multiple inheritance in Famix/Fame
Locked
multiple inheritance in Famix/Fame
 
|
I believe I already said here that I don't like very much the "default" FAMIX metamodel in Moose because it is the union of Smalltalk and Java specificities. For 2 languages it works (more or less), but if we want to add C#, C++, and whatever-else (not talking about non OO languages), it quickly starts to be a huge mess. It would be preferable to have a generic metamodel and specific ones for each languages. But then, we will run into cases where we want to have some kind of multiple inheritance. For example we could want to say that a Java method and a C function are behaviouralEntities with return type and Java methods and Smalltalk methods are behavioural entities with a received (when they are called) and Java method now have two superclasses. So the question arise why Fame does not have multiple inheritance? - no specific reason? - fundamental design decision? - historical reason? nicolas -- Nicolas Anquetil -- RMod research team (Inria) _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: multiple inheritance in Famix/Fame
|
|
Hi, We do not want to use multiple inheritance as a design tool. We should use Traits instead, as we know it scales better. Alain Plantec already did a prototype for having Traits in Fame, but it was not integrated. It would be great if someone would look into this. The only thing to keep in mind is that we would have to also update the Java version of Fame.
Cheers, Doru On Thu, Mar 7, 2013 at 2:14 PM, Nicolas Anquetil <[hidden email]> wrote:
"Every thing has its own flow"
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: multiple inheritance in Famix/Fame
|
|
On 03/07/2013 02:45 PM, Tudor Girba
wrote:
do we? :-) We were discussing this over lunch with Damien Cassou (our trait expert) It seems to me traits are a behaviour reuse mechanism, Famix is mostly about typing and structure. Maybe conclusions drawn from programming languages implementation do not apply to models. I don't think models are just another kind of programming language, they were primarily intended to be conceptual tools that could be derived to programs. A programming language must be implementable on a computer (or fail to present much interest). A model must be mostly transformable into a program, but some manual tweaking seems acceptable. nicolas
-- Nicolas Anquetil -- RMod research team (Inria) _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: multiple inheritance in Famix/Fame
|
|
Yes, we do :). A model without the appropriate behavior has little value. For example, we want to have traversals next to the model. It is for this reason that we should use Traits.
Cheers, Doru On Thu, Mar 7, 2013 at 4:05 PM, Nicolas Anquetil <[hidden email]> wrote:
"Every thing has its own flow"
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: multiple inheritance in Famix/Fame
|
|
In reply to this post by Nicolas Anquetil
On Mar 7, 2013, at 2:14 PM, Nicolas Anquetil <[hidden email]> wrote: > > I believe I already said here that I don't like very much the "default" FAMIX metamodel in Moose because it is the union of Smalltalk and Java specificities. > > For 2 languages it works (more or less), but if we want to add C#, C++, and whatever-else (not talking about non OO languages), it quickly starts to be a huge mess. > > It would be preferable to have a generic metamodel and specific ones for each languages. > > But then, we will run into cases where we want to have some kind of multiple inheritance. > For example we could want to say that a Java method and a C function are behaviouralEntities with return type and Java methods and Smalltalk methods are behavioural entities with a received (when they are called) and Java method now have two superclasses. > > > So the question arise why Fame does not have multiple inheritance? simplicity at the time as reaction of MOF2.0 > - no specific reason? > - fundamental design decision? > - historical reason? > > nicolas > > -- > Nicolas Anquetil -- RMod research team (Inria) > > _______________________________________________ > Moose-dev mailing list > [hidden email] > https://www.iam.unibe.ch/mailman/listinfo/moose-dev _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: multiple inheritance in Famix/Fame
|
|
In reply to this post by Tudor Girba-2
On Mar 7, 2013, at 5:08 PM, Tudor Girba <[hidden email]> wrote:
I do not see why we could not use traits to make fame modular.
It depends and I do not see the point.
_______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: multiple inheritance in Famix/Fame
|
|
On 03/08/2013 08:10 AM, stephane
ducasse wrote:
I did not say I don't want to use it But I don't believe it covers all the needs. For me a model and therefore Famix is inherently typed because it is about describing things. For exmple I fighted with Usman and Guillaume because they were using FamixReference to represent an import between two programs when it should have been some other association. It worked, but it was not the right semantic. And models are essentially tools to describe semantics, not implementation. This is why I don't think traits are the solution in this case. They are a reuse mechanism, not a typing one.
The points is: we know that multiple inheritance is a pain in the back to implement in programming language. Yet, we also know that everything everywhere uses something that in lack of better image I would call multiple inheritance. So I would like to see if we can define some kind of multiple inheritance to describe things (fame/famix), and then some clever mechanism to translate this multiple inheritance into real program (pharo, java, whatever). nicolas
-- Nicolas Anquetil -- RMod research team (Inria) _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
Re: multiple inheritance in Famix/Fame
|
|
Hi Nicolas,
I definitely do not want multiple inheritance in Fame. I do not see what you can model with multiple inheritance and cannot model. I think the problem you seem to see comes from you not perceiving Traits as Types. But, they are types, just like classes are. And they should be captured as types in Fame as well. Perhaps we can even make the MetaDescription be a Trait by default and have no distinction between Class and Trait at the level of Fame. Furthermore, the pragmatic reasons for using Traits should not be ignored lightly: Traits exists as first class in Pharo, which makes it straightforward to map the TraitDescription to the actual implementation. This is a huge plus. So, I want to explore the Traits as an option before we discard it. And we already have an implementation, as I mentioned. Cheers, Doru On Mar 8, 2013, at 9:41 AM, Nicolas Anquetil <[hidden email]> wrote: > > On 03/08/2013 08:10 AM, stephane ducasse wrote: >> >> On Mar 7, 2013, at 5:08 PM, Tudor Girba <[hidden email]> wrote: >> >>> Yes, we do :). >>> >>> A model without the appropriate behavior has little value. For example, we want to have traversals next to the model. It is for this reason that we should use Traits. >> >> Yes >> >>>> >>>> We do not want to use multiple inheritance as a design tool. We should use Traits instead, as we know it scales better. >>> do we? :-) >>> >>> We were discussing this over lunch with Damien Cassou (our trait expert) >>> >>> It seems to me traits are a behaviour reuse mechanism, Famix is mostly about typing and structure. >> >> I do not see why we could not use traits to make fame modular. > I did not say I don't want to use it > But I don't believe it covers all the needs. > > For me a model and therefore Famix is inherently typed because it is about describing things. > For exmple I fighted with Usman and Guillaume because they were using FamixReference to represent an import between two programs when it should have been some other association. > It worked, but it was not the right semantic. > And models are essentially tools to describe semantics, not implementation. > > This is why I don't think traits are the solution in this case. > They are a reuse mechanism, not a typing one. > >>> I don't think models are just another kind of programming language, they were primarily intended to be conceptual tools that could be derived to programs. >>> >>> A programming language must be implementable on a computer (or fail to present much interest). >>> A model must be mostly transformable into a program, but some manual tweaking seems acceptable. >> >> It depends and I do not see the point. > > The points is: we know that multiple inheritance is a pain in the back to implement in programming language. > > Yet, we also know that everything everywhere uses something that in lack of better image I would call multiple inheritance. > > So I would like to see if we can define some kind of multiple inheritance to describe things (fame/famix), and then some clever mechanism to translate this multiple inheritance into real program (pharo, java, whatever). > > nicolas > >>>> >>>> Alain Plantec already did a prototype for having Traits in Fame, but it was not integrated. It would be great if someone would look into this. The only thing to keep in mind is that we would have to also update the Java version of Fame. >>>> >>>> Cheers, >>>> Doru > > -- > Nicolas Anquetil -- RMod research team (Inria) > > _______________________________________________ > Moose-dev mailing list > [hidden email] > https://www.iam.unibe.ch/mailman/listinfo/moose-dev -- www.tudorgirba.com "We are all great at making mistakes." _______________________________________________ Moose-dev mailing list [hidden email] https://www.iam.unibe.ch/mailman/listinfo/moose-dev |
«
Return to Moose
|
1 view|%1 views
| Free forum by Nabble | Edit this page |