Re: discriminant in constraint must appear alone

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

"Vincent Smeets" <No.Spam@T-Online.de> wrote in message
news:bqiq1m$uis$00$1@news.t-online.com...
> Hallo,
>
> I want to define a discriminated record type with two arrays. One array
only half the size
> of the other. Below is the type definition that I want to use.
>
>   type R (D : Positive) is
>      record
>         A : String (1 .. D);
>         B : String (1 .. D / 2);
>      end record;
>
> This type definition can't be compiled by the GNAT compiler. It gives the
error message
>    "discriminant in constraint must appear alone"
>for the record component B. I know this isn't correct Ada, but how should I
define the type
>in a correct way?

I don't think you can.

>I have defined B as
>    B : String (1 .. D);
>and only used the first half of it, but this way I waist memory and can't
have any Ada
>constraint checks for component B. So this is not the way I want to do it.
>
>Are there other possiblities?

The only thing that comes to mind is to have a second discriminant. The
problem with that is that the relationship between them can't be defined
formally.

   type R (D1, D2 : Positive) is
     record
        A : String (1 .. D1);
        B : String (1 .. D2);
     end record;

  Obj : R (Max_Size, Max_Size/2);

I suppose you could have Initialize check if they don't match and raise an
exception:

   type R (D1, D2 : Positive) is new Ada.Finalization.Controlled with
     record
        A : String (1 .. D1);
        B : String (1 .. D2);
     end record;

   procedure Initialize (Obj : in out R);

   procedure Initialize (Obj : in out R) is
   begin
      if Obj.D1 /= Obj.D2*2 then
          raise Constraint_Error;
         -- Or better Ada.Exceptions.Raise_with_Message
(Constraint_Error'Identity, "D2 is not half of D1");
      end if;
   end Initialize;

  Obj : R (Max_Size, Max_Size/2);
  Obj2 : R (Max_Size, Max_Size); -- Would raise Constraint_Error.

But that seems like a lot of mechanism for a simple check. (OTOH, my theory
is that virtually all types ought to be derived from Controlled or
Limited_Controlled anyway, so this doesn't add much in that case.)

                      Randy.