Re: numbers as 'generics' parameters

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

On Sun, 18 May 2008 13:37:47 -0700 (PDT), Maciej Sobczak wrote:

> On 18 Maj, 16:52, "Dmitry A. Kazakov" <mail...@dmitry-kazakov.de>
> wrote:
> 
> Note also that the constraint to have equal strings is only an example
> and probably does not show the full problem.

(The problem; I don't understand, BTW)

> I might have written my code like this instead:
> 
> template <size_t N>
> void foo(const fixed_size_string<N> & s1,
>          const fixed_size_string<2 * N> & s2);

Easily:

generic
   N : Natural;
package Silly_String is
   type Pascal_String is new String (1..N);
end Silly_String;

with Silly_String;
generic
   with package One_Silly is  new Silly_String (<>);
   with package Twice_As_Silly is
      new Silly_String (One_Silly.Pascal_String'Length * 2);
procedure Foo
          (  S1 : One_Silly.Pascal_String;
             S2 : Twice_As_Silly.Pascal_String
          );

>> If you replaced the type declaration in Strings_Nobody_Ever_Needed to
>>
>>    subtype Pascal_String is String (1..N);
>>
>> you would start to see what's wrong with all this... (:-))
> 
> No, I don't follow. Could you please explain?

This Pascal_String would clash when multiple instantiated. The constraint
in your examples is a part of the type. This is wrong. The constraint of
the subtype is a proper constraint. To violate it is *legal*, so in order
to check it statically you need other (finer) means.
 
>> Talking about strings, they are allowed to have statically unknown length
> 
> Yes - the string was only a vehicle to show the problem. It might be a
> matrix of size N or just about anything else.

Still, for a vast majority of them, it is wrong to have the constraint a
part of the type. This is a fundamental issue, why generics are not a
solution.

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