On 3 févr. 09, at 18:05, Tudor Girba wrote:
Hi Simon,
iPlasma exports for Moose/Meta/FAMIX 2.1, and the file is incompatible with Moose/Fame/FAMIX 3.0 because of two reasons:
- because the MSE syntax changed slightly
- because FAMIX changed
The syntax of MSE is stable now. When we will get a stable version of FAMIX, I will work to create another exporter from iPlasma.
I'm cool with that, but do you have a date in mind?
Still I think it will be good to use the FAMIX2/FAMIX3 convention suggested below to distinguish between both model in MSE, simply to avoid name mismatch. It will be a minor fix and for older importer, a batch replacement of FAMIXx by FAMIX is not very difficult.
PS: my personal opinion is that the Famix3 implementation in SqMoose is now usable, although some tests are still red (more related to Moose than to Famix) and there should still be some bugs in some place. (disclaimer :) currently the Smalltalk importer does not infer previous/next attributes in Famix Association)
So it's not an official announcement, but we will begin to use it for our own work.
Cheers, Doru
On Feb 3, 2009, at 5:59 PM, Simon Denier wrote:
Hi everybody
I'm trying to use iPlasma to parse Java source files, export the model as MSE, then import them in SqMoose using Fame.
I ran into a couple of problems when doing this, specificall the MSE format built with iPlasma is not well parsed by Fame.
Here is a sample generated by iPlasma:
Beginning of MSE file:
(Moose.Model (entity
A) According to the MSE spec, the file cant begin directly with a name like that, should be ((Moose.Model (entity (....)) Besides, I dont see the point of this header, it works fine without it.
(FAMIX.Namespace (id: 6) (name 'java::lang') )
B) Fame infers accessor 'smalltalk-style', ie it will call #name: to set the name of the Namespace entity in the example above. It works in Famix3 but not in Famix2, where the setter is declared as #setName: . It's alright with the inference rule of Meta (as I far as I remember) but not with the new one. We should add those accessors in Famix2 for seamless access.
(FAMIX.Class (id: 3) (name 'Object') (belongsTo (idref: 6)) (isAbstract false) (isInterface false)
C) Fame does not recognize the idref: syntax, only the ref: syntax. The MSE spec is inconsistent about that: the grammar only defines the ref: syntax as Fame does, but the doc talks about both an IDREF command and a 'REF command for metamodels only'.
D) Finally, the biggest hurdle is that we dont know directly with which version of Famix the file was created. For example, the FAMIX package in SqMoose matches Famix3, and Famix2 is defined in a FAMIX2 package. This leads to a mismatch when trying to load the above file which was generated with Famix2 (unless batch editing all FAMIX -> FAMIX2)
My suggestion is that each Famix metamodel should explicitely declare its version in its Fame package name, either Famix2 or Famix3, like that: FAMIXClass class>>annotation <MSEClass: #Class super: 'FAMIX3.Type'> <package: #FAMIX3> ^self
Does it sound ok?
-- Simon
Moose-dev mailing list Moose-dev@iam.unibe.ch https://www.iam.unibe.ch/mailman/listinfo/moose-dev
-- www.tudorgirba.com
"There are no old things, there are only old ways of looking at them."
-- Simon