In a presentation recorded at QCon San Francisco, CORBA guru Steve Vinoski introduces REST from the perspective of a traditional SOA person. He explains the goals of the various constraints REST imposes, and the desirable properties one can gain from adhering to them. In a hypothetical discussion with a "SOA guy", Steve addresses various frequent doubts people express when they first look at REST.
Steve Vinoski is a very well-known expert on middleware, mostly known for his long-time involvement with CORBA. He is a member of technical staff at Verivue and was previously chief architect and Fellow at IONA Technologies for a decade. Over the past 15 years, Steve has authored or co-authored over 80 highly-regarded publications on distributed computing and enterprise integration.
QCon is a conference that is organized by the community, for the community.The result is a high quality conference experience where a tremendous amount of attention and investment has gone into having the best content on the most important topics presented by the leaders in our community.QCon is designed with the technical depth and enterprise focus of interest to technical team leads, architects, and project managers.
backlash against interface/contract based programming in general.
raises the issue of versioning/deprecating interfaces (or resources).
For me, I see issues with service discovery (a la WSDL) with REST, as there is no way that I can see for a consumer to learn a REST service's API without referring to human-readable documentation. With SOAP, we have the ABCs (address-binding-contract). With REST, the A and B are implicit, which leaves C.
What I'd like to see happen for REST to mature is a simple way to
describe the exchange of documents (messages), and not just serialize
objects, which is all the SOAP/WSDL/XSD mess really describes. In the
end, I guess REST is about semantics and SOAP is about syntax.
There is certainly some room for REST to "grow up" with little used HTTP 1.1 verbs such as OPTIONS, HEAD, TRACE, and CONNECT and header fields.
You said: "Specialized interfaces inhibit scalability" and inhibit reuse as they require custom code clients. I fully agree with that remark.
However does the need for specialized data not inhibit reuse and scalability just as much as specialized interfaces? Don't we spend far more lines of application code on setting up our specialized data structures than on calling the actual specialized interface of a service?
Does offering a uniform interface while still having specialized data structures really increase reusability and therefore scalability? I kind of question that but would like to get others opinions as well.
As I once remarked in a blog entry, I've never seen anyone develop an IDL/SOAP/WSDL-based client without referring to human-readable documentation. Nobody writes a client that simply goes out, discovers such services, and starts using them. Among other problems, the specialized interfaces required to communicate with the newly-discovered service makes this hard to do. Keep in mind that each specialized service interface is effectively a new application protocol.
IMO you have a better chance at this with REST due to the very important HATEOAS constraint: "hypermedia as the engine of application state." The representations returned by resources direct the client through the application state by giving it hyperlinks and form metadata so it knows what to do next. The client must understand the media type of the resource representation, of course, but the fact that media types are globally standard types registered with the IANA means that clients and servers can be independently developed against them. This is quite different from the specialized data types for each specialized interface that IDL/WSDL encourage, which ends up discouraging independent development of client and server.
This is a REST frequently-asked question, and if I recall correctly it was even asked by some of the attendees (Glen and Sanjiva IIRC) during the presentation. It's such a common question that I wrote a whole column about it earlier this year: please read "Demystifying RESTful Data Coupling" as I believe it will answer your question.