Re: Unconstrained Arrays

[Robert A Duff <bobduff@shell01.TheWorld.com>]

sjw <simon.j.wright@mac.com> writes:

> That makes so much sense! I think my brain freezes when I see the
> compiler quoting rules about statically matching, so I just remember
> the first 2 lines of the message (as you say, pretty explicit) and
> pass over the reason why in the last 2 lines. After all, GNAT is much
> more likely to be right than I am.

Well, GNAT is right in the sense that it obeys the Ada RM on this point.
But some people (including some of the folks who wrote GNAT) think the
language designers were wrong on this point.  It's an efficiency issue.
You can certainly come up with cases that make a big difference:

    subtype S is String (1..4);
    type A is array (Positive range <>) of aliased S;
    X : A (1..1_000_000);
    
    type S_Ref is access all S;

An object of subtype S is likely 4 bytes without dope, 12 bytes with
dope.  Removing this restriction from the language would triple
the size of X from 4 million bytes to 12 million.  If you only
wanted to point at components of X from objects of type S_Ref,
(as opposed to some "access all String" type) then that would be
annoying (why store 1 million copies of the compile-time-known
values 1 and 4?).

Memory is not cheap -- using more memory tends to cause cache misses.

On the other hand, the restriction is confusing, and not strictly
necessary.

- Bob