Re: Working with incompetent adaists / unsafe typing war story

[<adaworks@sbcglobal.net>]

"Anonymous Coward" <anonymous@coward.org> wrote in message
news:p9cJf.34455$lo3.34072@trnddc07...
>
> Maybe I'm just dealing with a tough crowd.  A systems fella ordered a
> change on one of my private types, forcing me to represent the same
> type with a different set of numbers (a change in coordinate systems
> to be exact).  Because it was a private type, I only needed to make
> that change in ONE package; which took me about 15-20 minutes (the
> same change would have taken most of the day if the type were public).

Someone recently encountered a problem with a package I wrote years
ago that had a limited private type.  It seems I had overlooked one possible
requirement (method)  for that type. The need for the method is rare but I
should
have anticipated it.   No one ever encountered the problem before, as far as
I know.

It took me five minutes to add a specification for that method and about as
long to write an implementation for it in the package body.  The algorithm
was simple to implement, but did involve an if statement.   Had programmers
been able to write their own code for the functionality of this method, the
probability of some of them getting it wrong was low, but still possible.  In
this
case, the algorithm was testable in one place rather than scattered about.

If I had declared the type as non-limited, or had used a pre-defined type,
the absence of the method would never have manifested itself.   Users would
have simply carried on without thinking about it.  From this example, some
designers might suppose that the type should not have been limited.   My
own view is that we learned something from discovering the need for this
method.   Also, I ended up thinking more deeply about what other methods
might be required in the future.  I did not come up with anything else.

Unfortunately we did have to recompile the package specification, but that did
not seem to present any hardship. More important, the internal strucutre of
instances of the type continue to be unreachable by users of the package.

Richard Riehle