ISSUE 8: SCDL artifact resolution underspecified (schema adjustments for uniqueness?)

["Moberg Dale" <dmoberg@axway.com>]

Michael Rowley states

 

I believe that SCA definitions (composites, binding types, intents, etc) should all be unique within a single “contribution context”, by which I mean a contribution and the exported definitions of all of its dependent contributions.  If others agree, I’ll add that text.

 

Dale Moberg>> WSDL and also schema enforce uniqueness for named things that they intend to reference by qnames. For example, WSDL  1.1 uses the key constraint

·  (unique) the Identity-constraint definition asserts uniqueness, with respect to the content identified by [selector], of the tuples resulting from evaluation of the [fields] XPath expression(s).

·  (key) the Identity-constraint definition asserts uniqueness as for unique. key further asserts that all selected content actually has such tuples.

as follows:

 

  <xs:element name="definitions" type="wsdl:tDefinitions" >

    <xs:key name="message" >

      <xs:selector xpath="wsdl:message" />

      <xs:field xpath="@name" />

    </xs:key>

    <xs:key name="portType" >

      <xs:selector xpath="wsdl:portType" />

      <xs:field xpath="@name" />

    </xs:key>

    <xs:key name="binding" >

      <xs:selector xpath="wsdl:binding" />

      <xs:field xpath="@name" />

    </xs:key>

    <xs:key name="service" >

      <xs:selector xpath="wsdl:service" />

      <xs:field xpath="@name" />

    </xs:key>

    <xs:key name="import" >

      <xs:selector xpath="wsdl:import" />

      <xs:field xpath="@namespace" />

    </xs:key>

  </xs:element>

 

While this will only enforce uniqueness of named component, er thing, within the “targetNamespace”, it might be useful to add the appropriate constraints to the scdl schemas, especially if you add in the Assembly specification, a requirement for uniqueness of named scdl things that are to be referenced by qnames.

 

 

For XML Schema and WSDL definitions that are found by QNames, I think they can be ambiguous, but we can use the standard __Location attributes.  I think that should cover the practical issues with existing badly behaved artifacts.

 

I don’t know what to do about Java.  I’m inclined to continue to say very little.  We currently say that “the installed contribution provides the context” in which fully qualified classnames are resolved, but it doesn’t say _how_ they are resolved (in fact it says there may be multiple ways).  I think we should continue along those lines.