[news.eclipse.modeling.mdt.uml2.uml] Re: Questions about code generation

["Kenn Hussey" <khussey@xxxxxxxxxx>]

Sunny,

Note sure why one would need to reverse-engineer the UML2 code generator 
(the source is openly available and the logic is in the templates), but I'll 
try to answer your questions as best I can... see below.

"Sunny" <sunfire001@xxxxxxxxx> wrote in message 
news:b2a5fbd6b836f32c4dc8173a3aaac382$1@xxxxxxxxxxxxxxxxxx
> Hi,
>
> I am trying to reverse engineer associations from the uml2's 
> implementation (based on UML2.0 spec. and uml2_1.1.1). I would like to 
> know the generation rules for its code:
>
> 1. in what situation a field is (or is not) generated for a navigable 
> association end in the implementation class? For example, a field 
> Classifier#generalization is generated for the association from Classifier 
> to Generalization (pp.26), but not for the association from ActivityGroup 
> to ActivityNode (pp.288)
>

The logic for whether to generate a field is fairly complex, but generally 
one is generated iff it meets the usual EMF criteria and such a field does 
not already exist in the class hierarchy, the feature is not a union, and 
the feature has not been redefined. See the source for 
org.eclipse.uml2.codegen.ecore.genmodel.impl.GenClassImpl#getDeclaredFieldGenFeatures() 
for details.

> 2. in what situation, a createRoleName() is generated (e.g., 
> Classifier#createGeneralization())?

These factory methods are generated iff the option has been turned on in the 
generator model, the feature is changeable, the feature is effectively a 
containment feature (or a subset of one) and the feature's class has at 
least one child class. See the source for 
org.eclipse.uml2.codegen.ecore.genmodel.util.isFactoryMethods(GenClass, 
GenFeature) for details.

>
> 3. is it true that only when the type of an association end is a subclass 
> of NamedElement, is a method getRoleName(String name) generated (e.g., 
> Classifier#getRedefinedClassifier(String name))?

No, these look-up methods are generated for any multiplicity-many reference 
feature which is typed by a type that has one of its features marked as a 
"key" in the generator model. See the source for 
org.eclipse.uml2.codegen.ecore.genmodel.util.getKeyGenFeatures(GenFeature) 
for details.

> 4. how are association ends subsetting other ends specially handled in the 
> generated code? (e.g., what kind of code will be generated if an 
> association end subsets another end)
>

This depends on whether the feature and its superset are multiplicity-many 
or not... Generally, the getter for a multiplicity-many subset is generated 
to use a special subset/superset list - see 
org.eclipse.uml2.codegen.ecore.genmodel.impl.GenClassImpl#getSubsetListConstructor(GenFeature) 
for details - whereas a scalar feature gets extra code appended to its 
setter to also add elements to the superset feature  - see 
templates/model/Class/setGenFeature.post.insert.javajetinc for details.

> 5. how are association ends redefining other ends specially handled in the 
> generated code?

The accessor for the other end is generally overridden to call the accessor 
for the redefining end.

> I checked the book "EMF Modeling Framework" and could not find all the 
> corresponding rules to generate the uml2_1.1.1's code. For example, it 
> only says that a field won't be generated for a volatile or non-changeable 
> reference? It says nothing about the handling for subsetting or 
> redefinition...

Of course not - all of the support for UML constructs like subsetting, 
unions, redefinitions, etc. isn't part of EMF, but rather is provided by the 
UML2 code generator. I apologize for the lack of documentation...

>
> Your help is much appreciated,
>
> -- Sunny
>
>