Hi Blaise,

First a question: How does the algorithm react when I use Department as the root of my graph? If I understand correctly, it won't create an "orphan" element in the XML. How does the algorithm react here? Does it produce an unclosed graph? If not, where are the non-contained addresses serialized?

Thanks for the description of the use-case. My first reaction is that I'm not so sure how much it matters that the tech root is "unexpected": the main thing is that we produce something that can be described in a WSDL, and validates against it's XSD. This can be achieved with by adding a tech root. And as the model gets more and more complex, won't the orphan properties that collect at the root start making it look more and more like a technical root, that is, the presence of the orphans will become just as unexpected?

Maybe you should describe what you think the expectations are that are coming from these other frameworks? The expectation from a WebService is that the payload can be validated against the WSDL, right? And having a schema, we can also generate JAXB, so integration with JAVAEE is also covered. I also think that as soon as we are dealing with references to objects that don't have @XmlID, then we are losing the semantics of the graph anyway (until the JAXB integration solves this for us!)

Actually, I think SDO-66 is orthogonal to the current discussion. SDO-66 is all about adding to a the metadata used in a data graph, to make the model usable in different scenarios. Let me describe where I think SDO-66 can be used to solve some of the problems associated with containment, especially if you combine them with the Oracle's proposal for defining keys as part of the SDO metadata. I've been playing with our JPA integration lately, and I've been wondering if, in a world where we define some objects as having keys, the requirement of having datagraphs be closed isn't excessive. In a complex data model, we're always going to have to decide which paths to follow, and where to trim the graph. Is it necessarily a problem to use a key that is not resolved in the graph? Image a model like this: Customer has Orders, Orders have parts, Parts have Suppliers, Suppliers have Addresses, Addresses have Countries. For a particular WebService, we may need the customer orders, and for every order the part, but not for every the supplier. In this case, it might be reasonable to ship the PK of the supplier to the client, and if he wants, he can follow the link (by using a DAS, or whatever). Or take the list of countries. Is it really efficient to return this list with every call that returns an address? Won't we want to return only the country's PK as part of the graph, and leave it to the application to manage the cache of countries.

What we need for this is SDOs that have primary keys, and we need to say that a type is, in the SDO-66 sense, compatible with its key type. Then, when I project from my JPA model (which has the complete domain), into my XSD model (which has some complex objects replaced by keys), I can produce XML that is pruned in all the right places. Actually, I think the approach is not that different from what AquaLogic does. When you go over a webservice, you have to replace relations that are there in the domain model with keys, and you leave it to the application to call the DAS and follow the keys, as needed. The difference is, that AquaLogic sees the boundaries as being part of the type definition, with SDO-66 the boundries are not part of the type definition, but added into it, to support specific scenarios.

This should be contrasted with the proposals for SDO-124, both these issues effectively serialize the complete transitive closure reachable from the root of the data graph. This is not a criticism of the algorithms, it's just pointing out how they differ from SDO-66.

But it does impact a little on the discussion of the use case. If there's a non-SDO client there that cares about the shape of the XML generated by the SDO service, then, to me, it's very unlikely that the client will be satisfied with the schema produced by either of the SDO-124 proposals. In order to bring the graph into an acceptable form, you're probably going to have to feed in the expected form via SDO-66, and project into it. That is, I don't think we can solve your second use case with either algorithm. This makes, to my mind, the use-case where SDO is on both sides of the web service the more compelling one.

Ron

------------------------------------------------------------------------------

Von: Blaise Doughan [mailto:blai...@oracle.com] Gesendet: Mittwoch, 23. April 2008 18:10 An: sd...@lists.oasis-open.org Betreff: Re: AW: [sdo] Containment discussion Hello All,

XML serialization refers to two concepts in SDO: 1. The XML representation used when a data object goes through Java serialization (see SDO 2.1 section 6 "Java Serialization of DataObjects"). Here we are sending data objects from one SDO environment to another SDO environment. In this arena we can focus on making the transfer of data as efficient and portable as possible (which may require Xcalia's technical root). 2. The XML representation used when a data object is marshalled by XMLHelper. Here we are sending a message that may or may not be received by an SDO client. This is the compelling use case, as it relates to how SDO interacts with other technologies (Web Services, Java EE, .Net, etc.). In this case if we marshal/save an employee data object it would be unexpected to have the result wrapped in a technical root. Here is where the Oracle proposal (SDO-124) fits.

Oracle Proposal (SDO-124) and SDO 2.1

SDO 2.1 allows the current metadata: Department --containment --> Address Department --containment --> Employee Employee --non-containment --> Address

With this SDO 2.1 compliant metadata it is possible to create a graph that can not be serialized. All you need to do is to specify an instance of Address on an instance of Employee that is not containment by an instance of Department. The algorithm in the SDO 2.1 spec requires that you specify enough containment relationships (and obey them). The Oracle algorithm (SDO-124) makes the assumption that if you specify containment relationships you are specifying enough (just like SDO 2.1), in addition if no containment relationships are specified then there is special treatment for them.

Comparing the Containment Proposals

Oracle Proposal (SDO-124) One of the interesting aspects of the Oracle proposal (SDO-124) is that XML messages can be sent relative to a particular SDO type. This is very useful for DAS implementations that want to have per type find operations and then send this as an XML message. If the underlying metamodel had a containment relationship between Department and Address then the Oracle proposal (SDO-124) could produce an XML message wrt Department and another XML message wrt Employee.

Xcalia Proposal The above use case is not addressed by the Xcalia proposal.

SAP's proposal (SDO-66) The SAP proposal (SDO-66) does address this use case, my understanding of that proposal is that a set of metadata would need to be defined per root type and stored in its own instance of HelperContext. DAS would realize data in a HelperContext, and then project that data into another HelperContext that corresponds to the root type of the query. The onus would be on the user to ensure that the types between the HelperContexts are compatible.

-Blaise

Barack, Ron wrote:

Hi,

I'd like to revive this mail thread. I think the first question has been answered in Xcalia's document. To me, the compelling use-case is that we have two SDO based applications that want to communicated. Neither cares about specifying an XSD to which the message should conform, they just want to send SDOs over an XML wire.

That leaves the second question... now, the question is specifically to Oracle. Technical Root can transform *any* graph to XML. I think there is no tuning that can be done to the Oracle proposal to achive this, because Oracle's proposal tries to do something that is impossible: it tries to add properties to the basic structure to achieve closure *based on examining the metadata*. TechRoot does something else... it works directly with the data in the graph, not with the metadata.

That being the case, what are the reasons for preferring Oracle's algorithm over TechRoot?

Ron

------------------------------------------------------------------------------

Von: Barack, Ron [[1]mailto:ron....@sap.com] Gesendet: Mittwoch, 9. April 2008 00:33 An: [2]sd...@lists.oasis-open.org Betreff: [sdo] Containment discussion Hi Blaise and Xcalia

Two questions...

1. Before we embark on a long discussion of this (or any other) proposal for dealing with containment, I think it's first important to get our use-cases straight. Which use cases are you aiming at here, and why is the functionality provided by issue 66, and in particular the convenience methods based on it, not the best means to deal with it. As I understand your use-case, it seems to fit very well indeed to issue-66.

2. This propoal does not generate a closed XML document from an arbitrary SDO data graph. There are other algorithms that do, for instance, Xcalia's TechnicalRoot proposal. I would think that the main requirement of such an approach to containing is that it can produce XML out of any data graph. What is the advantage of prefering this approach to Technical Root?

Best Regards, Ron

--------------------------------------------------------------------- To unsubscribe from this mail list, you must leave the OASIS TC that generates this mail. You may a link to this group and all your TCs in OASIS at: https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php

References