Question on variant/discriminated records

Question on variant/discriminated records

[reinert]

Hello,

Assume the following program and change "y1" to "y". 

procedure test3 is
 type options1_Type is (a,b,c);
 type rec1_Type(option1 : options1_Type) is
   record
      x : Float;
      case option1 is
        when a => null;
        when b => y : float;
        when c => y1,z : float;
      end case;
   end record;
begin
   null;
end test3;

My compiler does not like this change.  I find it not natural to invent
a new name "y1" instead of "y". What should I do?

reinert

Re: Question on variant/discriminated records

[J-P. Rosen]

Le 31/12/2020 à 11:38, reinert a écrit :
> Assume the following program and change "y1" to "y". 
> 
> procedure test3 is
>  type options1_Type is (a,b,c);
>  type rec1_Type(option1 : options1_Type) is
>    record
>       x : Float;
>       case option1 is
>         when a => null;
>         when b => y : float;
>         when c => y1,z : float;
>       end case;
>    end record;
> begin
>    null;
> end test3;
> 
> My compiler does not like this change.  I find it not natural to invent
> a new name "y1" instead of "y". What should I do?
> 
Yes, it is a rule that identifiers in parallel branches cannot be
identical. And there is a good reason for this: in Ada, when you access
a component, the compiler checks that the component exists, according to
the value of the discriminant. Assume that you write (if V is of type
rec1_type):
   V.Y := 1.0;

If both components are named Y, should the discriminant be B or C?
-- 
J-P. Rosen
Adalog
2 rue du Docteur Lombard, 92441 Issy-les-Moulineaux CEDEX
Tel: +33 1 45 29 21 52, Fax: +33 1 45 29 25 00
http://www.adalog.fr

Re: Question on variant/discriminated records

[reinert]

Here must be a hole in my understanding. I thought that the discriminant must be known
before V.Y := 1.0; and hence there should be no confusion?

reinert


torsdag 31. desember 2020 kl. 11:52:26 UTC+1 skrev J-P. Rosen:

> Yes, it is a rule that identifiers in parallel branches cannot be 
> identical. And there is a good reason for this: in Ada, when you access 
> a component, the compiler checks that the component exists, according to 
> the value of the discriminant. Assume that you write (if V is of type 
> rec1_type): 
> V.Y := 1.0; 
> 
> If both components are named Y, should the discriminant be B or C? 
> -- 
> J-P. Rosen 
> Adalog 
> 2 rue du Docteur Lombard, 92441 Issy-les-Moulineaux CEDEX 
> Tel: +33 1 45 29 21 52, Fax: +33 1 45 29 25 00 
> http://www.adalog.fr

Re: Question on variant/discriminated records

[Gautier write-only address]

On Thursday, December 31, 2020 at 1:24:25 PM UTC+1, reinert wrote:
> Here must be a hole in my understanding. I thought that the discriminant must be known 
> before V.Y := 1.0; and hence there should be no confusion? 

Imagine you have a function Evaluate (r : rec1_Type) return Float.
Except in trivial cases like
  if r.option1 = b then result := result + r.y; end if;
the compiler cannot guess the value of the discriminant.

Re: Question on variant/discriminated records

[reinert]

Approaching :-) But the interpretation of (the meaning of) r.y could in principle be 
calculated on the fly (when called) based on the received discriminant -
but arguments related to  security/reliability  are against?

Maybe nice to find errors at compile time as compared to experiencing costly runtime errors :-)

reinert

torsdag 31. desember 2020 kl. 13:46:47 UTC+1 skrev gautier:
> On Thursday, December 31, 2020 at 1:24:25 PM UTC+1, reinert wrote: 
> > Here must be a hole in my understanding. I thought that the discriminant must be known 
> > before V.Y := 1.0; and hence there should be no confusion?
> Imagine you have a function Evaluate (r : rec1_Type) return Float. 
> Except in trivial cases like 
> if r.option1 = b then result := result + r.y; end if; 
> the compiler cannot guess the value of the discriminant.

Re: Question on variant/discriminated records

[DrPi]

Le 31/12/2020 à 13:46, Gautier write-only address a écrit :
> On Thursday, December 31, 2020 at 1:24:25 PM UTC+1, reinert wrote:
>> Here must be a hole in my understanding. I thought that the discriminant must be known
>> before V.Y := 1.0; and hence there should be no confusion?
> 
> Imagine you have a function Evaluate (r : rec1_Type) return Float.
> Except in trivial cases like
>    if r.option1 = b then result := result + r.y; end if;
> the compiler cannot guess the value of the discriminant.
> 
Isn't the discriminant stored in the record ?

Re: Question on variant/discriminated records

[Dmitry A. Kazakov]

On 2020-12-31 11:38, reinert wrote:

> Assume the following program and change "y1" to "y".
> 
> procedure test3 is
>   type options1_Type is (a,b,c);
>   type rec1_Type(option1 : options1_Type) is
>     record
>        x : Float;
>        case option1 is
>          when a => null;
>          when b => y : float;
>          when c => y1,z : float;
>        end case;
>     end record;
> begin
>     null;
> end test3;
> 
> My compiler does not like this change.  I find it not natural to invent
> a new name "y1" instead of "y". What should I do?

Use nested choice:

    type rec1_Type (option1 : options1_Type) is record
       x : Float;
       case option1 is
          when a =>
             null;
          when b | c =>
             y : float;
             case option1 is
                when c =>
                   z : float;
                when others =>
                   null;
             end case;
       end case;
    end record;

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de

Re: Question on variant/discriminated records

[Gautier write-only address]

> > Imagine you have a function Evaluate (r : rec1_Type) return Float. 
> > Except in trivial cases like 
> > if r.option1 = b then result := result + r.y; end if; 
> > the compiler cannot guess the value of the discriminant. 
> >
> Isn't the discriminant stored in the record ?

Yes. But if both "y"s were allowed, the compiler should for the non-trivial cases, generate a hidden "case" instruction to select the right "y" depending on the value of option1 (both "y"s are not necessarily at the same position in the record). Certainly the language designers have considered the additional complexity and performance cost of that point. Fortunately the nested choice as suggested by Dmitry solves the problem.

Re: Question on variant/discriminated records

[reinert]

Thanks, problem clarified and solved :-)

reinert

torsdag 31. desember 2020 kl. 16:03:57 UTC+1 skrev Dmitry A. Kazakov:

> Use nested choice: 
> 
....