Re: superimposing the Fielding and TBL architectures

Pat,

I'd like to know more about your (bad?) reaction to the "generic
resource" theory ( http://www.w3.org/2001/tag/awwsw/ir/20110625/ ). It
sounds like you consider it flawed, wrong, distasteful, or all three.
I would like to know why, and consider whether there is an
alternative. I certainly don't like it, but I haven't come up with
anything better. Discarding it means, I believe, abandoning
httpRange-14 and "information resource" altogether, which is fine by
me but would be a big step.

What we're trying to explain is a URI which when used for retrieval
yields content that's the same every time in all important ways, but
different each time in unimportant ways. For example, the content
could be XML, and the differences might be insignificant whitespace
(e.g. in between attribute/value pairs). We want the URI to name the
document modulo the unimportant differences. (A quotient
construction.)

But what constitutes "important" differs from one URI to the next.
Maybe for a different application (such as one that depends on the
SHA1 hash) the whitespace *would* be significant.

Under this theory there is a class of "generic resources" and a
partial order ≤ among them that might be read "specializes". There are
least elements, which I'll call "points" (you didn't like
wa:Representation for good reason). Each point is a pair of a content
(which is an octet sequence) and a media type (another octet
sequence). The space probably has lubs and other structure that I
don't want to bother to nail down unless we can get through the rest
of the discussion.

Now let C be any class. Say that C is ≤-closed if the following holds:
    for all generic resources g and h, C(g) and h ≤ g imply C(h).
(One might say C is "heritable under specialization".)

Now (the weak part of the theory) we say that "important" classes are
≤-closed.  Clearly not all classes, but the ones we mainly care about,
e.g. "was written by Charles Darwin" or "is about 80 pages long" or
"does not use the letter 'e'" (see
http://www.w3.org/2001/tag/awwsw/ir-axioms/20110225 for a longer
list).

OK. There is another theory, which says that terms that might be
interpreted as GRs in the above theory, might be interpreted as
points, but that the theory is incomplete, so we do not always have a
way to distinguish which point. Any reasoning we do regarding the
referent of such a term has to be robust to reinterpretation of the
term as a different point (specializing the same GR). We might even
have a satisfying interpretation that contains something like GRs,
suggesting this treatment may be indistinguishable from the previous
one.

Theory #2 is not Tim's, but it is simpler and leads to the same
conclusions, and maybe it's what you have had in mind. Is it? Or do
you have yet another approach in mind?  Maybe you wanted (theory #3)
to fix the equivalence relation at the top, so that there is no GR
ordering, but the GR that the URI names is determined by looking at a
single point?  That doesn't really work since it doesn't distinguish,
say, the referent of http://www.w3.org/TR/2004/REC-rdf-mt-20040210/
from the referent of http://www.w3.org/TR/rdf-mt/ ; among other
things.

Thanks
Jonathan

Received on Wednesday, 5 October 2011 21:50:05 UTC