Re: ADA 95 Child Packages and UML

[Richard D Riehle <laoXhai@ix.netcom.com>]

In article <38132103.380DA624@club-internet.fr>,
	Philippe Dalleau <phda@club-internet.fr> wrote:

>Hello,
>
>Peter Lindgren wrote:
>> 
>> Hello!
>> How do I model Ada95 child packages in UML?
>> Is it possible at all?
>
>Sure but it depends a lot on the UML tool you choose.
>With Rational Rose/Ada 95 : via a class or a component whose name
>follows the rule : identifier.{identifier}.

No. It is not possible to correctly model child library units
in UML.  There is a workaround, as described above, but it is 
something of a kludge. There is no UML way to properly represent 
an architecture built on child and private child library units.

In its defense, UML is a modeling tool.  Asking a question such
as "Can I model xxxx in UML," is a lot like asking, "Can I program
the cache in C++ (or Ada)?"  One must consider the level of 
abstraction being addressed by a linguistic protocol.  Many
software architects believe UML and its various associated methods
to be just the right tool for abstracting away the kind of detail
that distracts from architectural considerations.

On the Ada side, UML includes mechanisms that cannot be easily 
designed in Ada.  The most compelling examples of these are 
the "association class" and multiple inheritance.  

This problem is not unique to Ada. UML does not represent every C++
construct, and vice versa.  It highlights a principle described 
in Bertrand Meyer's, "Object-Oriented Software Construction" which
he calls, "Linguistic Modular Units."  The idea is that the language
of modeling must correspond to that of design, that of programming,
and that of testing, etc. throughout the software lifecycle.  At every
place in our software effort in which there is a linguistic disconnect,
we increase the probability for failure.  I don't know that anyone
has ever done serious research on this issue.  It would probably make
a good Ph.d. thesis for someone.

Richard Riehle
http://www.adaworks.com