Re: Opinions on Programming by Contract

On 17 Jun, 20:56, "Malcolm McLean" <regniz...@btinternet.com> wrote:
> "James Harris" <james.harri...@googlemail.com> wrote in message news:
> > Is it worth including in a language postconditions and class (or any
> > other) invariants? I'd be interested in the views of other programmers
> > and language designers.
>
> If you're writing a high-level language, it's worth including ranges for
> variables.
>
> eg
>
> int i <0, N-1> /* i is a counter */
>
> you can extend this idea to input parameters and the return value. Obviously
> things get a bit trickier when you pass in vectors or structures.

Interesting. I am tenuously thinking of the compiler maintaining
possible range information as it looks at the dynamic flow of
computation. This is for optimisation purposes.

I am torn between checking and not checking actual variable ranges for
performance reasons. I can see that if I allow pre and post assertions
this is a natural extension, i.e. an invariant on a variable
declaration which, I believe is not supported in Eiffel (as Eiffel
only allows on new object types) but I may be wrong.

Speaking of language design and performance there may be a tie-in
between variable ranges and the saturated arithmetic performed by
Intel/AMD CPUs. In this case, a sum exceeding the limit would be
_silently_ pegged back to the limit. I presume preconditions and
postconditions are normally expected to raise an exception if the
condition is violated. Hmm....

--
James