Re: toward HTTP traces for SPARQL (fromUnionQuery, CONSTRUCT)

On Thu, May 05, 2005 at 10:32:21PM -0500, Dan Connolly wrote:

> > (Hence my skepticism when anyone says "the names don't matter"...!)
> 
> Yes, they're clearly arbitrary, and yet agreement is
> critical to interoperability, and syntax engineering is
> a big part of what makes the web go around, apparently.

(I don't agree that the name is "arbitrary"; it's a matter of
convention, but that's not the same thing as arbitrariness, that is,
something that is (or should be) the result of chance, whim, or
caprice, lacking any substantive, reasonable, or rational
considerations. In this case, we obviously *do* have such
considerations: (1) not too long; (2) not too short; (3) not
misleading; (4) informative; (5) congruence...maybe others.)

> Technically, one character names would suffice and would
> save bytes on the wire. But short words seem reasonable.
> multiple-word-names seem like overkill.

Well, you say "overkill" and I say "unambiguous" -- honestly, I have
no idea what "data" means... And it's fine to name the parameter
"data", but as you pointed out, there's no named graph parameter to go
along with "data" yet; and once there is, "data" may not seem as good.

Anyway, as I pointed out, I'm pretty sure that in the actual bindings
in WSDL, the parameters can get new names; as things stand right now,
"default-graph-uri" and "named-graph-uri" are names of abstract
things. I believe we have a choice as to whether they also end up
being names of concrete binding parameters in HTTP.

> That seems like more complexity than I'm interested in.

Yes, I'm not proposing anything like that.

My real point is that I reach for my gun every time anyone says "the
names don't matter", since they obviously do. 

Kendall Clark

Received on Friday, 6 May 2005 12:59:09 UTC