Re: jettisoning baggage

At 11:33 PM 6/14/00 -0500, Al Gilman wrote:
>>So long as we have unique identifiers, we can safely build infrastructure
>>to support them.  There is nothing sacred about namespaces beyond 'unique
>>identifiers' and a set of syntactical tools for making those unique
>>identifiers work in XML.
>>
>>I suppose that makes me a strong advocate of 'literal'.
>
>Or does it make you a likely customer for the urn:oid scheme?

It could, but that's not a likely solution.

>Literal comparison doesn't get you uniqueness.  That's inverse logic.

No, but it does remove the complex issues involving context and makes it
clear to users that 'what you write is what you get'.  Adding that to "thy
shalt not covet thy neighbor's identifier" makes it quite plain that
context isn't going to take care of this for you.

It also gets rid of the philosophical spin we've enjoyed here thus far.

>What selecting literal, or selecting absolutized, as the basis of
>comparison does is this: it makes precise the terms of reference for
>accounting how far off you are from achieving uniqueness.  Neither one will
>ipso_facto get you there.  That's a whole 'nother battle.

Using literal shortens the road to getting there quite dramatically, and
removes the issues regarding the absolutization process and whether or not
anything might be at the end of the road simultaneously.

>This is the baggage we need to jettison.  The sacred 'unique' of the
>identifiers.  The idea that somehow global uniqueness is a logical
>prerequisite for using namespaces.  Global uniqueness will never be
>achieved beyond the attainment of low collision rates.  The statistics can
>be improved through use of devices such as the urn:oid: scheme.  But
>guaranteed uniqueness is not in the cards.  Don't loose sleep over it.

Guaranteed anything is not in the cards anyway - URIs are unfortunately not
up to guaranteeing very much of anything, and we're stuck in a salvage
operation that is showing very little return.  Using 'literal' at least
lets us see what we're doing plainly, without concern for context.

There is no 'sacred unique' in the Namespaces spec, except insofar as
attribute names need to be kept out of each others' way.  I suspect that's
the result of the natural limitations of URIs, which while 'Uniform' make
no claims to 'unique', and hence are not a great choice for schemes that
rely on them for parts of names.

At this point, I'll be happy to settle for low collision rates (occasional
drunks), provided that the means of determining collisions is clear,
lightweight, and doesn't bring a few thousand philosophical issues into the
room.

To me, that still screams 'literal'.

Simon St.Laurent
XML Elements of Style / XML: A Primer, 2nd Ed.
http://www.simonstl.com - XML essays and books

Received on Thursday, 15 June 2000 08:36:08 UTC