I argued with the author that what he was showing wasn’t a RESTful interface, and at the beginning he was surprised by the importance I gave to hypermedia:
I have never seen the point of jamming links into the response, that is what documentation is for.
[…] As for HATEOAS? I don’t see the point. I’d be interested to see some examples of it being useful […]
that leaded to my response:
To just make an example… without using hypermedia relations you need your client to make assumptions about resource location ( such information can be retrieved on your documentation, for instance ).
When you change the implementation ( for example, URI schemas ), your client is broken: driving application’s state by hypermedia means that clients are loosly coupled with your service and consume it without any assumption.
My point was mainly against URI schemas as far as we are intending them nowadays: the resource which the consumer is relating to during the service’s consumption.
Although URI schemas are generally awesome, because it’s easy to start consuming an application implementing a widely-known point of access, they’ve gained too much power in our interfaces; and like singletons ( and statics ), we become slaves of this power.
We don’t want coupling inside our code, so why would we like to have tight coupling inside our architecture?