One idea would be to create some small language where one can:

- define "modules": a module is a group of something (MouseGroup). Modules could be defined from name patterns (e.g. "Test*") ; scope membership (org.eclipse.core.*) ; entity type (Methods, Class, ...) ; ...

- define relationship between modules: the 4 Famix relationships spring to mind (Access, Invocation, Inheritance, Reference) + possibly a generic Dependency (union of the 4). One could also think about intermediate generalization (e.g. TypeAccess=Inheritance+Reference ; MemberAccess=Access+Invocation ; ...) or more specialized relationship (e.g. implements=inheritance of interface ; extends=inheritance of class ; ...)

- define rules on modules and relationship: logical expressions on relationship between modules (e.g. "only M1 can-invoke M2": only methods of module M1 can invoke methods of module M2)


We would rather not implement all this, because it can go very far.
For example, there are people here working on matching a FeatureModel to an existing system to transform it into a Software Product Line.
This could possibly be handle by this language with arki behind checking the rules ...

So our line of thought was: What is the smallest amount of effort we can spend to have something usable and extensible ?

The next step after that could be to be able to change the description of a given architecture and have a tool showing us how to implement it in the source code and helping us to do it  ...

nicolas


De: "Usman Bhatti" <usman.bhatti@gmail.com>
À: "Moose-related development" <moose-dev@iam.unibe.ch>
Envoyé: Mardi 12 Juillet 2011 23:17:15
Objet: [Moose-dev] Re: Famix ADL

On Tue, Jul 12, 2011 at 5:51 PM, Tudor Girba <tudor@tudorgirba.com> wrote:
Hi Usman,

On 12 Jul 2011, at 16:38, Usman Bhatti wrote:

> Hi all,
>
> We are working to develop an Architecture Description Language (ADL) in Moose. The objective of the development is to define a language that allows to specify different components of an architecture so that these entities can be manipulated directly (analysis, visualization, etc). The architecture definition will be used to check rule conformance, for example. However, we would not want to restrict ourselves to any particular usage of the ADL.
>
> Today, we have implemented a preliminary version of ADLFamix by implementing modules that actually contain MooseGroups. Based on these modules, we can now write Arki queries for rule-checkling, for example. Now, we are contemplating about the next step because different people implement different things in an ADL. Some describe rules that specify connectors that can exist between modules. However, this approach ties connectors to rules and we cannot define a connector without any rules associated. Connectors can be defined separately or these can also be inferred from Famix associations of the Famix entities contained in modules. Also, rules can be built into modules so that each one has its own repository (however, it is not always possible to associate a rule to any particular module).

Do you have code snippets?

Here you can find some code snippets but todays task was more of a reflective effort to understand and assess things to implement in Moose. 
Gofer new
squeaksource: 'DelaunayTmpStuffs';
package: 'ADLFamix';
load.
 

Btw, I worked with Andy Kellens a while ago on getting Soul to work with Moose / Pharo. We got pretty far and got a prototype working, but it did not get any further. Perhaps it would be an interesting thing to look into this, because then we get the reasoning mechanism for free.

In any case, you do not want your ADL to have anything to do with FAMIX.

What is of interest here is to see how the following scenario can be implemented with such an ADL:
only A can-create B: only classes declared in module A can create objects of classes declared in
module B.

For the above scenario, we can construct two modules according to user specification (can represent 2 groups of packages or classes). So, modules are aggregations of Famix (Named?) entities. But the module still remains independent of any particular Famix abstraction: for example, a module can represent all class in a package com.example.mine.* or methods in all classes having a particular name.

Now, I am not sure if Andy's rule engine would be useful, but we've found that Moose API can be sufficient for describing rules and Arki for specifying the above-mentioned scenario and checking it.

It should define constraints based on any input objects. Maybe I did not understand what you are doing, but why do you need MooseGroups? These are specific to Moose, but you should not necessarily need them for describing your constraints.

MooseGroup is only used to specify famix entities encapsulated inside a module. Now, we need to turn our attention to other entities in an ADL (connectors and rules) so that the ADL remains generic and doesn't get tied to the task of rule checking. 

thanx
 

Cheers,
Doru


> The purpose of the mail is to get feedback from the people on the group about an ADL in Moose and its features. We are thinking in terms of, but not limited to:
>
> 1) fundamental features (modules, connectors, rules, ??)
> 2) objectives: rule checking, architecture inference (e.g. reconstructing plug-ins from java models in moose), ??
> 3) ??
> thanx
>
> Moosecians @ RMod
>
> _______________________________________________
> Moose-dev mailing list
> Moose-dev@iam.unibe.ch
> https://www.iam.unibe.ch/mailman/listinfo/moose-dev

--
www.tudorgirba.com

"It's not how it is, it is how we see it."


_______________________________________________
Moose-dev mailing list
Moose-dev@iam.unibe.ch
https://www.iam.unibe.ch/mailman/listinfo/moose-dev


_______________________________________________
Moose-dev mailing list
Moose-dev@iam.unibe.ch
https://www.iam.unibe.ch/mailman/listinfo/moose-dev