Service operation (invocation contract) only supports one fault type

[objectiser]

I have a BPMN2 choreography that is generating a switchyard bean service where one of the operations returns two faults, and I get the following exception:

org.switchyard.exception.SwitchYardException: Service operations on a Java interface can only throw one type of exception.

    at org.switchyard.metadata.java.JavaService$OperationTypeQNames.fault(JavaService.java:241)

    at org.switchyard.metadata.java.JavaService.fromClass(JavaService.java:103)

    at org.switchyard.deploy.internal.Deployment.loadServiceInterface(Deployment.java:319)

    at org.switchyard.deploy.internal.Deployment.getComponentServiceInterface(Deployment.java:303)

    at org.switchyard.deploy.internal.Deployment.deployImplementations(Deployment.java:392)

    at org.switchyard.deploy.internal.Deployment.start(Deployment.java:140)

Was wondering why the invocation contract only supports one fault response type?

Regards

Gary

[kcbabo]

The issue with multiple faults is how we handle transformation.  Let's say we have this Java interface for the provider:

Bar submit(Foo foo) throws ABC, XYZ;

And this WSDL definition for the consumer:

           <input message="tns:foo"/>

           <output message="tns:bar"/>

  <fault message="tns:kaboom"/>

  <fault message="tns:kachow"/>

If the provider throws ABC, do we transform to tns:kaboom or tns:kachow? 

[objectiser]

The same issue would apply if we had polymorphic operations - so I guess they are not supported?

Keith Babo wrote:

If the provider throws ABC, do we transform to tns:kaboom or tns:kachow? 

My suggestion would be to transform it to the type that has a registered transformer for the pair - as it is unlikely that transformers would have been registered for both ABC to kaboom and ABC to kachow.

[kcbabo]

The same issue would apply if we had polymorphic operations - so I guess they are not supported?

Correct.  FWIW, this practice is specifically prohibited in WS-I basic profile and support was also removed in WSDL 1.2.

My suggestion would be to transform it to the type that has a registered transformer for the pair - as it is unlikely that transformers would have been registered for both ABC to kaboom and ABC to kachow.

That could work, particularly with the rather basic mapping we have in place today.  Would you mind filing a JIRA and we can pick it up for 0.6?

[objectiser]

Done - https://issues.jboss.org/browse/SWITCHYARD-982