Re: network endpoints

On Apr 28, 2008, at 1:51 AM, Booth, David (HP Software - Boston) wrote:
> I'm just starting to catch up on this discussion after being away.
>
> I do *not* think we should introduce the term "endpoint" or  
> "network" into this discussion on the definition of  
> awww:InformationResource at this point, because I think we are  
> discussing the concepts at a level of abstrations that would be  
> inappropriate for those terms.  The definition of  
> awww:InformationResource (IR) that I proposed that is relevant to  
> the current level of discussion was:
> http://dbooth.org/2006/identity/#propdefir
>
>         "a function from time and requests to representations"

The definition of yours that *I* was referring to was this one:

	"a URI-named network source of representations" [1]

So you are proposing to add another class of entities to the story.  
That's fine and interesting.

My model for the way a discussion like this one should be conducted  
is the book Counterexamples in Topology. Rather than redefine the  
same term repeatedly as we learn more, and argue over which  
definition is the "best" or "correct" one, mathematicians generally  
give different names to each object (thing, class, space, type,  
etc.), and then try to prove theorems that relate them to one  
another. You can start out with A and B defined in wildly different  
ways, and then discover after study that they are the same, or at  
least symmetrically related (e.g. in 1-1 correspondence). So the  
process is: make as many names and definitions as you need to, and  
then through exploration learn about relationships between them -  
equivalence, subsumption, symmetry, quotients, approximations, and so  
on. Useless definitions fall by the wayside.

So I do not consider this a discussion on the definition of  
awww:InformationResource. It may be unclear but we are not at liberty  
to redefine it - it's published. I consider what we're doing to be  
(in part) an attempt to articulate the intended semantics and  
recommended practice around 200 responses. Whether the intent, or the  
right thing, coincides with what AWWW says, is a separate question.

Your new definition has no reference to any physical apparatus  
(network/web), and therefore is close to Tim's and AWWW's, by my  
reading, and that's fine. We can place it in our landscape of classes  
and figure out how it relates to awww-IRs. Your older definition, the  
one that mentions "network", is closer to Stuart's notion of "the  
web" and, as I read it, is closely tied to a physical apparatus. To  
me a big question is how these two very different kinds of things  
relate to one another. Having specified a particular function-from- 
time-and-requests-to-representations X, does some particular uri- 
named-network-source-of-representations Y emit awww-representations  
of the awww-information-resource that is (or is pretty much the same  
as) X under the right circumstances?  This could be tautological if X  
were defined by construction from Y (that is: an information resource  
is whatever the web tells you it is, and the web cannot be wrong),  
but in general it might be a question of implementation (Y)  
correctness relative to some specification (X). A similar question:  
Is "the web" handling requests at a given URI in a way that is  
compatible with some desirable principle, such as information  
accessibility, bookmark support, or awww-representation consistency  
across variants or through time? Can we characterize the difference  
between desirable and undesirable behaviors? Maybe behaves-admirably  
is yet another class that might be proved to relate to one of the  
others.

The consequential outcome of any analysis of IRs would be an answer  
to the question of when any particular 200 is OK (taking into account  
"web architecture", past behavior, and anything else known about the  
IR) and when it's not. As I said, if we assess the situation and  
decide that answering this is too difficult a job for this group, or  
is not appropriate, we don't have to take it on.

> An IR is thus a function f with two parameters:
>
>   f: Time x Request --> Representation
>
> The Request parameter includes things like language and mime  
> preferences, request URI(!) and cookies.  On the other hand, the  
> function Roy describes, which I'll call fRoy, and the model  
> diagrammed at
> http://sw.neurocommons.org/2008/inforesource.pdf
> have only one parameter: time.  But note also that Roy's function  
> does not go directly to representations, it goes to representation  
> *sets*, so in essence I think Roy has pushed the Request input into  
> the function result.  I.e., Roy's function fRoy can be viewed as a  
> "curried"
> http://en.wikipedia.org/wiki/Currying
> form of my function f:
>
>   fRoy: Time --> (Request --> Representation)
>
> The main problem with the diagram at
> http://sw.neurocommons.org/2008/inforesource.pdf
> is that any "dereference" function needs *both* Time and Request  
> parameters, because the resulting Representation may depend on both.

I think there are other ways one might consider F&T resources as  
abstracting over request logic. You can say that they are  
existentially quantified, i.e. that a value is a member of the value  
set iff there exists a request that elicits that value, or you can  
say that a request would serve to select a value from the value set  
in a deterministic way (CN) that is the same for all resources. In  
any case, the definition is what it is, and it clearly doesn't do  
anything very particular with requests or permit for differentiation  
between resources based on what they do in detail with requests. One  
may usefully define other kinds of entities that do, and look to see  
how they relate.

You are right that eventually we need to talk about requests, but I  
think it's too early for that and we can introduce them when we need  
to. Let's suppose them to be existentially or universally quantified  
for now, as appropriate.

> Also, I don't see the need for ValueCloud.

There is no need. I just wanted to refer to Fielding and Taylor since  
it was mentioned (by Stuart, I think) and because the paper makes  
more sense to me than AWWW. (Yes I know it has a different purpose  
and says different things.) It provides a published example of a  
formal definition of (information) resource, and therefore a  
potential foothold in perilous territory. It may end up not being a  
useful definition, but I see little harm in putting it in the diagram  
for the time being, unless you think that I'm imposing on your time  
by doing so.

Check the newer diagram [2] and tell me if you like it any better.  
The word "dereference" has gone away.

Jonathan

[1] http://dbooth.org/2006/identity/slides.html#[2]       -- n.b. the  
bracketed part is part of the URI
[2] http://sw.neurocommons.org/2008/inforesource2.pdf

Received on Monday, 28 April 2008 14:00:45 UTC