> -----Original Message-----
> From: Robert Sanderson [mailto:[log in to unmask]]
> Sent: Tuesday, September 10, 2002 11:21 AM
> But that table won't be machine readable. For every index set
> that anyone
> creates, we'll need to add logic for index structure.

Of course it will be machine readable!  Why wouldn't it?  This is the most
trivial variant on Explain imaginable.

> Secondly, that table may not exist if the underlying
> implementation isn't
> Z39.50.  I'm sure that Amazon would create indexes like
> 'actor' 'format'
> 'director' and so forth... Which don't have attribute
> combinations readily
> available.  Do we then create a Z39.50 attribute set to map them into?

The strenth of the whole attributes discussion over the years is that we
have done an excellent job of identifying the semantic components of the
concept of index.  So, yes, I think that if Amazon wants to expose an index,
they need to define the underlying semantic concepts of the index and our
attribute combinations is exactly how they should do that.

Now, we may want to come up with some sort of semantic web name for z39.50
attributes, for political purposes.  But the concepts contained in them are

> > But if I define index "pooh" as being use=1016,
> structure=6, ... is there
> > any problem?  The name of the index is just a string.  It
> has no implicit
> > semantics.
> Then how do we (being the client writers) interoperate without prior
> knowledge of your index naming system?

Because in my explain record I said what index sets I support.  I said that
by giving the URL of the index set's definition.  That definition had a list
of every index in the set and the attribute combination that defined it.