Re: [xml-dev] Re: determining ID-ness in XML

[Jeff Rafter <jeffrafter@defined.net>]

Tim Bray wrote:

> Giving something a unique address is arguably 100%
> orthogonal to validity.  The only reason we mix these two
> up is that IDs were one of the many things that got
> thrown into the DTD basket by SGML.

I have been listening in for the most part throughout this thread-- but it
strikes me-- what if we started fresh today? Which of these solutions would
win out? It seems that something along the line of xml:id would win.
Although I came out against this, I feel that it is the most structured way
of handling IDs-- the way we would handle them if we started from scratch
today. I feel that the PI solution is by far the least intrusive and a very
useable solution-- but utilizing xml: seems correct-- if there is such a
notion.

In general, if we accept the argument that IDs *are* orthogonal to both
Validation and Well-Formedness then removing them to an xml:id attribute
would be the best solution. I am struggling to juxtapose this notion with
the use of the xml:idatts approach. It seems that declaring which attributes
are ID attributes brings us back to step one (though the inheritance for
child nodes would save on bandwidth-- it is less direct, and therefore
requires more time to learn). If we started over I think xml:id would get my
vote. It would support the streamed model and would eliminate some of the
copy/paste hand editing problems that have been brought up because it would
be attached to the element.

OTOH one of the primary benefits of xml:idatts is that it would support
David Brownell's suggestion to allow multiple (labeled) IDs. We could make
an xml:ids that was type NMTOKENS but each ID would not be labeled.

Jeff Rafter
Defined Systems
http://www.defined.net
XML Development and Developer Web Hosting