[Prev][Next][Index][Thread]

Re: XML catalog draft



[Apologies if you get this twice; the first seems to have evanesced into
 the aether. -T.]

On reflection, and although I agree with pretty well every word in 
the XML catalog draft, and think it's a fine piece of work, I cannot 
support its inclusion in XML.  Background:

1. We left out PUBLIC originally because we didn't want to bless 
   a syntax (e.g. FPI or URN) or a resolution mechanism.  Making
   everything a URL has the virtue that it works.
2. People said they wanted PUBLIC.  Michael asked: "even without a single
   resolution mechanism?"  Several, most eloquently Deb Lapeyre,
   said "Yes."  I for one (and I think most of the ERB) can see
   no good reason to resist this desire.
3. The recent draft blesses a resolution mechanism (Socats) which it
   specifies at some length, in a process that packs in a lot of design
   decisions (DELEGATE/CATALOG, prefix rules) very compactly, but does
   not really provide a mechanism to find the catalog.  I think the
   design decisions are good.

What bothers me is: 

a. The length and complexity of the discourse on catalog usage;  I think
   the XML spec should be shorter, not longer.
b. The fact that this discourse overlaps with TR9401.  I think that we
   should avoid rewriting parts of other specs, with the obvious exception
   of ISO 8879.  We do not do this with URLs, for example.
c. The fact that some but not all of TR9401 is here; notably missing is
   the find-the-catalog logic.  Are the parts of 9401 where there is 
   overlap guaranteed not to change?  Are the parts of 9401 that are 
   omitted notably lacking implementations?

I think we should:

 - bless the PUBLIC keyword & associated identifier
 - make a decision as to what should be done when both PUBLIC and SYSTEM
   are there [I can't recall detecting a consensus; did I just miss it?]
 - provide a pointer to TR9401 in the write-up on the PUBLIC identifier,
   saying that this is one proven-in-practice way to resolve them
 - improve support for catalogs by providing a special PI, as recommended
   by a couple of WG-ers, to help define the BASE and thus find the
   catalog
 - investigate the problem of what seems like the unnecessary
   restrictions on MINIMUM LITERAL; I don't think it's desirable to
   say that a PUBLIC identifier can't be a URN, which this would do.

I'm not sure what the right thing to do is with the current
catalog proposal.  It seems to represent the best thinking, by the people
who know, on how to get good mileage out of Socats; it would be a 
pity to lose that.  But at the moment it just doesn't seem like a good call 
to wire this into XML.

Cheers, Tim Bray
tbray@textuality.com http://www.textuality.com/ +1-604-708-9592