Re: Ada vs Eiffel - Ada programmer approach

["Randy Brukardt" <randy@rrsoftware.com>]

"Robert A Duff"  writes:
...
> An invariant (whether it be a pragma, or some special syntax) would be
> attached to a type, and would mean "all objects of this type have
> this property, always" (except that maybe objects can temporarily
> violate their invariant, so long as the code puts it back).
>
> A related concept that has been discussed is "user-defined constraints".
> I'm not sure what the difference is, exactly -- I think with user-defined
> constraints, you can't "temporarily violate".

A user-defined constraint applies to a view (usually via a subtype), while 
an invariant applies to a type (always). For instance, a user-defined 
constraint can apply to a formal parameter but not to the actual parameter 
object (just as any other constraint). Calling such a thing an "invariant" 
would be completely wrong, because it's not invariant at all - the values 
are only checked against the constraints at the points where Ada does 
subtype conversions (and possibly a few others - this is TBD and hopefully 
will be discussed at this week's ARG meeting).

Tucker explains the distinction somewhat in his new draft of AI-146 (to be 
posted tonight, along with the ARG meeting agenda). There also is a partial 
user-defined constraint AI (AI-153), but I got stuck over how invariant we 
want these to be. Check them both out tomorrow.

                                           Randy.