Re: Catalog Requirements (resend after 2-day bounce)

[John Cowan <jcowan@reutershealth.com>]

Norman Walsh wrote:

> 1. At the moment, catalog lookup is non-recursive and terminates at
> the first match. I worry that if we add a URI keyword, we'll find
> users that want lookup to be recursive: map the public id to a URN
> (via PUBLIC), then map that URN to a URL (via URI) and I don't think
> we want to go there.

I agree, but we can just say loudly that lookup is never recursive
(except for delegation); it maps "what you see" into "what you get",
period.

> 2. We're going to introduce a thorny precedence question. If a URI
> matches both a SYSTEM keyword and a URI keyword, which one wins?
> There are four answers, I think:
> 
>   a. SYSTEM always wins
>   b. URI always wins
>   c. SYSTEM and URI are equal, the first one encountered wins
>   d. It depends on what you're looking up

I think SYSTEM is applied only to system identifiers, so I think
SYSTEM wins for system identifiers, URI wins for other URIs.
This is a variant of d.

> The last is maybe the most "correct", but it's also the most difficult
> to explain. And we'll have to answer the question: if you attempt
> SYSTEM lookup and fail, can you then (in conformance) attempt URI
> lookup?

Answer: no.  SYSTEM is applied when you have a system identifier
(which must be a URI, to be sure); other lookups are applied to
other kinds of URIs.

> 3. I think it will open the door to the question of how many different
> sorts of URIs are there, and shouldn't I be able to map them all
> differently?
> 
>   URI "uri" "uri"
>   NAMESPACE "uri" "uri"
>   STYLESHEET "uri" "uri"
>   SCHEMALOC "uri" "uri"
>   ...
> 
> And I don't know if I want to go there or not. I want to keep this
> simple and as close to TR9401 as is practical.

I do think that if we are introducing general URI-URI rewriting, we do
have to go here, although the syntax URI NAMESPACE "uri" "uri" might be
better.  (I think generalized 9401 parsers can parse that even if
they don't understand it, right?)

That would leave URI for all other URIs, ones that are application-specific.

Of course, we could dispense with this by not doing generalized
URI-URI rewriting, but only rewriting system ids and public ids.
Maybe not as universal, but easier to delimit.

-- 
There is / one art                   || John Cowan <jcowan@reutershealth.com>
no more / no less                    || http://www.reutershealth.com
to do / all things                   || http://www.ccil.org/~cowan
with art- / lessness                 \\ -- Piet Hein