> Think of validating a 20Gb document made up of:
> 
>    <logs>
>      <log .... />
>      <log .... />
>      <log .... />
>      ...
>    </logs>
> 
> Very simple to validate as a pattern.  A *big* burden to 
> support assertions for the entire tree.
> 
> I'm not on the committee, but I suspect the requirement to 
> keep an entire document tree in memory for validation is a 
> non-starter.  Whereas dealing with all of the information 
> about an element and its attributes can be done easily 
> because that is where the validator is at the time.
> 
> . . . . . . . . . . . Ken

If you change that to 

<logs count="28654">
  <log>
  ...
</logs>

Then I think it's perfectly reasonable to have an assertion at the top level
that says

test="@count = count(log)"

And indeed the specification allows you to do so. Some implementations will
be able to handle this efficiently; some won't. Users will have to make
decisions (about designing their schemas, and about choosing their products)
that take this into account.

So I don't think the argument based on performance is overriding. The main
reason for not allowing navigation outside a subtree is the theory that a
type defines a set of instances, and that the set of instances of a type
doesn't depend on the phase of the moon.

Regards,

Michael Kay
http://www.saxonica.com/
http://twitter.com/michaelhkay