Getter dispatch in case of default value in K3

[fcoulon]

For the case:

• An EClass containing an attribute foo
• An Aspect over this class declaring foo with a default value
  Can we consider the aspect has the priority?

It is breaking a test case in Melange. We load a model of merged Ecore + Aspect and try to access the attribute declared in the Aspect but we access the null value of the model.

k3/k3.eclipse/fr.inria.diverse.k3.al.annotationprocessor.plugin/src/fr/inria/diverse/k3/al/annotationprocessor/Aspect.xtend

Lines 800 to 817 in 5c35a83

	val gemocHackGetter = '''
	try {
	for (java.lang.reflect.Method m : _self.getClass().getMethods()) {
	if (m.getName().equals("«IF f.type.simpleName == "boolean" || f.type.simpleName == "Boolean"»is«ELSE»get«ENDIF»«f.simpleName.substring(0,1).toUpperCase() + f.simpleName.substring(1)»") &&
	m.getParameterTypes().length == 0) {
	Object ret = m.invoke(_self);
	if (ret != null) {
	return («f.type.type.qualifiedName») ret;
	}«IF !f.type.primitive» else {
	return null;
	}
	«ENDIF»
	}
	}
	} catch (Exception e) {
	// Chut !
	}
	return «PROP_VAR_NAME».«f.simpleName»;

[ebousse]

Are we talking about an EClass generated by Melange, where the foo attribute is automatically added with the aspect EAnnotation? Or are we talking about a user-made EClass which happens to have an aspect defining the attribute foo a second time?

• In the first case, I would suggest to set the default value in the generated Ecore file, if possible.
• In the second case, we would need to check if this is the first time that the attribute is being accessed, and if it is, we can indeed probably pick the aspect default value if the EMF value

[fcoulon]

I was talking of the first case but the second one is relevant too.

[ebousse]

So for the first case, the best would be to patch Melange to generate a more complete EClass, where the EStructuralFeature foo has its defaultValue set to the one that was written in the aspect.

But can this default value be accessed when the aspect is parsed?

[fcoulon]

I guess we have to deal with Expression for default value at the Xtend side
Maybe the Java compiler is smart enough to evaluate it

[ebousse]

Ah you are right, the expression might not be a primitive/serializable value...

In that case, yes you are right something smarter can be done in K3, such as checking only on the first access whether there is a default expression specified in the aspect.