside the narrow scope of HTTP. HTTP was, of course, designed (at least in part) around those concepts. Possibly Atom is, too. No other protocol, so far as I know, is, and the mappings of REST to those protocols are hugely less than convincing. Under that circumstance, REST-specific attributes/properties belong in HTTP (or HTTP-based) bindings (or perhaps someone wants to create a generic "REST" binding?). So I remain strongly opposed to setting HTTP attributes on the interface. Amy! (cleaned up cc list ... could folks please try to do this from time to time?) -- Amelia A. Lewis Senior Architect TIBCO/Extensibility, Inc. alewis@tibco.com
Received on Thursday, 1 July 2004 13:35:18 UTC