Classification: UNCLASSIFIED Caveats: NONE
Hi, Simon and others, I'm reposting this because I found my earlier post was grouped under an old topic and it may be why I did not receive any response, which I badly hope to hear. To provide a background, I'm trying to write a Policy extension to add a new WS-security implementation to WS binding in Tuscany. In my earlier post (see http://www.mail-archive.com/[email protected]/msg15333.html), I described the problems with Tuscany 1.6.x Policy extension framework using PolicyInterceptor approach and got some pointers from Simon, one is to try Tuscany 2.0 and the other is to use the PolicyHandler extension in Tuscany 1.6.x. I did not get any response for my request for help on the 2.0 Policy issues (see http://www.mail-archive.com/[email protected]/msg15353.html), so I went ahead down to the second path Simon suggested - to work with PolicyHandler extension point in 1.6.x. Everything went well on the reference side until I was testing on the service side. I think the current PolicyHandler extension framework has a defect when processing the PolicyHandler.afterInvoke() call for the service side and the model does not fit the need to add WS-security to the binding.ws.axis2 binding implementation. Today PolicyHandler.afterInvoke() is call with the OMElement response and MessageContext inMC, which is still the message context for the inbound request. In order to add WS-security to the response, I need to access the message context for the outbound response. So I modified Axis2ServiceInOutSyncMessageReceiver and Axis2ServiceProvider classes to delay the calling of PolicyHandler.afterInvoke() from inside Axis2ServiceProvider.invokeTarget() before outMC is constructed to after outMC has been constructed in Axis2ServiceInOutSyncMessageReceiver.invokeBusinessLogic(). After the change, my prototype worked and it also passed all unit tests. Now I would like to hear Tuscany developers' opinions and/or get your blessing. If this change is fine, how can it be pulled back into the code base? If not, what can be done to support what I would like to do, which is pretty natural and generic. I would really appreciate responses so our project can make a decision. Thanks in advance, Gang PS: The following is the code segments to high light the change I made (with "Gang: ..." comments): Axis2ServiceInOutSyncMessageReceiver.invokeBusinessLogic(...) { .... OMElement responseOM = (OMElement)provider.invokeTarget(operation, args, inMC); /* for ( PolicyHandler policyHandler : policyHandlerList ) { policyHandler.afterInvoke(operation, args, inMC, responseOM); } */ SOAPEnvelope soapEnvelope = getSOAPFactory(inMC).getDefaultEnvelope(); if (null != responseOM ) { soapEnvelope.getBody().addChild(responseOM); } outMC.setEnvelope(soapEnvelope); outMC.getOperationContext().setProperty(Constants.RESPONSE_WRITTEN, Constants.VALUE_TRUE); // Gang: Added for ( PolicyHandler policyHandler : policyHandlerList ) { policyHandler.afterInvoke(responseOM, outMC); } ... } Axis2ServiceProvider.invokeTarget(...) { ... // find the runtime wire and invoke it with the message RuntimeWire wire = ((RuntimeComponentService)contract).getRuntimeWire(getBinding()); Object response = wire.invoke(op, msg); // Gang: Commented out // for ( PolicyHandler policyHandler : policyHandlerList ) { // policyHandler.afterInvoke(response, inMC); //} return response; } Classification: UNCLASSIFIED Caveats: NONE
