Re: ada95 and uml

["Peter Dencker" <dencker@web.de>]

"Simon Wright" <simon@pushface.org> schrieb im Newsbeitrag
news:x7vsnj0go62.fsf@smaug.pushface.org...
> Robert Palasek <bobpalasek@home.com> writes:
>
> > "Riehle, Richard" wrote:
> >
> > > From my study of UML and my knowledge of Ada, I have concluded
> > > there is a substantial linguistic discontinuity.  Ada has a rich
> > > model of program development that cannot be > completely represented
> > > with UML.  Translating from UML to Ada or Ada to UML is guaranteed
> > > to lose something in the translation.
> >
> > I know a little about each, but am not an expert yet with either.
> >
> > But I don't understand the idea that you are supposed to be able to
> > generate one from the other.  That seems wrong and misguided to me.

The idea of round-trip is carried forward by folks whose reverse engineering
tool is as good as its forward generating code generator. That must
obviously be an isomorphic mapping. Thus, the picture can't tell you more
than the code and vice versa. If that is all they want, fine.

> >
> >  A model is an abstraction that carries some proper subset of
> > salient points of the thing being modeled.  For it to be useful, the
> > model has to leave other stuff out.  If it doesn't leave anything
> > out, it's just an alternate representation.
>
> True; model + translation rules -> code. This is possible provided
> your model is expressed with the precision required by the translation
> rules[1].
>
> This has been a principle of Shlaer-Mellor Recursive design for a
> while now (unfortunately the book hasn't come out).
>
> You can get a commercial UML->Ada generator now from Aonix that does
> this (for code frameworks only). I'm sure there are others, perhaps
> not for Ada!

Yes, indeed. You get StP/UML + ACD  and it does more than just code
frameworks only, e.g. generates all the automaton code for state transition
diagrams.

To get a first idea read http://www.aonix.com/content/downloads/stp/ACD1.pdf

It also tells you why round-trip does not help much if you see your design
as an abstraction over the final code.

>
> What you can't reasonably expect to do is to take the result of such a
> code-generation, change it, and round-trip back to the model. No great
> loss (IMO).
>
> [1] As a simple example, you might have to specify role names for each
> end of an association.