Re: Ichbiah's Letter

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

"Jacob Sparre Andersen" <jacob@jacob-sparre.dk> wrote in message 
news:87mw8ivlr4.fsf@adaheads.sparre-andersen.dk...
>> https://duckduckgo.com/l/?kh=-1&uddg=http%3A%2F%2Fweb.elastic.org%2F~fche%2Fmirrors%2Fold-usenet%2Fada-with-null
>
> Reading the letter, I must say that I disagree with Ichbiah on some of
> the Ada 9X features which he wanted dropped:
>
> + Aliased objects:
>
>  I have more than one program I don't know how I could have implemented
>  sensibly without aliased objects.
...
> - Accessibility (checks):
>
>  It seems like this definitely is a part of the language, which has
>  ended up being too complicated for anyone to understand, but I am not
>  in a position to say anything sensible about how to avoid them in the
>  language.

If you don't have aliased objects, then you don't need accessibility checks 
(because you can't get the access value of an object). So those two go 
together. Of course, getting the access value of a local object was one of 
the requirements for Ada 9x, so Ichbiah was essentially arguing to ignore 
one of the language requirements. The MRT had some leeway in doing that, but 
hardly enough to drop a number of major requirements.

>+ Tagged types and dispatching:
>
>  Well.  Although Ichbiah puts tagged types on his list of complicated
>  features, he still argues for them further down in the letter.  I
>  suppose that we agree.

No, he argued for Simula classes; the fact that he lost that vote was the 
tripping point for his leaving the Ada 9x process and writing the letter in 
question.

Why he thought that adding a new complex form of visibility was somehow 
better than extending an existing form of type (the Ada 83 derived type) 
still escapes me to this day. (At worst, they're about the same in overall 
complexity.) There may have been compelling arguments in favor of a class 
type, but "complexity" wasn't one of them. (Indeed, I recall arguing that 
basing anything on derived types, a very obscure corner of the language, was 
going to make things harder for Ada users. But the class type would not have 
helped that at all -- something totally new isn't easy to learn, either. And 
the actual problems in practice, particularly overriding, would have been 
just as messed up either way (the syntax has nothing to do with it).)

                                          Randy.