I have a meta-model where the following classes and 
aggregations are present:

LogicGraph - aggregates a collection of of LogicNode
LogicNode - has two aggregation associations of LogicLink. 
One aggregation represents the node's input links, and one 
aggregation represents node's output links.

If I create a model with two LogicNode objects, and one 
LogicLink between the two LogicNodes, then one of the the 
LogicNodes has the LogicLink in its input collection and 
the other LogicNode has the LogicLink in its output 
collection. This is modelled OK (using the Unisys MOF 
plugin for Rose), and I can successfully add instances to 
the MDR using the following code: (where "pkg" is an MDR 
RefPackage extent)

        // Create a simple Graph and two nodes
        LogicGraph graph = pkg.getLogicGraph
().createLogicGraph();
        graph.setName("MyGraph");
        StartLogicNode start = pkg.getStartLogicNode
().createStartLogicNode();
        start.setName("Start Node");
        EndLogicNode end = pkg.getEndLogicNode
().createEndLogicNode();
        end.setName("End Node");
        // Add both nodes to the graph's nodes Collection
        graph.getNodes().add(start);
        graph.getNodes().add(end);
        // Add a link between the nodes by addding the link
        // to one node's input collection and the other 
node's
        // output collection
        LogicLink link = pkg.getLogicLink().createLogicLink
();
        link.setName("StartToEndLink");
        start.getOutLinks().add(link);
        end.getInLinks().add(link);

This is also persisted OK using 
org.netbeans.api.mdr.XMIWriter.getDefault() which produces 
the following XMI fragment:

  <XMI.content>
    <APML.LogicGraph xmi.id = 'a1' name = 'MyGraph'>
      <APML.LogicGraph.nodes>
        <APML.EndLogicNode xmi.id = 'a2' name = 'End Node' 
success = 'false'>
          <APML.LogicNode.inLinks>
            <APML.LogicLink xmi.id = 'a3' name 
= 'StartToEndLink'/>
          </APML.LogicNode.inLinks>
        </APML.EndLogicNode>
        <APML.StartLogicNode xmi.id = 'a4' name = 'Start 
Node'>
          <APML.LogicNode.outLinks>
            <APML.LogicLink xmi.id = 'a3' name 
= 'StartToEndLink'/>
          </APML.LogicNode.outLinks>
        </APML.StartLogicNode>
      </APML.LogicGraph.nodes>
    </APML.LogicGraph>
  </XMI.content>

However, when attempting to load the XMI file into an 
extent using the XMIReader 
(org.netbeans.api.mdr.XMIReader.getDefault()) the 
following exception is thrown.

org.netbeans.lib.jmi.util.DebugException: The same value 
of xmi.idref used second time: a3
	at org.apache.crimson.parser.Parser2.parseInternal
(Parser2.java:524)
	at org.apache.crimson.parser.Parser2.parse
(Parser2.java:305)
	at org.apache.crimson.parser.XMLReaderImpl.parse
(XMLReaderImpl.java:442)
	at javax.xml.parsers.SAXParser.parse
(SAXParser.java:393)
	at org.netbeans.lib.jmi.xmi.XmiSAXReader.read
(XmiSAXReader.java:99)
	at org.netbeans.lib.jmi.xmi.XmiSAXReader.read
(XmiSAXReader.java:71)
	at org.netbeans.lib.jmi.xmi.XMISaxReaderImpl.read
(XMISaxReaderImpl.java:37)
	at 
net.agentis.apml.domain.examples.persist.LoadExtent.execute
(LoadExtent.java:99)
	at 
net.agentis.apml.domain.examples.persist.LoadExtent.perform
Action(LoadExtent.java:70)
	at 
org.openide.util.actions.CallableSystemAction.actionPerform
ed(CallableSystemAction.java:69)
	at org.netbeans.core.ModuleActions$1.run
(ModuleActions.java:105)
	at org.openide.util.Task.run(Task.java:152)
	at 
org.openide.util.RequestProcessor$ProcessorThread.run
(RequestProcessor.java:622)
javax.jmi.xmi.MalformedXMIException: 
org.netbeans.lib.jmi.util.DebugException: The same value 
of xmi.idref used second time: a3
	at org.netbeans.lib.jmi.xmi.XMISaxReaderImpl.read
(XMISaxReaderImpl.java:43)
	at 
net.agentis.apml.domain.examples.persist.LoadExtent.execute
(LoadExtent.java:99)
	at 
net.agentis.apml.domain.examples.persist.LoadExtent.perform
Action(LoadExtent.java:70)
	at 
org.openide.util.actions.CallableSystemAction.actionPerform
ed(CallableSystemAction.java:69)
	at org.netbeans.core.ModuleActions$1.run
(ModuleActions.java:105)
	at org.openide.util.Task.run(Task.java:152)
	at 
org.openide.util.RequestProcessor$ProcessorThread.run
(RequestProcessor.java:622)
javax.jmi.xmi.MalformedXMIException: 
org.netbeans.lib.jmi.util.DebugException: The same value 
of xmi.idref used second time: a3

Clearly the SAXParser complains when both LogicNode 
objects refer to the same instance of LogicLink ('a3').

Should the MDR implementation allow such aggregations, or 
is there a problem with XMIReader?

I notice that if I try to aggregate the same instance of 
an object to multiple container objects via specific 
attributes, e.g.

      // Create one instance of a Graphic
	Graphic myGraphic = pkg.getGraphic().createGraphic
();
      // Create two Logic Graphs
	LogicGraph graphA = pkg.getLogicGraph
().createLogicGraph();
	LogicGraph graphB = pkg.getLogicGraph
().createLogicGraph();
      // Attempt to add the same Graphic to multiple graphs
      graphA.setGraphic(myGraphic);
      graphB.setGraphic(myGraphic); // Oh dear - this 
won't work

Then an exception is thrown when attempting to aggregate 
to the second "graph" object in the MDR.
So should I be able to aggregate to multiple objects via 
Collections (as the original LogicGraph, LogicNode, 
LogicLink example does) or not. If so then I should be 
able to persist and reload the model. If not, then MDR 
probably shouldn't let me aggregate the same object 
instance to multiple aggregated collections.