Re: Relative Namespaces

[Richard Tobin <richard@cogsci.ed.ac.uk>]

> I remember that someone, very recently, posted here that "setting the
> namespace to the empty string" was intended to mean that this was a way to
> signal that you want to remove a namespace that might otherwise apply (change
> it to a null, that is).  It wasn't supposed to mean that a null namespace is
> actually the same as an empty string namespace.

That's right, but it only applies to unprefixed names.

According to the namespaces spec, the namespace name in a namespace
declaration may not be empty in the xmlns:prefix case, but only in the
xmlns case.  This corresponds to it being possible for an unprefixed
name to be in no namespace, but not for a prefixed name.

A digression:

There are three states a name can be in w.r.t. namespaces:

  (1) in a namespace (which is defined by an absolute URI)
  (2) in no namespace
  (3) undefined (ie it's illegal to use the name in a context where
      it would be interpreted as a qualified name)

Prefixed names start in state (3), and namespace declarations of the
form xmlns:prefix="URI" move them into state (1).  After that, all you
can do is keep them in state (1) and change the namespace by using
further declarations of the same form.

Unprefixed names start in state (2), and can be moved to state (1) by
namespace declarations of the form xmlns="URI".  You can change the
namespace by further declarations of this form, or move them back to
state (2) by a declaration xmlns="".

So there are three states, but only two syntaxes (absolute URI and
empty string) for moving between them.  Prefixed names can only ever
be in states (1) and (3), unprefixed names only in states (1) and (2).
It would be nice to be able to move prefixed names back to state (3),
so that you could include an XML element from one document in another
without the risk of inadvertent prefix capture.  The syntax
xmlns:prefix="" could be used for this (but it is not so defined at
present).

-- Richard