[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Re: WG: [sdo] ISSUE 66: Moving Data Between Contexts
Hi Ron, Thanks for kicking off this discussion. I agree it's a very important issue. For question #1, I think the answer needs to be c) O1 becomes invalid/undefined. I think this is necessary to allow for the possibility of O2 being the same instance as O1. I think the most efficient implementation of cast() would return the same object if possible, just with the type changed. For example, the HelperContext.cast() method might do something like the following: DataObject cast(DataObject input) { Type inputType = input.getType(); Type resultType = getTypeHelper().getType(inputType.getURI(), inputType.getName()); // check if types are compatible ... // Unrelated question: do we need to recursively call cast() on the rest of the reachable objects ??? ... // change the type ((DataObjectImpl)input).setType(resultType); return input; } This may not be always possible, but if the spec doesn't say that the original object becomes suspended, there is no way we could ever implement cast this way - it would always require at least a wrapper DataObject. For question #2, I think that C1.cast(C2.cast(O1) should == O1. It certainly would be the case if cast() is implemented as above. If, on the other hand, a proxy is returned, I still think their can only be one DataObject instance representing the same object in a given context. Otherwise, we're talking about a copy - not a cast. Which brings me to a question. Do we really think that cast() should ever return a disconnected copy. In Java, cast and copy are very different - one changes the type, the other changes the instance. I think that for copy cross-scope copy we should allow/use cross scope CopyHelper. For example: O2 = C2.getCopyHelper().copy(O1); This would create a copy of O1 in the context C2 (the same compatibilty requirement as cast() would need to apply). O2 = C2.cast(O1): This returns the same conceptual object, just represented in the new metadata (type) scope. This way, cross-scope cast and copy are two different things with clear APIs for both. One more thought I had about cast() is that the argument to the cast method should maybe be Object - instead of DataObject. That way, you could use it to cast a POJO (or JAXB) object to a DataObject: DataObject cast(Object object) Now, if the input object is not already an instance of DataObject then the default mapping rules (maybe there are also annotations) are used to check compatibility, etc., before returning a DataObject proxy, for example. That's all for now. Frank. "Barack, Ron" <ron.barack@sap.com> 02/21/2008 05:20 AM To <sdo@lists.oasis-open.org> cc Subject WG: [sdo] ISSUE 66: Moving Data Between Contexts Hi Everyone, OK, here we go... defining the cast method. I think the concept is actually simple, but it was hard for me to write down without having to resort to this quasi-mathematical language. Sorry about that. Maybe, once the concepts are clearer, someone else can do a better job. Anyway, we decided at the F2F that is was a high priority issue, and to get started on it right away. Please spend a few cycles on this topic, so we can begin discussing it. I've been trying, but I don't think it's possible to describe this functionality without some definition of what it means for types to be compliant. So let me start out with the following definition, which I believe to be too restrictive, but which at least allows us to speak concretely about proposals. The concept can always be loosened up in a separate step. Definition: Compliance A data type DT1 is compliant with a data type DT2 if they have the same instance class A complex type T1 is compliant with a type T2. The types are compliant iff 1. The URI and name of both types are equal. 2. For every property P1 in T1 there is a property P2 in T2 such that a) P1.getName() is equal to P2.getName() b) P1.getType() is compliant with P2.getType() c) P1.isMany() is identical to P2.isMany() d) if P1.opposite() is non-null, the P2.opposite is also non-null. Of course, the key point is what is not looked at in determining compliance. This includes 1. The instance class (static SDO of the complex type) 2. Whether the value of P1.isContainment() is identical to the value of P2.getContainment() 3. The XML or Java names of the properties. I believe we will someday be able to loosen this definition, for instance, that the URI-Name pair does not have to match, or we can allow the properties in T1 to be a subset of those in T2. But we can already do some pretty useful things with the definition. ____________________________________ Definition: CAST Using this definition of compliance, I would like to define the following operation in the HelperContext interface DataObject cast(DataObject) with the following characteristics 1. If there is no type in the current HelperContext that is compliant with the type of the argument, an exception is thrown. 2. If there is such a compliant type, the object returned will have this type. 3. The values of data type properties in the returned object have the same value as the corresponding properties of the original data object. 4. The value of complex typed properties in the returned data object are data objects whose types are compliant with the type of the data object returned by retrieving the corresponding property on the original data object. This is maybe clearer using the following semi-mathematical formulation. Consider object O1, where O1.getType() is T1 and where T1.getHelperContext() is C1. Consider a context C2, having a type T2 that is compliant to T1. Now DataObject O2 = C2.cast(O1) For each property P1 in T1 there is a property P2 in T2 such that P1.getName().equals(P2.getName() ) If P1.getType().isDataType() is true then O1.get(P1)==O2.get(P2) If P1.getType().isDataType() is false, then if O1.get(P1) is null, O2.get(P2) is null, otherwise O2.get(P2)==C2.cast(O1.get(P1)) __________________________________________________________________________ Now for some details I'd like to discuss: 1. What is the state of O1 after casting to O2? In our implementation, and also in the description of the cast operation that I gave at the F2F, each data object (O1 and O2) was depicted as a "view" of the same underlying data. The actual motivation for this implementation was performance, that is, we wanted to avoid a copy step. This has the effect that both O1 and O2 are "live", that is, changes to O2 are immediately visible in the O1. This *could be* useful, but I would also say that the vast majority of time, either O1 or O2 is active, not both. I see three possibilities for what we say in the spec: a) We require both O1 and O2 to be live views of the same underlying data. b) We say that O1 is unchanged by operations on O2. This more or less requires a copy step, which would make the cast operation expensive. c) We say that the state of O1 is undefined. In order to "reactivate" O1, we would have to cast it back from O2, that is we would have to explicitly say O1=C1.cast(O2), which would make O2 invalid, but bring O1 back into a valid and consistant state. 2. Does cast create objects, even if there are the underlying data already has a DataObject in the context? Is C1.cast(C2.cast(O1)) == O1? Or, more fundamentally, does a data object have only one ?view? in a particular context. Imagine that the object that is casted is the root of a complex graph, and that there are multiple paths (i.e., through non-containment references) to some node in the graph: do we get the same object, no matter which path we take? I would say that all three of these behaviors are important. To achieve them, the mental model is that the underlying data representation ?knows? which data object represents in which context. 3. How is the value of O2.getChangeSummary() related to the value of O1.getChangeSummary()? Ron Von: Barack, Ron [mailto:ron.barack@sap.com] Gesendet: Montag, 18. Februar 2008 11:27 An: sdo@lists.oasis-open.org Betreff: [LIKELY JUNK][sdo] ISSUE 66: Moving Data Between Contexts http://www.osoa.org/jira/browse/SDO-66 Von: Barack, Ron [mailto:ron.barack@sap.com] Gesendet: Mittwoch, 13. Februar 2008 14:47 An: sdo@lists.oasis-open.org Betreff: [LIKELY JUNK][sdo] New Issue: Moving Data Between Contexts Motivation: HelperContexts were motivated principally by the need to allow JEE applications to maintain independent type registries, protected from possibly conflicting types that may be defined in other applications. In an SCA environment, we can imagine HelperContexts being associated with contributions, since these normally define the boundries of artifact resolution, which includes the XSDs, java classes, or whatever else is being used to define types. If we envision SDO being used as data transport in a SOA, for example, in SCA wiring, then it becomes clear that SDO must also provide some mechanism for moving data between contexts. This is, in my mind, the most common use-case for moving data between contexts. On the other hand, it is completely possible for a single application to want to manage multiple contexts, and to want to move data between contexts. In the discussion that follows, moving data between applications should be considered as a common use-case for a functionality that can also be used within a single application. More specifically along the lines of our charter, the ability to move data between contexts is the central concept behind our approach towards relaxing containment requirements. We are envisioning that some applications within the SOA environment will not have need for or interest in signifying some relationships as being containment, while other relationships are references. Other applications will require containment, perhaps because they will serialize to XML, perhaps because they use some other feature, such as ChangeSummary, CopyHelper, or EqualityHelper that relies on containment. If we are going to relax containment requirements, then we must either redefine these features (which would break backwards compatibility), or we must find a way to add containment information on the fly, so that this operations will still be available for data objects that come from contexts where containment does not make sense. This operation, adding containment information "on-the-fly", is, in this proposal, implemented as moving data objects from a context where the types are defined without containment information to a context where the types do have containment information. As previously stated, this could mean moving the data objects between applications, or between contexts in a single application. The cast() method, which I want to introduce with this issue, was discussed at the F2F, and identified as a high priority for SDO 3.0. For more information on where this is all headed (slowly, with baby steps) please see: http://www.oasis-open.org/apps/org/workgroup/sdo/download.php/26723/Containment%20in%20SDO%203.ppt and http://www.oasis-open.org/apps/org/workgroup/sdo/download.php/26711/SDO3%20Java%20Projections.ppt Description: Consider a DataObject O1, with Type T1, defined in context C1. Define an operation that will "cast" from O1 to a DataObject O2, with Type T2, defined in context C2. The data objects O1 and O2 would provide alternative views of the same data. The Types T1 and T2 may differ slightly, for instance, in how containment is defined. In order for casting to make sense, it will be necessary to define some concept of "compatibility" of types, that is, what must T1 and T2 have in common, in order for the cast operation to make sense. I would, however, like to leave the exact definition of compatibility for a later issue, and discuss the operation in principle first. The details of the relationship between O1 an O2, for instance, if changes to O2 are visible in O1, however, should be determined as part of the resolution of this issue. Proposed Assignment JIRA Components: Topic 10: Containment Topic 7: TypeSystem and Helpers
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]