Re: Is the Writing on the Wall for Ada?

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

Wes Groleau wrote:

> Dmitry A. Kazakov wrote:
>> This is what I want. And this is what arrays cannot:
>> 
>> type Element (Constraint : Natural) is ...;
>> type Array_Of_Elements is array (...) of Element;
>> 
>> This is not allowed, because I cannot constrain Element, and there is no
>> way to do it other than create a constrained subtype of Element.
> 
> I realize this is not exactly what you meant,
> but is it not possible to do this:
> 
>     type Element (Constraint : Natural) is ...;
> 
>     type Array_Of_Elements is array (...) of Element;

No. Consider a very simple thing:

   type String_Array is array (...) of String;

You cannot do that because the element is not constrained. The compiler does
not know that in our case all elements are of same length. How to tell him?

You have to do something like:

   type String_In_Record (Length : Positive) is
      The_Body : String (1..Length);
   end record;

and even this artificial construct does not work:

   type String_Array is array (...) of String_In_Record; -- Error

So you have to go further:

   type String_In_Record (Length : Positive := 80) is
      The_Body : String (1..Length);
   end record;

This would result in Storage_Error. Defaulted discriminants are not what
people usualy think about them!

And even if you replace Positive with something of lesser range, you will
not solve the problem:

   type Index is range 1..1024;
   type Yet_Another_String is array (Index range <>) of Character;
   type String_In_Record (Length : Index := 80) is
      The_Body : Yet_Another_String (1..Length);
   end record;
   type String_Array is array (...) of String_In_Record;

Observe that this disgusting construction does not serve the purpose,
because it allocates the memory for the worst case.

And also this does not work for all the cases where the defaults are not
allowed for the discriminants. For example (a very important one), the
tagged types cannot have defaulted discriminants.

Presently this is unsolvable without generics.

-- 
Regards,
Dmitry A. Kazakov
www.dmitry-kazakov.de