Re: attributes containing subordinate attributes

[James Clark <jjc@jclark.com>]

I think this is an instance of a more general problem.  It is possible
in TREX to write patterns that can never be matched by any input
document, because for example:

1. They require attributes containing structure
2. They require duplicate attributes
3. They require infinite structures

What should TREX say about these?

A related issue is patterns that allow, but do not require impossible
things such as 1-3.

Murata Makoto wrote:
> 
> The TREX spec allows patterns for attributes
> containing subordinate attributes.   For example:
> 
> <attribute name="foo"><attribute name="bar"><anyString/></attribute></attribute>
> 
> I do not see any values in allowing attributes to contain subordinate attributes.
> Certainly, this is not allowed in XML 1.0.

We need to be careful with terminology here.  There are two separate
issues.

1. Should the TREX data model allow attributes to contain attributes?

2. Should TREX place special restrictions on attribute patterns to make
it an error to write "nonsense" patterns, for some appropriate
definition of "nonsense"?

On 1, the TREX data model does not allow attributes to contain
attributes, and so is consistent with XML 1.0.

On 2, TREX currently does not have any such restrictions. At the moment,
such nonsense patterns will simply fail to match any input document. The
difficulty is in coming up with an appropriate definition of "nonsense".

I would like to consider this in a general way, rather than on a case by
case basis.  I am very keen to avoid complicating both the spec and
implementations with numerous ad hoc restrictions.

> I would argue that such patterns are useless, doubtful, and should be dropped.
> 
> Cheers,
> 
> Makoto