Re: Two generic families: I'm confused by error msg

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

On 2 Nov 2004 21:40:29 -0800, John Woodruff wrote:

> I'm puzzled by an Ada error.  The error arises in the attached
> program, which uses two different (parent-child) pairs of generic
> packages.  And the error seems to indicate a hard-for-me-to-understand
> issue of visibility.
> 
> The first two packages (numeric_io and it's child *.matrix_io) do some
> IO on vectors and matrices of floating numbers.  Just about like the
> new Ada.Numerics.Generic_Real_Arrays.Generic_IO.
> 
> Two more packages (name_io and its child *.matrix_name_io) add
> services that parse name-directed input for object-oriented
> applications.  This pair uses instances of numeric_io, so that vectors
> are the same regardless of which services are invoked.
> 
> It all works fine, and I'm in the final stages of preparing this to
> share.  Maybe on adaworld or adapower if they'll have me....
> 
> BUT
> 
> Numeric_io defines a subtype "real" that is just the actual type given
> for its argument.  That's how the two children manage to work on the
> same types (they are children of different, unrelated, parents).
> 
> So along comes the client (I named it "Unreal" in this exhibit). It
> laces the two generic families together, and it works perfectly  IF and
> ONLY IF it *happens* to use the identifier "Real" somewhere.  This
> client does not need to declare any *usage* of the subtype "Real", nor
> use it in instantiating the generics.
> 
> Unfortunately, poor client has no way of knowing he needs to declare
> "real".  And I'm not sure what is the language-law that connects the
> type Scalar_IO.Real in the generic actual package with the subtype
> identifier Real in the client.
> 
> Can someone help me make sense of this?  Can someone suggest a way to
> rephrase these relationships so I don't require the client to "say the
> magic word"?
> 
> <<< gnatchop the stuff after this line to see the puzzle >>>
> 
> generic
>    type Floating_Type is digits <> ;
> package Numeric_Io is
>    subtype Real is Floating_Type ;
> end Numeric_Io ;
> 
> 
> generic
>    type Vector_Ix is  (<>) ;
>    type Vector is array (Vector_Ix range <>) of Real ;
> package Numeric_Io.Matrix_IO is
> end Numeric_Io.Matrix_IO ;
> 
> with Numeric_Io;
> generic
>    type Floating_Type is digits <> ;
>    with package Scalar_Io is new Numeric_Io (Floating_Type) ;
> package Name_Io is
> end Name_Io;
> 
> with Numeric_Io.Matrix_Io ;
> generic
>    type Vector_Ix is  (<>) ;
>    type Vector is array (Vector_Ix range <>) of Scalar_Io.Real ;
>    with package Matrix_Io is new Scalar_Io.Matrix_Io
>      (Vector_Ix, Vector) ;
> package Name_Io.Matrix_name_IO is
> end Name_Io.Matrix_name_IO ;

Why do not you do it as:

with Numeric_Io.Matrix_Io ;
generic
   with package Matrix_Io is new Scalar_Io.Matrix_Io (<>);
package Name_Io.Matrix_name_IO is

Here you can refer the actual types of Matrix_IO instantiation as
Matrix_IO.Vector_Ix and Matrix_IO.Vector. This would also make
instantiations easier.

> with Numeric_Io.Matrix_Io ;
> with Name_Io.Matrix_Name_Io ;
> 
> procedure Unreal is
>    type Unreal is digits 6 ;
>    
> --------------------------------------------------------------------------
> -- If this line "subtype Real is Unreal"  is ABSENT, then error at line 26:
> -- Component subtype of actual does not match that of formal "Vector" 
> --------------------------------------------------------------------------
>    subtype Real is Unreal;
>    
>    type Vector is array (Integer range <>) of Unreal ;
> 
>    package Scalar_IO is new Numeric_IO (Unreal) ;
> 
>    package Matrix_IO is new Scalar_IO.Matrix_IO
>      (Vector_Ix  => Integer,
>       Vector     => Vector);
> 
>    package Nio is new Name_Io (Floating_Type => Unreal,
>                                Scalar_Io     => Scalar_IO);
> 
>    package Mat_IO is new Nio.Matrix_Name_IO
>      (Vector_Ix  => Integer,
>       Vector     => Vector,   -- the error is here
>       Matrix_Io  => Matrix_IO) ;

Yes, I saw that problem too and many times. To me it looks like a compiler
bug, but wait what our language lawyers will say.

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