|
|
|
|
|
|
|
|
Re: intermediate classes in library? [message #1695948 is a reply to message #1695943] |
Wed, 20 May 2015 13:41 |
Ed Willink Messages: 7669 Registered: July 2009 |
Senior Member |
|
|
Hi
Your issue asked the question. It did not ask for a change. The
resolution answered the question in accordance with the prevailing
wording. And added:
"Future: if the dual Package/Class inheritance of Module is resolved to
make Library Package-like and
OperationalTransformation Class-like, promoting intermediates to Module
will create more problems to
solve."
Eclipse QVTd trims a QVTc/QVTr Transformation to be just a Class and
wrapping it in a Package appears to work fine. The only issue is whether
an implicitly named wrapping Package is unnamed or duplicates the
transformation name.
Current work on the Pivot-zed QVTo seems to come to the same conclusion.
Therefore if a Library evolves to be Package-like, intermediate classes
would appear to be no problem.
But one major problem with the current QVTo metmodels are that they are
very limited (no 'unique' collections) and severely underspecified
('ordered' not in the grammar). Chasing 100% of relevant UML
functionality is probably a loser; I can see a case for adding
StateMachines. But providing a lightweight capability for richer data
declarations could be useful.
If you provide a motivating example, it might be clearer what is needed.
Contrast the two approaches, forced use of an external model wrt use of
an internal model. Would the internal model be re-useable?
Regards
Ed Willink
On 20/05/2015 13:43, Christopher Gerking wrote:
> I actually filed an OMG issue on exactly this topic years ago:
> http://www.omg.org/issues/qvt-rtf.clos.html#Issue18325
>
> Ed Willink wrote on Mon, 04 May 2015 06:31
>> I don't use intermediate classes, certainly not for any non-trivial
>> purpose. I consider them to be unhelpful inferior language bloat.
>
>
> Can't reproduce. They are very helpful to store intermediate data,
> which doesn't make it into your target model, but is still required to
> build up that target model.
>
> Ed Willink wrote on Mon, 04 May 2015 06:31
>> Much better to design a proper Ecore model with whatever facilities
>> you want in it.
>
>
> Intermediate classes actually are a proper Ecore model, which just
> emerges dynamically from the textual syntax used in QVTo. I consider
> an external Ecore to be an unnecessary overhead.
>
> I still don't see a reason why intermediate classes should be
> restricted to transformations, and excluded for libraries.
|
|
|
|
|
Powered by
FUDForum. Page generated in 0.04743 seconds