larry's position on URIs as names

Larry,

I wanted to ask this while it's still fresh... at my presentation on
web semantics I was making the point that GET/200 exchanges
("corresponds to" relationships in 2616 parlance) are of little use on
their own in discerning what entity the "URI owner" means to name with
the URI, since any set of responses is compatible with the entity
being a wide variety of things.  You made a comment about ambiguity
that I think was in agreement with this.  But I wasn't clear on your
exact position around the use of http: URIs as names because, as I had
just arrived at (3) below, providing a list of possible sources of
"other evidence", you appeared to disagree with one of my premises.  I
suspect that was a misunderstanding.

Can you tell me which of the following most closely matches your view
of best practice?

1) http: URIs don't ever make reliable names/designators for use as
the subjects of metadata assertions.  Use tdb: or guids or ____ instead.

2) http: URIs make reliable names only when HTTP exchanges with the
URI as request-URI play no role in determining what they name.

3) http: URIs make reliable names but only in the presence of evidence
other than HTTP exchanges with the URI as request-URI.  (If exchanges
are useful it's only when they're used in combination with that
evidence.)

4) http: URIs make reliable names.

A consequence of 2) might be, for example, that if you knew (via
independent communication with the "URI owner") that
http://example.com/x was intended to name some unchanging particular
PNG file, and did a GET soon thereafter on http://example.com/x and
got a PNG file, you could not infer that http://example.com/x was
intended to name *that* PNG file, absent other metadata.  Whereas with
3) you could.

If that's not good enough, the "other evidence" in 3) might be an SLA
that you and the URI owner have entered into that guarantees response
uniqueness over some period of time ("trust"), or metadata giving
enough checkable information about the image (e.g. a date/time or
checksum) that the chance of accidental collision is satisfyingly
small (verifiability).  Would evidence like this, in addition to the belief
(above) that the representation is fixed, enable use of HTTP
exchanges as a way to help figure out what entity is meant?

Just trying to figure out where you set the bar.

Jonathan

Received on Saturday, 27 March 2010 23:52:32 UTC