[
Lists Home |
Date Index |
Thread Index
]
"Henry S. Thompson" wrote:
> The logic of this decision is as follows:
>
> 1) schemes are going to be popular (Simon himself has already
> defined several);
> 2) short scheme names are likely to collide;
> 3) collision is an interop nightmare;
> 4) scheme names therefore need to incorporate domain names in some way;
> 5) we'll reuse the nsdecl/nsprefix design pattern.
>
> Constructive discussion on where the above logic falls down and/or how the above
> goals can be met in other ways will be welcome, here and/or to
> www-xml-linking-comments@w3.org.
The logic falls down with step 3, leaving 4 and 5 falso sequens. Mere lexical
collision is inevitable in any scheme of sufficient scale (and therefore the wrong
place to look for solutions for interop). Workable interop at global internetwork
scale cannot feasibly be based on a priori agreements embracing all likely
counterparties and resulting in names somehow globally unique. Interop at global
internetwork scale can be achieved only by the autonomous functioning of
independent processing nodes, each of which is locally responsible for
discriminating among the data sources it requires. Luckily, internetwork creation
is predicated upon universal addressing schemes (IP at one level, URL at another),
so a sufficient basis for each node to discriminate the provenance of each of its
data sources (and for that matter, its data or message targets) is already in
place. QNames are therefore otiose in a universe of such processing nodes, as well
as being practically unworkable at the scale desired.
Respectfully,
Walter Perry
|
