Re: limited types (Was: Records that could be arrays)

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

On 28 Feb 2006 11:58:11 -0800, Matthew Heaney wrote:

> Dmitry A. Kazakov wrote:
>> On 28 Feb 2006 09:24:35 -0800, Matthew Heaney wrote:
>>
>> I still don't see CS.
> 
> I wrote the declaration wrong; it should have been:
> 
> generic
>   type ET is private;  --or: type ET (<>) is limited private;
>   type CT is private;  --cursor type
>   with function E (C : CT) return ET is <>;
> procedure Generic_Algorithm (C1, C2 : CT);
> 
> The cursor pair [C1, C2) describes a range of elements.  It might be
> the entire range of elements in the container, or just a subrange.  The
> algorithm doesn't care.

But still S <: T does not imply CS <: CT. 

>> BTW, if I had CS, would need not to instantiate Generic_Algorithm once
>> more.
> 
> (I assume "CS" means "container of element type S, and S derives from
> type T.")

Yes and CS should be a subtype of CT, so that CS could be passed as an "in"
where CT is expected. It also could be as an "out", but then with a chance
of Constraint_Error at run-time.

> As far as generic algorithms are concerned, it doesn't matter that type
> S derives from type T.  That's what "generic algorithm" means.

It means "works on a set of types." Generics are only one [weakest] form of
polymorphism.

> And yes, you have to instantiate the algorithm twice, since the cursor
> types come from different instantiations of some generic container
> package.

That is the whole point. BTW, you will need to instantiate it more than
twice for cross combinations, like when C1 is CS and C2 is CT. It
geometrically explodes.

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