> I would say that it's SOMETIMES better than nothing. It depends on what
> you're doing, what your requirements and goals are. Not every application
> needs long-term persistence of URLs -- whether through an 'abstraction
> layer' or not. ('abstraction layer' is just an implementation detail to get
> long-term persistence of URLs accross systems changes, right?  You don't
> always need something called an 'abstraction layer' to do that).  Almost
> every application does need bookmarkable URLs for the short/medium-term
> though.  If you're sacrificing short-term bookmarkable URLs for
> long-term-goal persistent but confusing/non-transparent/not-discoverable
> URLs, that may or may not be a good trade off.

Agreed. The operative phrase here is "trade off" as these inevitably come
into play when choices are made.

With very few exceptions, it's not realistic to believe that conceptual
problems can be solved outright. Consequently, solutions based on the notion
that they can inevitably ignore the costs associated with those paths.

A URL is already an abstraction, as is a domain name, as is a file system.
The question for any addressing issue is what level of abstraction is most
appropriate for the task at hand.