[
Lists Home |
Date Index |
Thread Index
]
From: "John Cowan" <jcowan@reutershealth.com>
> An interesting idea. All schema languages known to me are weak in
> supporting mixed content; even RNG, which is the strongest, cannot
> express this constraint. (I don't really know Schematron: can it
> cope?)
Not nicely.
The approach I favour is to use W3C's Martin Duerst's hulls and kernels
approach[1] on all the children to any level in an element.
If the selector or validator mechanism also provided a filter mechanism
(e.g. Schemachine's <exclude>[2]) then that would filter out any annotation elements
that you are not interested in, or elements marked in other namespaces.
This helps move XML away from the backdoor third normal form
and static typing of WXS and towards supporting mixed content
and instance-based inheritence, which is useful sometimes.
I think there are three gaps in schema languages at the moment:
one is concerned with localized data and complex values, another
is to do with link integrity checking and controlled vocabularies,
and the third has to do with supporting inherited effects (such
as SVG's [or is it FOs?] effect that a parent can have any
attribute that is propert to a child, if that is really what is going on)
> It also turns out to be a hard problem to say what "Dutch characters"
> are. Even English has corner cases like coöperate and façade.
Pick a dictionary and enumerate the characters! Finding that unusual cases
are present is probably the point of validating rather than being
an argument against it! (I was surprised that a European mentioned
it, because it is something that I would more associate with CJK.)
One argument I have heard against providing this kind of feature
is that it would discourage full Unicode use or entrench minimal
character sets.
Cheers
Rick Jelliffe
[1] http://www.w3.org/TR/1999/NOTE-charcol-19991105/
[2] http://www.topologi.com/public/Schemachine.pdf
|
