Re: Can everyone be happy?

>Actually there seems to be fairly clear consensus that nothing in
>particular need be identified by the namespace name if used as a URI.
>The only two people I've seen speak against this with any real
>conviction are Dan Connolly and Tim Berners-Lee.

... I'm not sure even TB-L has spoken against this. His assertion, if I
understood it correctly, was that the namespace's identity was in fact the
URI. While dereferencing that URI might be useful at some point in the
future, Tim seemed more concerned with the fact that simply being a URI
buys you a potentially useful set of implications. Some parts of the URI
space have concepts of "ownership" and the like associated with them, and
using URIs as the name would permit implicitly tapping into those concepts.

Dereferencing the URI really is a completely independent issue. Any name
can be "dereferenced" by architecting appropriate catalog schemes or
ancillary binding declarations.  Hence this point really can be set aside
and dealt with after the question of naming itself has been resolved.

I believe that conflating these two points has been a major impediment to
understanding in this debate.  Perhaps unavoidable, since most of us think
of URIs as directly naming something specific. But in fact they don't; they
are pointers which occupy an N-dimensional plane that lies _between_ the
application and the data, and it's possible and (occasionally) useful to
discuss these pointer objects -- the URIs themselves -- without going
further.

Took me a long time to get my head wrapped around that, so I don't blame
others for being confused.



Having said all that: It still remains unclear that URIs do, in fact, have
the characteristics desired for use in identifying namespaces. The behavior
will not match most user expectations, and in fact the behavior must be
constrained (stronger inequality) in a way that diverges somewhat from the
normal interpretation of URIs. I'm willing to accept those weaknesses if
that's what it takes to put this debate behind us.

And I remain _extremely_ unconvinced that premitting relative references to
namespace names is desirable, even if this way of viewing the problem makes
the resulting breakage architecturally meaningful. The behavior will be
excessively fragile, and the benefits seem excessively weak. Better to
remove the land mines than to put up signs and maps.

______________________________________
Joe Kesselman  / IBM Research

Received on Thursday, 22 June 2000 11:17:35 UTC