Mike - We can create meta schema and write XSL utility to create ultimate DFDL schema or code it by hand; the latter is a one time effort and undoubtedly there would be some overhead to update the schema by hand from one version to another but I don't expect many drastic changes to come from one version of the spec to other. The key part of the issue I am trying to resolve is whether the attributes in the special annotations will be namespace qualified or whether they will be local attributes (ie without namespace qualification). It affects how the ultimate DFDL schema describing annotations will be used by users and tools..

Say if we go with local attributes in complex types and global attributes (which by the way be required to support short form of annotations), then there will be a processing overhead as you cannot easily determine if a particular attribute inherited from parent container or specified through short from is overriding the attribute in the special annotation or not.. While making such decisions you would have to do comparison based on base name of the attribute and not explicit NCName of the attribute etc. Also you would have to ensure that you take only those attributes from short form of annotations which belong to dfdl namespace, strip their namespace part for comparing with local attribute name and ensuring that the local attributes being matched are part of the enumeration set defined through dfdl spec etc..

However if you make the attributes in complex type of special annotations namespace qualified, then you have somewhat usability issue as I hilited in the example but processing wise it is quite efficient and consistent.

Suman Kalia
IBM Toronto Lab
WMB Toolkit Architect and Development Lead
WebSphere Business Integration Application Connectivity Tools

http://www.ibm.com/developerworks/websphere/zones/businessintegration/wmb.html

Tel : 905-413-3923 T/L 969-3923
Fax : 905-413-4850 T/L 969-4850
Internet ID : kalia@ca.ibm.com

From:	"Mike Beckerle" <mbeckerle.dfdl@gmail.com>
To:	Suman Kalia/Toronto/IBM@IBMCA, "'Alan Powell'" <alan_powell@uk.ibm.com>, "'Steve Hanson'" <smh@uk.ibm.com>
Cc:	<dfdl-wg@ogf.org>
Date:	04/22/2009 08:30 AM
Subject:	Go Meta - RE: DFDL Schema - Mulitple forms of Annotation syntax

These difficulties, all seem to arise from the intellectual objective to leverage XML Schema validation against DFDL schema annotations maximally. Given the 3 syntaxes, this is a bit painful if done directly.

I suggest the following approach. Write a "Meta XML Schema" for DFDL annotations. This XML document would contain exactly one list of all the DFDL annotations. E.g., something like:


<dfdl:defineProperty name="initiator" allowedOn="complexType sequence element" type="string" expressionAllowed="yes"/>
<dfdl:defineProperty name="terminator" allowedOn="complexType sequence element" type="string" expressionAllowed="yes"/>
...

This document could be turned into the set of redundant annotations on a "real" XML Schema for DFDL annotations by a XSL stylesheet translation. This would actually be a pretty easy XSL stylesheet. Each dfdl:defineProperty in the meta schema would end up putting content into the big enum, a local and global attribute definition, and anywhere else needed in the "real" XML Schema for DFDL annotations.

I would suggest that doc strings might also be good to put in: e.g.,

<dfdl:defineProperty name="initiator" allowedOn="complexType sequence element" type="string" expressionAllowed="yes">
Defines a string used to recognize the beginning of a construct..blah blah blah
</dfdl:defineProperty>

This could then be used to generate online doc also.

I always use this sort of "go meta" approach whenever I have redundant, but isomorphic things happening in code. I always create a spec and generate the redundant instances from that. This helps tremendously with QA, because whole classes of inconsistency errors are ruled out.

...mike

1. Shot form syntax
<xs:element name="foo" type="xs:int" maxOccurs="unbounded" dfdl:representation="text" dfdl:initiator="{" dfdl:terminator=”}”/>

Note: It requires all properties to be defined as global attributes in schema as each attribute has to be uniquely identifiable and accessible within the DFDL namespace

2. Long form syntax ( used for dfdl:format and specialized annotations such as dfdl:element etc)

<xs:element name="foo" type="xs:int" maxOccurs="unbounded">
<xs:annotation><xs:appinfo source=”http://www.ogf.org/dfdl/”>
<dfdl:format representation="text" initiator=”{“ terminator="}"/>
</xs:appinfo></xs:annotation>
</xs:element>

Note: Attributes specified here are not qualified with dfdl namespace prefix. It means the attributes have to be defined locally within the complex type are not accessible directly (like in the short form syntax)

3. Long form property representation to ease syntactic expression difficulties with data in attribute values:
<xs:annotation>
<xs:appinfo source=”http://www.ogf.org/dfdl/”>
<dfdl:format>
<dfdl:property name='encoding'>utf-8</dfdl:property>
<dfdl:property name='separator'>\n</dfdl:property>
<dfdl:property name=’initiator’><[CDATA[<!-- ]]></dfdl:property>
</dfdl:format>
</xs:appinfo>
</xs:annotation>

Note: The name attribute will be enumeration of base name of all DFDL properties (ie without namespace prefix). On the specialized annotations (ie dfdl:element) we will also need to have a repeating child element definition for dfdl:property. This annotation will appear on specialized annotation element for dfdl attributes whose value cannot be stored as XML attribute.

PS: Alan - when you update the spec for specialized annotation please mention this use case and put an example also.

To support syntax form 1 & 2, it requires defining all attribute definition for each DFDL property multiple times ie one global attribute and a local attribute in the complex type of each long form annotation element. Using Attribute groups we can address the latter issue ie define related local attributes once in the attribute group and include the attribute group in one or more complex type. But it still requires defining the attribute definition for each DFDL property atleast twice (ie one global and one local attribute definition). This duplication can potentially lead to inconsistency and inadvertently updating one definition and not the other.

One option that we can consider to remove this redundancy is to define only the global attribute definition for each DFDL property and include it through attribute reference in each of the complex type of dfdl:format and specialized annotations. The downside of this approach is that attributes will appear qualified in the dfdl:format and specialized annotation elements. Given below is how the dfdl:format annotation will appear in the user defined schema where each attribute is prefixed with dfdl (hilited in green).

<xs:element name="foo" type="xs:int" maxOccurs="unbounded">
<xs:annotation><xs:appinfo source=”http://www.ogf.org/dfdl/”>
<dfdl:format dfdl:representation="text" dfdl:initiator=”{“ dfdl:terminator="}"/>
</xs:appinfo></xs:annotation>
</xs:element>

From technical perspective (e.g. schema integrity , avoiding duplication definitions etc.), this is better but it does have usability issue.

Suman Kalia
IBM Toronto Lab
WMB Toolkit Architect and Development Lead
WebSphere Business Integration Application Connectivity Tools

http://www.ibm.com/developerworks/websphere/zones/businessintegration/wmb.html

Tel : 905-413-3923 T/L 969-3923
Fax : 905-413-4850 T/L 969-4850
Internet ID : kalia@ca.ibm.com
Fax : 905-413-4850 T/L 969-4850
Internet ID : kalia@ca.ibm.com