Problem with generic parameter

Problem with generic parameter

[Jacob Sparre Andersen]

When I try to compile the program included below (cut down version of
a test program) GNAT 3.15p reports:

-----
Compiling: demo.adb (source file time stamp: 2005-06-22 10:10:41)

    16.                               Measurement_Array => Measurement_Array);
                                                           |
        >>> component subtype of actual does not match that of formal "Measurement_Array"
        >>> instantiation abandoned

 20 lines: 2 errors
-----

I can't see _how_ it doesn't match.  Can somebody figure out what I'm
doing wrong?  And maybe even suggest me an elegant solution to the
problem?

The reason I have Measurement_Array as a generic parameter is that I
also use it elsewhere, and that I may want to work with differently
indexed measurement arrays.  Does it - despite that - make more sense
to declare the types "Measurement" and "Measurement_Array" together?

/Jacob

------------------------------------------------------------------------------
--  Source code:

with Generic_Root;
with Generic_Root.Child;

procedure Demo is

   type Scalar is digits 15;

   package Measurements is
      new Generic_Root (Scalar => Scalar);

   type Measurement_Array is array (Positive range <>)
     of Measurements.Measurement;

   package Measurement_Text_IO is
      new Measurements.Child (Indices           => Positive,
                              Measurement_Array => Measurement_Array);

begin
   null;
end Demo;

generic

   type Scalar is digits <>;

package Generic_Root is

   ---------------------------------------------------------------------------
   --  type Limit:

   type Limit is (Below, Exact, Above, Undefined);

   ---------------------------------------------------------------------------
   --  type Measurement:

   type Measurement (As : Limit := Exact) is
      record
         case As is
            when Below | Exact | Above =>
               Value : Scalar;
            when Undefined =>
               null;
         end case;
      end record;

   ---------------------------------------------------------------------------

end Generic_Root;

with Ada.Text_IO;

generic

   type Indices is (<>);

   type Measurement_Array is array (Indices range <>) of Measurement;

package Generic_Root.Child is

end Generic_Root.Child;

Re: Problem with generic parameter

[Martin Dowie]

Jacob Sparre Andersen wrote:
> When I try to compile the program included below (cut down version of
> a test program) GNAT 3.15p reports:
>
> -----
> Compiling: demo.adb (source file time stamp: 2005-06-22 10:10:41)
>
>     16.                               Measurement_Array =>
>
>         Measurement_Array); | >>> component subtype of actual does
>         not match that of formal "Measurement_Array" >>>
> instantiation abandoned
>
>  20 lines: 2 errors
> -----
>
> I can't see _how_ it doesn't match.  Can somebody figure out what I'm
> doing wrong?  And maybe even suggest me an elegant solution to the
> problem?

Use a more compliant compiler! This works just fine with ObjectAda 7.2.2
(U18).

Actually, given that it doesn't work with GNAT GAP 1.1.0 either, I'll leave
it to a language lawyer to decide which is more compliant! :-)

Cheers

-- Martin

Re: Problem with generic parameter

[Marius Amado Alves]

I think this is legal, so the compiler is buggy. But I find it an  
unlogical generic idiom, and probably the bug has passed unnoticed  
before for this reason. That is, nobody wrote this before. What I find  
unlogical is the Generic_Root.Child parameters. I would find it logical  
to represent the array type as an exported entity instead of a generic  
parameter.

generic
    type Indices is (<>);
package Generic_Root
    type Measurement is array (Indices) of Measurement;

On 22 Jun 2005, at 11:21, Jacob Sparre Andersen wrote:

> with Generic_Root;
> with Generic_Root.Child;
>
> procedure Demo is
>
>    type Scalar is digits 15;
>
>    package Measurements is
>       new Generic_Root (Scalar => Scalar);
>
>    type Measurement_Array is array (Positive range <>)
>      of Measurements.Measurement;
>
>    package Measurement_Text_IO is
>       new Measurements.Child (Indices           => Positive,
>                               Measurement_Array => Measurement_Array);
>
> begin
>    null;
> end Demo;
>
> generic
>
>    type Scalar is digits <>;
>
> package Generic_Root is
>
>     
> ----------------------------------------------------------------------- 
> ----
>    --  type Limit:
>
>    type Limit is (Below, Exact, Above, Undefined);
>
>     
> ----------------------------------------------------------------------- 
> ----
>    --  type Measurement:
>
>    type Measurement (As : Limit := Exact) is
>       record
>          case As is
>             when Below | Exact | Above =>
>                Value : Scalar;
>             when Undefined =>
>                null;
>          end case;
>       end record;
>
>     
> ----------------------------------------------------------------------- 
> ----
>
> end Generic_Root;
>
> with Ada.Text_IO;
>
> generic
>
>    type Indices is (<>);
>
>    type Measurement_Array is array (Indices range <>) of Measurement;
>
> package Generic_Root.Child is
>
> end Generic_Root.Child;

Re: Problem with generic parameter

[Jacob Sparre Andersen]

Marius Amado Alves wrote:

> I think this is legal,

Thanks.

> so the compiler is buggy.

:-(

> But I find it an unlogical generic idiom, and probably the bug has
> passed unnoticed before for this reason. That is, nobody wrote this
> before. What I find unlogical is the Generic_Root.Child
> parameters. I would find it logical to represent the array type as
> an exported entity instead of a generic parameter.
> 
> generic
>     type Indices is (<>);
> package Generic_Root
>     type Measurement is array (Indices) of Measurement;

The reason that I don't do that is that that would have to instantiate
a copy of Generic_Root _and_ Generic_Root.Child for each choice of
array indices.

I will report the problem to ACT and Debian. - And use your suggestion
as a workaround.

Jacob
-- 
Would you also prefer not to have had to read Victor Hugo?
Vote for EU software patents!

http://www.guardian.co.uk/online/comment/story/0,12449,1510566,00.html

Re: Problem with generic parameter

[Marius Amado Alves]

>> generic
>>     type Indices is (<>);
>> package Generic_Root
>>     type Measurement is array (Indices) of Measurement;
>
> The reason that I don't do that is that that would have to instantiate
> a copy of Generic_Root _and_ Generic_Root.Child for each choice of
> array indices.

No you don't. (The above example was misleading.)

generic
    type Indices is (<>);
package Generic_Root.Child_By_Marius is
    type Measurement_Array is
       array (Indices range <>) of Measurement;
end;

with Generic_Root;
with Generic_Root.Child_By_Marius;
procedure Marius_Demo is
    type Scalar is digits 15;
    package Measurements is
       new Generic_Root (Scalar => Scalar);
    package Measurements_Text_IO is
       new Measurements.Child_By_Marius
         (Indices => Positive);
    use Measurements_Text_IO;               --|
    Array_1 : Measurement_Array (1 .. 10);  --| *HERE*
    Array_2 : Measurement_Array (2 .. 20);  --|
begin
    null;
end;

Re: Problem with generic parameter

[Marius Amado Alves]

I, Marius Amado Alves, just wrote:

> I think this is legal, so the compiler is buggy. But I find it an 
> unlogical generic idiom, and probably the bug has passed unnoticed 
> before for this reason. That is, nobody wrote this before. What I find 
> unlogical is the Generic_Root.Child parameters. I would find it 
> logical to represent the array type as an exported entity instead of a 
> generic parameter.
>
> generic
>    type Indices is (<>);
> package Generic_Root
>    type Measurement is array (Indices) of Measurement;

Here's the complete 'logical' idiom, which also corrects faults above:

generic
    type Indices is (<>);
package Generic_Root.Child_By_Marius is
    type Measurement_Array is
       array (Indices range <>) of Measurement;
end;

with Generic_Root;
with Generic_Root.Child_By_Marius;
procedure Marius_Demo is
    type Scalar is digits 15;
    package Measurements is
       new Generic_Root (Scalar => Scalar);
    package Measurements_Text_IO is
       new Measurements.Child_By_Marius
         (Indices => Positive);
    -- now here you have your unconstrained array type
    -- it is Measurements_Text_IO.Measurement_Array
    -- and you have defined it only once
    -- and the whole thing compiles
begin
    null;
end;