Answers inline...
Richard Jacob wrote:
OFF5997A80.C7B3B6A8-ONC125765D.003BB4BF-C125765D.003D7203@de.ibm.com" type="cite">Hi Kevin,
I agree with your observation here.
As far as I understood your problem description this interop issue arises when portlets start to throw events for which they didn't declare any type definition neither in the WSRP description nor in the JSR deployment descriptor (which both specs allow).
Yes.
OFF5997A80.C7B3B6A8-ONC125765D.003BB4BF-C125765D.003D7203@de.ibm.com" type="cite">Let it try to describe it in my words to see if we have a common understanding:
For the JSR case not specifying the type possible because the container would just pass around the serializable object and portlets knowing the type could cast explizitly and they are done, right?
Yes, or the receiving portlet can examine the event payload object to determine its type. For example, a portlet expecting a number as a payload could accept an Integer object, a String representation of the number, a Long, etc. This flexibility becomes much more useful when the portlet is designed to be reusable with other portlets, and applies especially in the case where event aliasing may be used to wire portlet events together.
OFF5997A80.C7B3B6A8-ONC125765D.003BB4BF-C125765D.003D7203@de.ibm.com" type="cite">Now as soon as we go over the wire we can serialize the object because it is either a JAXB-annotated complext type or an implicitly supported simple type.
So serialization just works.
Yes, BUT- there is no "standard" way to turn the implicitly-supported simple types into XML, which is what I would like to see if there is interest in defining for interoperability.
OFF5997A80.C7B3B6A8-ONC125765D.003BB4BF-C125765D.003D7203@de.ibm.com" type="cite">Now, when the event is distributed as the WSRP payload (with no type annotation), then the producer/consumer which want to deserialize that payload can not reliably do so, because the root element qname does not provide any hint of the type. And since the type in this case is not specified in the deployment descriptor this should fail....
Right- since the type was not declared in the event descriptor in the serviceDescription, by WSRP spec the consumer must treat the payload as opaque. I would like to change that slightly, to allow for some standard, universally-recognized simple types, such that when the event is received and has one of these special standard simple type event payload types, we all know what it is and how to deserialize it. In my opinion, the payload's root element name should not be used for any sort of type determination (since there can be complications with aliases); only the event payload type field should be used in the payload type determination.
OFF5997A80.C7B3B6A8-ONC125765D.003BB4BF-C125765D.003D7203@de.ibm.com" type="cite">I guess this is not only a problem for simple types (which could be solved e.g. in a manner you proposed) but also for complex types.
How would you point JAXB to the right annotated class allowing to create the complex bean object?
Yes, this is not just a problem for the simple types, but I think it can be handled for the complex types by configuring the JAXB context with a classpath pointing to all JAXB-bound event payload objects. JAXB then determines which type the XML is and what it should be unmarshalled as using the schema information.
OFF5997A80.C7B3B6A8-ONC125765D.003BB4BF-C125765D.003D7203@de.ibm.com" type="cite">I think this problem goes away if portlets in the JSR case always declare the type definition in the portlet.xml.
Maybe this is also worth mentioning as a best interop practice, even in the local case.
Yes and no... If I wanted to create a "flexible" portlet that displayed a map based on location (such as a US postal code / ZIP code) that is sent in via an event, it is easy to have the portlet code figure out whether the event payload is a Integer, Long, Short or String representation of the zip code so that it could accept any of those types. But if it is forced to declare a single type (and only a single type is allowed in the event declaration by both JSR286 and WSRP 2.0) then this cannot be done.
In addition, if we had two portlets (each on a different company's producer) which wanted to interoperate based on an event containing a simple String payload, they each declare the event payload type as java.lang.String in the JSR286 portlet.xml, but unless there is some standard WSRP 2.0 mechanism for describing the type java.lang.String (such as "xsd:string"), there might still be interoperability issues because each producer might marshal the String in its own way that the other producer doesn't recognize. This is a problem because you can't just give JAXB a String and have it marshal it, the String needs to live inside another JAXB-annotated Java object.
OFF5997A80.C7B3B6A8-ONC125765D.003BB4BF-C125765D.003D7203@de.ibm.com" type="cite">Another option besides definiting root element Qnames indicating the type could be to have the event flow with a type qname, which the spec alread has.
So if you would have a simple string being serialized, the event would flow as (simplified):
<event name="someNS:localPart" type="xsd:string">
<eventPayload>
<someRootElementName>StringValue</someRootElementName>
</eventPayload>
</event>
This way the wsrp-to-jsr bridge could pass the correct type class to the JAXB deserializer and retrieve the correct object (provided it has a type to class mapping at hand).
Exactly- this is exactly what I am proposing-- a standard set of "simple" event payload types, like "xsd:string" and "xsd:int" where we then know that the payload is serialized as some root element which contains the simple type. The name of the root element should not matter, but we could declare a standard set of payload root element names for the simple types if we wanted, which would make it a bit easier to configure JAXB for marshalling / unmarshalling these simple types.
Kevin
OFF5997A80.C7B3B6A8-ONC125765D.003BB4BF-C125765D.003D7203@de.ibm.com" type="cite">
Mit freundlichen Grüßen / Kind regards
Richard Jacob
Team Lead Portal Standards & WCM development
WSRP Standardization Lead
IBM Software Group, WPLC
WSS Websphere Portal Foundation Development 1
Phone: +49-7031-16-3469 IBM Deutschland
E-Mail: richard.jacob@de.ibm.com Schoenaicher Str. 220
71032 Boeblingen
Germany
IBM Deutschland Research & Development GmbH / Vorsitzender des Aufsichtsrats: Martin Jetter
Geschäftsführung: Erich Baier
Sitz der Gesellschaft: Böblingen / Registergericht: Amtsgericht Stuttgart, HRB 243294
From: Kevin Frender <kevin.frender@oracle.com> To: Nader Oteifa <nader2@netunitysoftware.com> Cc: wsrp-interop@lists.oasis-open.org Date: 10/15/2009 05:45 PM Subject: Re: [wsrp-interop] WSRP event payloads with "simple" types
Hello Nader,
I may be misunderstanding your suggestion, but my original statement of
the problem wasn't entirely clear either.
Yes, I know that JAXB can be set up so that arbitrary objects can be
marshalled and unmarshalled (as DOM elements), but the problem as I see
it is that for the simplest of data types explicitly supported by the
JSR286 spec as event payloads, there is no obvious, inter-operable way
to marshal these event payloads so that they will be guaranteed to show
up in the receiving portlet as the same simple type that was sent.
For example, if portlet A sends an event:
eventResponse.setEvent(new QName("namespace", "localpart"), "Simple
String payload");
and portlet B receives this event, I would expect portlet B to be able
to do this:
Event event = eventRequest.getEvent();
String payload = (String)event.getValue();
-- and not need to do anything like this:
String payload = ((Element)event.getValue()).getTextContent();
The problem isn't that I don't know of any way at all to marshal the
simple String payload into XML, or how to unmarshal that form back into
a String, but rather that I don't know how to distinguish between an
event payload coming from another (third-party) WSRP producer or
consumer that represents a marshalled simple payload type (like a
String) versus an event payload that was deliberately configured by the
portlet developer to be marshalled into XML. For example, if I get an
event payload in the WSRP event that looks like this:
<samplePayload>Simple String payload</samplePayload>
I could assume that this means I should return an Event object to the
portlet developer whose payload is a java.lang.String object, as
outlined in the sample above. However, if the portlet developer for
event-sending portlet "C" actually created their own event payload
object and gave it a valid JAXB binding, such as:
@XmlRootElement
public class SamplePayload implements Serializable
{
String value;
public SamplePayload(String s)
{
value = s;
}
}
eventResponse.setEvent(new QName("namespace", "localpart"), new
SamplePayload("Simple String payload"));
Then the developer actually intended for a "SamplePayload" object to be
sent, not a java.lang.String-- and a receiving portlet should probably
get an XML DOM element in the event payload instead of a simple
java.lang.String object (assuming that the receiving portlet's WSRP
implementation doesn't have a JAXB SamplePayload object configured).
So my proposal would be a set of extension elements for containing only
the simplest of event payload values-- the simple types explicitly
supported by the JSR286 specification-- so that various WSRP
implementations could recognize those elements as containing a simple
payload value (and ONLY a simple payload value). I would propose that
arrays of simple values would not be supported, as they are not
explicitly supported by the JSR286 spec as payload types, but arrays of
simple values can still be handled by JSR286 portlet developers by
specifying their own JAXB bindings.
For example, we might define in the wsrp-extra namespace elements such as:
<element name="stringPayload" type="xsd:string"/>
<element name="intPayload" type="xsd:int"/>
<element name="longPayload" type="xsd:long"/>
...
By being in the WSRP namespace (and being well-known elements) it would
be easy to distinguish these "simple" event payload types from "complex"
marshalled objects. I can create such schema definitions for sending
simple payload types myself and put them in our proprietary namespace,
and they would work just fine as long as the events were not delivered
to a third-party WSRP producer, which wouldn't recognize them as
intending to be "naked", simple values.
So that is the point of my post- for at least the simplest of event
payload types it seems like a WSRP-standard solution for encoding these
payload types would be beneficial for interop purposes.
Kevin
