Hi, On 23 Sep 2011, at 22:15, Stéphane Ducasse wrote: I do not understand. Do you refer to the name of the MetaDescription? If yes, that is because the qualified name is 'FAMIX.Attribute', given that it belongs to the FAMIX package. So, there is no need to prefix the class name. Ok, this is pretty much crappy :). #FAMIXStructuredEntity will perform a lookup of the class in the Smalltalk image. 'FAMIX.StructuredEntity' will perform it in the Fame meta repository. We should only use symbols because they are also affected by refactorings. I will create a rule :).

No. It should be: MSEClass: 'Attribute' super: #FAMIXStructuralEntity

MSEClass: 'Attribute' super: #FAMIXStructuralEntity

I know. That is why I said I did not like it (=crappy). We had this argument before, but Adrian wanted it this way, and it remained like this. The reason is that sometimes (in 0.001% of cases) you might have an entity that is only defined in Fame but not directly in Smalltalk. In that case, you want to look it up in Fame, not in Smalltalk. Now we can change it and only provide the Smalltalk based lookup.

You have the have the fully qualified name, not just the name of the meta description (you might have two in two distinct packages).

Fame already provides a lookup based on the fully qualified name, so we can safely use that.

> MSEClass: #Attribute# super: #StructuralEntity package: #FAMIX

See the above explanation. It should :).

#FAMIXStructuralEntity points to the name of the Smalltalk class. There is no other way around it.

This is the most robust way of referring to the superclass because if we rename the Smalltalk class, the pragma will be updated as well (it's a symbol).

So, everywhere we have superclass: we should have the name of the Smalltalk class.

Now, that we are at it, we should introduce also only FM3Class: pragma and rely on the default Smalltalk superclass information.

Like this we do not have to maintain it at all. Again, this is something I argued a while ago, but did not get introduced. Cheers, Doru

Hi Stef, I will reply in one batch.

Having meta descriptions is nice and powerful, but at least in Moose, we want to have a 1-to-1 mapping between the meta-description and implementation. That means that we will have one Smalltalk class corresponding to one Fame class. Also to make it simple, in most cases, we will have the superclass relationship kept in the same area. Suppose you have: SmalltalkClassA <-> FM3MetaDescription(P.A) SmalltalkClassB <-> FM3MetaDescription(P.B) And now, SmalltalkClassB superclass = SmalltalkClassA. In this case, the superclass of FM3MetaDescription(B) should be FM3MetaDescription(A). This is desirable in 99.99% of the cases. So, this is what we should optimize for. The only case we have until now where we do not want this, is in FAMIXEntity which has as superclass Object, instead of the implementation class MooseEntity. This is because we want to be able to generate FAMIX in Java without carrying the MooseEntity class with us. Now, let's go to our original example. Currently, you can define it in two ways: 1) SmalltalkClassB class>>annotation <MSEClass:'B' superclass: 'P.A'> <package: 'P'> 2) SmalltalkClassB class>>annotation <MSEClass:'B' superclass: #SmalltalkClassA> <package: 'P'> In our case, we do not want 1) (even if you write it like superclass: #A package: #P), because the issue is that if I later rename 'A' to 'X', the string in class B will not get refactored. But, in 2) I refer to #SmalltalkClassA, which is more robust. Furthermore, looking for the references in of SmalltalkClassB will get me to the pragma, which is another benefit. These are especially important when you build a new meta-model. I am not worried at all about mixing meta-levels. It is actually a way to enforce the practice of having a 1-to-1 mapping. In any case, the pragmatic solution is much more important. Btw, we currently have 108 subclasses of MooseEntity. So, all in all, we should point to Smalltalk implementations as much as possible. I like the idea of having another pragma that will be used for pointing to Fame strings, but its use should be discouraged in most cases. However, except when we want a different inheritance we should simply use: 3) SmalltalkClassB class>>annotation <FM3Class:'B'> <FM3Package: 'P'> In most cases, we do not want to get rid of the name because we have packages. So, FAMIXClass will correspond to FAMIX.Class in Fame. This is why we probably still want to define explicitly both the name and the package of the Fame class. But, maybe we can find a cleverer scheme. Does this answer your questions? Cheers, Doru On 24 Sep 2011, at 22:43, Stéphane Ducasse wrote: -- www.tudorgirba.com "In a world where everything is moving ever faster, one might have better chances to win by moving slower." _______________________________________________ Moose-dev mailing list Moose-dev(a)iam.unibe.ch https://www.iam.unibe.ch/mailman/listinfo/moose-dev