Drop “g-boxes”, talk about “stateful resources”

This is a proposal to drop the term “g-box” from our graph vocabulary, along with all the other terms it has inspired such as “layer” or “space”, and use the term “stateful resource” instead.


== The container metaphor is unhelpful ==

The terms “g-box”, “space”, “layer” and so on all follow a “container metaphor”. This seems like an obvious one to use. After all, what are RDF stores if not containers of triples? And isn't an RDF file just a container of triples? But my conclusion is that thinking about this in terms of containers is a mistake and is taking us down a dead-end. It doesn't match current (and conforming) SPARQL use. It doesn't work well for many use cases. It is disconnected from the terminology of REST. And it is plainly not how the web works.


== Stateful resources ==

So let's forget about g-boxes and let's talk about *stateful resources* instead.

Stateful resources are resources that have an associated state, and the state can be expressed as an RDF graph. We can accept the intuition that the state of a resource may change over time, and that it only has one state at any given time. A stateful resource doesn't necessarily ever have to change—it can be immutable. Since anyone can say anything about anything, there is nothing wrong with a resource whose state is a graph that contains a bunch of nonsense—that's an application problem, and we are defining infrastructure.

This is exactly what the word “resource” means outside of RDF (in REST, short for REpresentational *State* Transfer).


== What kinds of resources can have state? ==

I will purposefully avoid the question what sorts of things can have state, and what exactly may or may not be a reasonable state for particular kinds of resources. We can accept whatever answer works for REST.

But it is certainly the case that, if we dereference an IRI i and get back a 200 status code along with a representation that encodes an RDF graph G, then it would be reasonable to conclude that G is the state of (the resource denoted by) i.

Although for some use cases, some users would likely want to use other, non-standard, “state functions”, and that's ok as long as they're in their own closed environment.


== Extending the terminology ==

This naturally leads to a complete set of terminology (if we find that we need it).

*If* we want to be explicit about the denotations of the IRIs in <IRI,graph> pairs in RDF datasets, then let's say they denote *stateful resources*.

*If* we want to define a class for that, let's call it *rdf:StatefulResource*.

*If* we want to define a name for the <IRI,graph> pairs other than “named graph”, then let's call them *state pairs*.

*If* we want to give a name to the relationship that holds between the IRIs and graphs in these pairs, let's call it the *state relationship* or *state function*.

*If* we want to define this relationship as an RDF property, let's call it *rdf:state*.

We *could* define RDF dataset as “a default graph and zero or more state pairs, max. one per IRI.”


== How this is better ==

* It matches REST's notions of Resource and State.

* Weird uses of SPARQL named graphs can be explained simply as “using a non-standard state function”.

* It works even if you don't believe in httpRange-14; in this case, a person for example becomes a “stateful resource” that may have an RDF graph describing her associated via rdf:state. That sounds much less jarring than claiming that a person is a “g-box” or a “container of triples” or a “layer” or “space”.

* It works even with real-world web scenarios where servers return different content to different clients, e.g., depending on authentication or on Accept-Language content negotiation; we just have a state function that is different depending on who the client is.

* Other specifications can easily put conformance constraints on the state function: “In an XYZ-dataset, the state function that associates graphs with IRIs MUST be: derefencing with an Accept header asking for RDF/XML, and parsing any 200-result as RDF/XML.”

* It avoids the mistake that the TAG made when they defined “information resource” by appealing to the “nature” of the resource.

Received on Tuesday, 22 May 2012 11:54:18 UTC