[Standards-JIG] sequence vs. choice in Jabber/XMPP schemas

Trejkaz trejkaz at trypticon.org
Tue Aug 16 00:10:35 UTC 2005


Quoting Peter Saint-Andre <stpeter at jabber.org>:
> 1. Keep using xs:sequence and require existing implementations to pay 
> attention to the order of child elements.

Ouch.  That's going to be hell for people working with DOM-like APIs 
where it's
not so easy to insert new elements into them in the expected order.  On the
other hand, it's really easy for people generating their XML by string
concatenation.

> 2. Switch to xs:choice and care less about the number of allowable 
> child elements.

If anything, <xs:all> is probably a closer fit in most of these cases.

e.g. JEP-0077 (In-Band Registration) currently has:

    <xs:sequence minOccurs='0'>
      <xs:element name='username' type='xs:string' minOccurs='0'/>
      <xs:element name='password' type='xs:string' minOccurs='0'/>
      ...
    </xs:sequence>

It could be...

    <xs:all minOccurs='0' maxOccurs='1'>
      <xs:element name='username' type='xs:string' minOccurs='0'/>
      <xs:element name='password' type='xs:string' minOccurs='0'/>
      ...
    </xs:all>

That way it still allows them in any order, but doesn't allow people to 
use more
than one of each.  This is better than both <xs:sequence> and <xs:choice> for
the majority of situations you can find in the JEPs.

> 3. Define two different sets of schemas: a strict one following (1) 
> and a loose one following (2).

XHTML did this in version 1.0, and practically nobody used the Strict 
version as
there was no incentive given to use it.  They ended up making 1.1 effectively
strict, removing the non-strict version.

> 4. Switch from W3C XML Schema to a different schema language, such as 
> RELAX NG, Schematron, or Examplotron.

RELAX NG is certainly much more powerful than W3C's XML schema.  They 
even seem
to have acknowledged this themselves, as the drafts for XHTML 2.0 are entirely
written in RELAX NG schema. :-)

It allows situations which we have that we can't currently represent.  For
instance, we could specify that <status> and <show> elements don't 
occur inside
<presence> when type="subscribe"... and that's just the tip of the iceberg, of
course.

In addition to this, RELAX NG has a "Compact Syntax" which is just as flexible
as the normal syntax, but much easier to read and write.  It's easier to read
than DTDs, which were already a great deal easier to read than W3C XML 
schemas.
:-)

And of course, because RELAX NG is a superset of all other schema 
functionality,
there exist tools to transform it into other formats (approximating where
necessary, of course.)

My main choice then, is (4).  And (2) is a reasonable fallback as long as
<xs:all> is used in places where it makes more sense than <xs:choice>.

TX


----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.




More information about the Standards mailing list