Re: Problem space

Re: Problem space

[Alexandre E. Kopilovitch]

Simon Wright <simon@pushface.org> wrote:
>Can't help feeling that a problem that can be explained by source code
>is going to be on the small side.

Yes, but no-so-small problems aren't monolitic, each of them contains a number
of small ones (it isn't original observation, I think -:) . And while a natural
(human) language (note that it is not always English at all, and quite often
it is poor English) is most suitable for general overviews and causal remarks,
you can't cover all the details using the same natural language, unless you
are exceptionally good technical writer.

> And I don't fancy using source code to back up my design to customers.

Surely. During a presentation your customers need not *deal* with the source
code, they need not make immediate changes in it.

>On the other hand, I remember a colleague who claimed that his VDM
>spec was an excellent foundation for a customer review when annotated
>with English!

I see nothing surprising it that case. Good annotations in English should be
the best thing for customer review. But: 1) that was about spec, not about
implementation (including low-level specs); 2) what will remain from initially
excellent annotations after several years of maintenance and many changes?
  It is well-known problem with detailed comments: even initially perfect
comments eventually become poor or even inconsistent with the code after
substantial period of maintenance. "Things are best at their beginnings".

>...
>There is nothing to prevent definition of a subset ('profile') that is
>rigorously-enough defined to be translated. You can do that with a
>small core of UML. For example, http://www.projtech.com/

Well, but with that customization you lose "U" in "UML". And after several
such "rigorizations" of different but substantially overlapping subsets, what
will you have? Do you think that you will still dealing with the UML rather
then with some family of incompatible dialects?


Alexander Kopilovitch                      aek@vib.usr.pu.ru
Saint-Petersburg
Russia

Re: Problem space

[Simon Wright]

"Alexandre E. Kopilovitch" <aek@vib.usr.pu.ru> writes:

> Simon Wright <simon@pushface.org> wrote:

> >There is nothing to prevent definition of a subset ('profile') that is
> >rigorously-enough defined to be translated. You can do that with a
> >small core of UML. For example, http://www.projtech.com/
> 
> Well, but with that customization you lose "U" in "UML". And after
> several such "rigorizations" of different but substantially
> overlapping subsets, what will you have? Do you think that you will
> still dealing with the UML rather then with some family of
> incompatible dialects?

I doubt there is a project in the world that has used all the features
of the UML. Any project should subset the UML, just as you would
subset Ada. Certainly for translation, you need to restrict yourself
to that subset which includes the detail you need, with defined
semantics. You should also be able to include some of the woolly parts
to help explain what's going on without affecting the translation
(like comments).

No point in a universal language which no one can agree on the
semantics of!

I came across a vendor (I-Logix?) who had extended sequence diagrams
to allow their tool to perform automated testing, but at the cost of
non-standard usage I think.

UML includes extension mechanisms, a profile is (by one definition) a
statement of the subset and extensions (stereotypes and tagged values)
with their interpretations.