Re: disjoint ranges ?

[Robert A Duff <bobduff@world.std.com>]

Robert Dewar <robert_dewar@my-deja.com> writes:

> No, it could be over all damaging. Every new feature you add
> to a language damages it by adding to the understanding,
> programming, description, and implementation complexity.

Quite true, but you don't necessarily have to add features in order to
add functionality.  Sometimes, making a feature more powerful can add
functionality without increasing complexity.

For example, the approach I would take in this case would be to add
arbitrary boolean conditions as invariant conditions (somewhat as in
Eiffel), and then *get rid of* all the various kinds of special-purpose
constraints (range, index, discrim, accuracy).  This would lead to
adding the requested functionality, at a net *decrease* in language
complexity.

Note that Eiffel doesn't quite solve this problem: it has arbitrary
invariant expressions on types, but they can't be applied to integers
(there's only one integer type, and it can't have invariants).

> A feature like this is actually *quite* complex, think for
> example, that we would have to allow dynamic bounds for the
> ranges, and then the checks for disjointness get tricky
> to describe (e.g. can several ranges be flat or superflat?
> if they are, can bounds overlap for the flat ranges? etc)
> 
> That's only the tip of the iceburg complexity wise.
> 
> But the point is that you have to make sure that the value
> of the feature more than undoes the damage done by its
> addition. The burden of proof is tricky in this case, and
> I doubt a convincing argument can be made for adding this
> feature to the language.

Agreed.  Simply *adding* this feature as a feature is not a good idea.

- Bob