[
Lists Home |
Date Index |
Thread Index
]
>>>Why aren't defined patterns named with QNames? ...
>>
>>How would this aid portability?
>
>
> I'll give an (trivial) example that demonstrates. Say I want to
> validate against a custom version of XHTML that supports XLink. I
> include James Clark's Relax NG Schema for XHTML [1] and John Cowan's RNG
> for XLink [2]. Both schemas are OK by themselves, however, including
> them would give a conflict with start definitions and "title"
> definitions. The start definition would be expected; however, the title
> definition is not obvious - it's defined in a small module [3] among
> *many* definitions [4]. If I was combining two large languages (such as
> XHTML and MathML), then it would be hard to find all of the conflicting
> names.
Combining independently created schemas is an important requirement, but
I don't think using QNames for definitions is the solution.
For a start, note that using QNames for definitions wouldn't solve your
XHTML+XLink problem for two reasons. First, the XHTML schema is written
in a closed way and doesn't allow attributes from foreign namespaces.
Second, an instance needs to be validated concurrently against both the
XLink schema and the XHTML schema. What you need is a language that can
say: to validate this instance, use the XLink schema to validate the
complete instance and also use the XHTML schema to validate the instance
after stripping out attributes from the XLink namespace. NRL [1] is
designed to do this sort of thing. RELAX NG isn't intended to solve
every validation problem by itself.
For a case like combining XHTML and MathML, RELAX NG already provides a
solution to avoid conflicting definition names. The solution is nested
grammars. In RELAX NG, the grammar element is allowed anywhere not just
at the top level. When a grammar element occurs inside another grammar
element, the definitions in the nested grammar element are local to that
grammar and don't conflict with the definitions in the containing
grammar. Thus to make a XHTML+MathML schema, you can just do something
like:
<grammar>
<include href="xhtml.rng"/>
<define name="Inline.class" combine="choice">
<externalRef href="mathml.rng"/>
</define>
</grammar>
In RELAX NG, <include> merges grammars, whereas <externalRef> is
replaced by the pattern in the referenced file. So in this case, the
result will be that MathML grammar will be nested inside the XHTML
grammar, so the names in the MathML grammar will not conflict with the
names in the XHTML grammar. RELAX NG also provides a <parentRef>
element that allows nested grammars to selectively reference definitions
from their parent grammars. It's a little like the block scoping found
in many programming languages.
There are a couple of reasons I don't think QNames for definitions are a
good idea:
- URIs and QNames make sense for global objects, but definitions are not
global; they're local to a particular grammar. It would be a bit like
using package names for local variables.
- Some XML users make heavy use of XML namespaces; other XML users think
they are an unnecessary complication don't want to have anything to do
with them. RELAX NG is designed to work well for both kinds of user.
(In contrast, NRL doesn't have this goal.) However, the need to combine
schemas is not unique to XML namespace users. I might want to use a
math schema inside a document schema and not use XML namespaces for
either. Using QNames doesn't seem like a natural solution in a
non-namespace environment.
I think there is scope for improvement in RELAX NG in this area. At the
moment, the only mechanisms for referencing grammars are <include> and
<externalRef>. These mechanism both work by syntactically including the
referenced grammar. This doesn't work for mutually recursive grammars.
By contrast, the <ref> element can handle mutually recursive
definitions, because it doesn't work at a purely syntactic level: it
doesn't just replace the <ref> element by the content of the referenced
definition. I think there's scope for a grammar reference mechanism that
doesn't work just by syntactic inclusion; but how exactly it should work
is something I haven't yet figured out.
James
[1] http://www.thaiopensource.com/relaxng/nrl.html
|
