Re: Other Ada Standards (was Re: SIGada Conference)

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

Georg Bauhaus wrote:

> Dmitry A. Kazakov <mailbox@dmitry-kazakov.de> wrote:
> : I find "with"-ing boring. I cannot remember any case they helped me to
> : solve any problem.
> 
> They might help others solving a problem :-)
> with clauses might mark stars, followers, and independents in a network of
> dependent units. Many with clauses -> unit is very likely a star.
> If there are many stars, and no clique, something could be
> wrong. In this sense with clauses document design and might indicate
> problems.

I think that children packages do the job much better. They have a
substance, they extend/implement parents. As for "with", they lack any
substance. A program with a superfluos "with" remains valid, always. This
is an important sign. For example, "use" has a substance, because you can
break a program if you add a "use" causing a name conflict. BTW, if there
where a way to prevent name conflicts I would argue to get rid of "use".
Unfortunately there is no one.

> As an example, think of a class that takes on several roles but does
> so only by way of a single huge collection of state variables cobbled
> together.  (The (Java) class I am thinking of still exist; there is
> some hope that it will be replaced .-)
> That is, the states could have been grouped semantically
> but have not been grouped using language means for grouping. Can
> one build this class without a long list of with clauses somewhere
> (or by writing a larger number of "provider" packages nested in the
> same block so they can be seen where the type is declared,
> or worse, a large package providing everything)?

It is an interesting problem. There is a similar problem with ADT - "fat
classes". The base types tend to rapidly grow by adding more and more
primitive operations. It is very difficult to keep new pritive operations
in derived types, against the "gravity" force which attracts everything to
the base.

> Maybe it is easier to think about the correctness of what is happening
> inside a package if the number of with clauses is low?
> 
> : Moreover careless designers tend to "with" everything
> : they can.
> 
> So "with" is a useful indicator to maintainers?

Oh yes, but the number of source code lines is also an indicator. Should we
follow Basic by explicit numbering them? The point is that, yes, the
package dependencies is an important metric. But this by no means implies a
necessity to force a designer to *manually* measure it and then include it
in the code!

-- 
Regards,
Dmitry A. Kazakov
www.dmitry-kazakov.de