Web Service Contract Versioning

This book is great and has everything in it for everybody who is planning to create XML-based services of whatever flavour. I consider it a must-read (at least the core chapters)

Sadly, the number of pages accurately reflects the apalling complexity of XML-Schema + WSDL + versioning. I thank the authors for nailing down all aspects of this terrible subject.

Personally, I buried my optimism about the ambitions of developers to get a grip on all that. Ask anyone about the "unique particle rule". Or try to explain why naive XML-Schema-based web services are a very tight coupling, so much in conflict with the brainwashing SOA propaganda that is running for years now.

I see naive webservice/schema implementations all over the place, because those are easy to find with Google (that's how we learn new technology these days!), quick to implement, and they get the job done at least for short term. I am tired of arguing against this.

I take it from the description and the comment that there isn't a section on RESTful services and how to properly version them?

No, it's about WS.

That's a real good question. How do you handle versioning in REST? Sure enough you can version "resources", but I am not sure that's what you are looking for.

You should start by defining what you mean by a "RESTful" "Service". There is no notion of service in Roy's REST, let alone versioning.

I would actually argue that since the interface to a resource is uniform, why would you even need to version that interface? That sounds like an oxymoron.

The interface is what's uniform, not the resource. Inbound and outbound representation is bound to change over time and if a service needs to be backwards compatible, like mine does, you have to consider it.

I handle REST versioning via content negotiation with a version embedded in the mime-type. I came to this solution after doing a lot of Googling. There wasn't a clear winner as far as a standard stating "thou shalt do this". Which is why I asked the question in the first place. There are probably other ways, but this one seemed most clear and maintainable to me.

Dave:

yes, the only problem with that approach is the combinatory problem of Mime Types x Versions x Who knows what else. The content negotiation was never designed to handle versioning.

Unfortunately, since REST couples access with identity, you are out of luck, you are stuck with this kludge. One day you will appreciate the ability to assign a particular version to a single endpoint, independent of any resource identity.

yes, the only problem with that approach is the combinatory problem of Mime Types x Versions x Who knows what else.

I need to add that the chapters about XML Schema design and versioning are equally useful for all kinds of interfaces, which carry XML-Schema-based content.

I ask any colleague who is about to design a schema to read those chapters. It is the completest and most concise source I know.

I like the book, however I find the subject disgusting :-)

Hello David.
Yes, that is a way I've seen in several discussions, and it sees has some drawbacks.

But Jean-Jacques is right in the "RESTful" "Service" sense of what you are looking for. A service has some particularities, that will provide enough for a long discussion on how a Service can be made RESTful.

The most common confusion is that service consumption is just an RPC call. For that part, a RESTFul service is then an RPC call that is made in a RESTful way? We have problems there since REST is not about RPC. Actually, many so called RESTFul API were discredited as such by Roy himself because they were plain RPC calls.

Now, if you go the right way, a service being a business functionality that can be consumed sending asynchronous messages to it with documents, then we hit another wall there, since REST in its HTTP implementation, uses actual operations that are synchronous.

If you solve those conceptual issues, then you may be able to apply the considerations of the book about versioning, or you may find out REST may manage that easier due to hypermedia state control.

Cheers

William Martinez

As far as I understand the -- thorough -- discussion in the given chapters, adding an optional element (minOccurs="0") is generally considered backward compatible. This surely is correct for messages sent _to_ the provider of the web service. But what if this element is part of a message (response) sent from the provider to the "client". Then an additional element might be incompatible, if the client can not handle additional elements. And many frameworks, especially those mapping the message to objects, have problems with additional content.

other thing that this book doesn't consider is the transport protocol, if i use jms the "consumer" role is different and compatibility changes too.