generic parameter Copy for primitifs types.

generic parameter Copy for primitifs types.

[nblanpain]

Hello,

this is my problem :

-----
generic
type T_Item is private;
with procedure Copy (Left : in out T_Item; Right : in T_Item);
package Toto is
...
end Toto;
------

At instanciation, for T_Item = Integer for example, is there a methode
to take for Copy => ":=". Must I redefined Copy ? Can I say that, by
default, take ":=" ?

Thanks,

Re: generic parameter Copy for primitifs types.

[Marc A. Criley]

nblanpain@hotmail.com wrote:

> -----
> generic
> type T_Item is private;
> with procedure Copy (Left : in out T_Item; Right : in T_Item);
> package Toto is
> ....
> end Toto;
> ------
> 
> At instanciation, for T_Item = Integer for example, is there a methode
> to take for Copy => ":=". Must I redefined Copy ? Can I say that, by
> default, take ":=" ?

No, you can't.  The author of Toto presumably required the provision of 
a Copy procedure because what it's going to be doing may not be 
compatible with the default ":=".  (Though perhaps T_Item should then 
have been declared as "limited private".)  If a special Copy procedure 
wasn't needed, the implementation could've just used ":=", since that is 
allowed for non-limited private types.

Even if the type you're supplying, e.g. Integer, would be fine with the 
built-in assignment, the package author would want to ensure that you 
had thought about it, and by writing a simple little function to do the 
copying, he ensures that.  (Though I also think the "Left" parameter 
should probably be mode "out" rather than "in out".)

Marc A. Criley
www.mckae.com

Re: generic parameter Copy for primitifs types.

[Dmitry A. Kazakov]

On 23 Jun 2005 09:41:51 -0700, nblanpain@hotmail.com wrote:

> this is my problem :
> 
> -----
> generic
> type T_Item is private;
> with procedure Copy (Left : in out T_Item; Right : in T_Item);
> package Toto is
> ...
> end Toto;
> ------
> 
> At instanciation, for T_Item = Integer for example, is there a methode
> to take for Copy => ":=". Must I redefined Copy ? Can I say that, by
> default, take ":=" ?

No, for good or bad, but ":=" is not a procedure in Ada.

One point, the formal generic parameter T_Item is declared as "private".
This means that its actual counterpart must have assignment. So you can
write:

generic
  type T_Item is private;
package Toto is
   ...
end Toto;

and use assignment within the body of Toto:

package body Toto is
   ...
      X, Y : T_Item;
   begin
   ...
      X := Y;
   ...
end Toto;

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

Re: generic parameter Copy for primitifs types.

[Björn Persson]

nblanpain@hotmail.com wrote:
> generic
> type T_Item is private;
> with procedure Copy (Left : in out T_Item; Right : in T_Item);
> package Toto is
> ....
> end Toto;
> ------
> 
> At instanciation, for T_Item = Integer for example, is there a methode
> to take for Copy => ":=". Must I redefined Copy ? Can I say that, by
> default, take ":=" ?

The normal way when you need to do special things during assignment is 
to use a controlled type. That is, you derive from 
Ada.Finalization.Controlled and override Adjust:

    type Special_Type is new Ada.Finalization.Controlled with record ...

    procedure Adjust (Object : in out Special_Type) is ...

Then, instead of a Copy procedure you just use ":=", and Adjust will be 
called. (See RM 7.6.)

Your generic package would then look like this:

    generic
       type T_Item is private;
    package Toto is
       ...
    end Toto;

    package Integer_Toto is new Toto (Integer);
    package Special_Toto is new Toto (Special_Type);

Or is there a reason why controlled types wouldn't work in your case?

-- 
Bj�rn Persson                              PGP key A88682FD
                    omb jor ers @sv ge.
                    r o.b n.p son eri nu

Re: generic parameter Copy for primitifs types.

[nblanpain]

I can't use controlled types (aeronautic software).
Is there an other solution?

Re: generic parameter Copy for primitifs types.

[nblanpain]

But I want to use Copy for other types other than primitifs (integer,
float, boolean...). The best solution is to set ":=" by default but it
is not possible...

Re: generic parameter Copy for primitifs types.

[Marc A. Criley]

nblanpain@hotmail.com wrote:
> I can't use controlled types (aeronautic software).
> Is there an other solution?

Um...don't fight it, and just write Copy procedures for each type with 
which the generic is going to be instantiated?

procedure Copy(Left : in out Integer; Right : in Integer) is
begin
    Left := Right;
end Copy;

Re: generic parameter Copy for primitifs types.

[Marius Amado Alves]

> But I want to use Copy for other types other than primitifs (integer,
> float, boolean...). The best solution is to set ":=" by default but it
> is not possible...

If you really want a default you can do it with 2 levels of generics 
and an access-to-subprogram type, as below (not tested). In Ada 2006 I 
think you can do without the access type.

generic
    type T is private;
package Level_1 is
    type Copy_Ptr is access procedure (From : in T; To : out T);
    procedure Default_Copy (From : in T; To : out T);
    generic
       Copy : Copy_Ptr := Default_Copy'Access;
    package Level_2 is
       ...
    end;
end;

Re: generic parameter Copy for primitifs types.

[Jeffrey Carter]

nblanpain@hotmail.com wrote:
> 
> -----
> generic
> type T_Item is private;
> with procedure Copy (Left : in out T_Item; Right : in T_Item);
> package Toto is
> ...
> end Toto;

Not that both ":=" and Copy are available to Toto for type T_Item. That 
seems odd; Toto is probably incorrectly specified.

Making Left (in Copy) mode in out can cause a problem with elementary 
types. Such types are checked for subtype conformance on copy-in to 
subprograms, and again on copy-out (elementary types are passed by 
copy). Consider a type

type T is range 1 .. 10;

and a variable

V : T;

and a procedure Copy such as yours for T. If I write

Copy (Left => V, Right => T'First);

the value of V will be checked against the range of T on copy-in to the 
procedure. Objects of type T will probably take 8 bits (or more) on most 
processors; thus, it's likely that the initial value of V will not be in 
T, and the call to Copy will raise Constraint_Error.

While you cannot pass default assignment to the generic, you can create 
a helper generic to make this case easier:

generic -- Assignment
    type Item is private;
procedure Assignment (To : out Item; From : in Item);

procedure Assignment (To : out Item; From : in Item) is
    -- null;
begin -- Assignment
    To := From;
end Assignment;

Then you can write

procedure Assign is new Assignment (Item => Integer);
package Dunno is new Toto (T_Item => Integer, Copy => Assign);

If you change Copy to Assign and use a default parameter:

    with procedure Assign (To : out T_Item; From : in T_Item) is <>;

then the 2nd instantiation can become

package Dummo is new Toto (T_Item => Integer);

-- 
Jeff Carter
"Sheriff murdered, crops burned, stores looted,
people stampeded, and cattle raped."
Blazing Saddles
35

Re: generic parameter Copy for primitifs types.

[Dmitry A. Kazakov]

On 23 Jun 2005 12:41:44 -0700, nblanpain@hotmail.com wrote:

> But I want to use Copy for other types other than primitifs (integer,
> float, boolean...).

Though other types have assignment too. So the cleanest way would be either
to consistently use it or to make the formal parameter T_Item limited and
provide Copy procedure as you did.

> The best solution is to set ":=" by default but it
> is not possible...

True. You can try generic "specialization" like this:

------------
generic
   type T_Item is limited private;
   with procedure Copy (Left : out T_Item; Right : in T_Item);
package Toto is
   ...
end Toto;

with Toto;
generic
   type T_Item is private;
package Specialized_Toto is
   procedure Copy (Left : out T_Item; Right : T_Item);
   package Toto_Instance is new Toto (T_Item, Copy);
end Specialized_Toto;

package body Specialized_Toto is
   procedure Copy (Left : out T_Item; Right : T_Item) is
   begin
      Left := Right;
   end Copy;
end Specialized_Toto;
-------------

So instantiation with Integer might look like:

package Integer_Toto_Implementation is new Specialized_Toto (Integer);
package Integer_Toto renames Integer_Toto_Implementation.Toto_Instance;

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

Re: generic parameter Copy for primitifs types.

[Marius Amado Alves]

On 24 Jun 2005, at 08:28, Dmitry A. Kazakov wrote:
>> The best solution is to set ":=" by default but it
>> is not possible...
>
> True. You can try generic "specialization" like this:
> ...
> generic
>    type T_Item is private;
> package Specialized_Toto is
>    procedure Copy (Left : out T_Item; Right : T_Item);
>    package Toto_Instance is new Toto (T_Item, Copy);
> end Specialized_Toto;
> ...

True. But then the use is not uniform. A different generic package must 
be selected for each case:

package Default_Pkg is new Specialized_Toto (T);
package Special_Pkg is new Toto (T, Special_Copy);

The idiom I have posted earlier does not have this problem. But, 
granted, may have others instead, e.g. requiring two instantiations 
instead of one for each case.

All this would go away if packages were first class citizens of Ada, 
i.e. objects. This is one thing I'm trying to do with Ocpotus 
(softdevelcoop.org/ocpotus).

Re: generic parameter Copy for primitifs types.

[Dmitry A. Kazakov]

On Fri, 24 Jun 2005 12:45:19 +0100, Marius Amado Alves wrote:

> On 24 Jun 2005, at 08:28, Dmitry A. Kazakov wrote:
>>> The best solution is to set ":=" by default but it
>>> is not possible...
>>
>> True. You can try generic "specialization" like this:
>> ...
>> generic
>>    type T_Item is private;
>> package Specialized_Toto is
>>    procedure Copy (Left : out T_Item; Right : T_Item);
>>    package Toto_Instance is new Toto (T_Item, Copy);
>> end Specialized_Toto;
>> ...
> 
> True. But then the use is not uniform. A different generic package must 
> be selected for each case:
> 
> package Default_Pkg is new Specialized_Toto (T);
> package Special_Pkg is new Toto (T, Special_Copy);
> 
> The idiom I have posted earlier does not have this problem. But, 
> granted, may have others instead, e.g. requiring two instantiations 
> instead of one for each case.
> 
> All this would go away if packages were first class citizens of Ada, 
> i.e. objects. This is one thing I'm trying to do with Ocpotus 
> (softdevelcoop.org/ocpotus).

Or

- if true package specialization were available;

- if generic types were ADT (you could have generic formal types of your
own);

- if assignment were a primitive operation (discriminants/constraints
problem is solvable, IMO);

- etc.

Packages as objects is a fundamental question I believe. A potential danger
is that when packages become objects any difference between "type" and
"package" might disappear. The result would be a type system close to C++.
But we know that C++ has immense problems in the elaboration model, or
better to say with absence of any model. Is it curable? I can't tell. If
Robert Dewar were here, he would probably shed more light on the mechanics.

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

Re: generic parameter Copy for primitifs types.

[Marius Amado Alves]

>> All this would go away if packages were first class citizens of Ada,
>> i.e. objects. This is one thing I'm trying to do with Ocpotus
>> (softdevelcoop.org/ocpotus).
>
> Or
>
> - if true package specialization were available;
>
> - if generic types were ADT (you could have generic formal types of 
> your
> own);
>
> - if assignment were a primitive operation (discriminants/constraints
> problem is solvable, IMO);
>
> - etc.
>

Yes.

> Packages as objects is a fundamental question I believe. A potential 
> danger
> is that when packages become objects any difference between "type" and
> "package" might disappear.

I don't see it as a danger, but as something I want.

> The result would be a type system close to C++.

I don't see this. C++ has the same schizophrenia between OO and 
generics that Ada has.

> But we know that C++ has immense problems in the elaboration model, or
> better to say with absence of any model. Is it curable? I can't tell. 
> If
> Robert Dewar were here, he would probably shed more light on the 
> mechanics.

I think it's curable. However it may require relaxing the notion of 
static typing. Namely generalizing the notion of type checking w.r.t. 
bind times. End another aporia, that of static vs. dynamic, or compile 
time vs. run time. Again, Ocpotus. But we drift away from Ada. Better 
discuss this on the Ocpotus list, or privately.

Re: generic parameter Copy for primitifs types.

[Martin Krischik]

Hyman Rosen wrote:

> Dmitry A. Kazakov wrote:
>> But we know that C++ has immense problems in the elaboration model, or
>> better to say with absence of any model.
> 
> All C++ types are static - no type depends on a runtime value.
> So elaboration doesn't really mean anything for C++.
> What are these "immense problems" to which you refer?

Try:

class X
  {
  virutal inline F ();
  }

inline X::F()
  {
  return;
  }

Then use X from more then one object and or try to export/import it from a
DLL and or compile it with various C++ compilers.

Have fun.

Martin
-- 
mailto://krischik@users.sourceforge.net
Ada programming at: http://ada.krischik.com

Re: generic parameter Copy for primitifs types.

[Christoph Grein]

nblanpain@hotmail.com wrote:

>Hello,
>
>this is my problem :
>
>-----
>generic
>type T_Item is private;
>with procedure Copy (Left : in out T_Item; Right : in T_Item);
>package Toto is
>...
>end Toto;
>------
>
>At instanciation, for T_Item = Integer for example, is there a methode
>to take for Copy => ":=". Must I redefined Copy ? Can I say that, by
>default, take ":=" ?
>
As you should know by now from all the answers there is no way to avoid 
defining Copy.

As also others have pointed out, T_Item should probably be limited 
private. As it stands,
Toto has := and Copy:

package body Toto is
  ...
  X := Y;      -- Do you really want both?
  Copy (X, Y);  -- which to use?

If T_Item is limited private, := is no longer available in Toto.

What you perhaps don't know is that the actual type for a limited formal 
type need not be limited. If you say

generic
  type T_Item is limited private;

then you can take any definite type as actual.

Re: generic parameter Copy for primitifs types.

[Matthew Heaney]

Marius Amado Alves <amado.alves@netcabo.pt> writes:

> generic
>     type T is private;
> package Level_1 is
>     type Copy_Ptr is access procedure (From : in T; To : out T);
>     procedure Default_Copy (From : in T; To : out T);
>     generic
>        Copy : Copy_Ptr := Default_Copy'Access;
>     package Level_2 is
>        ...
>     end;
> end;

The access type is unnecessary.  The feature you want here has been
available since Ada83:

generic
   type T is private;
package P is
   procedure Default_Copy (From : in T; To : out T);

   generic
      with procedure Copy (From : in T; To : out T) is Default_Copy;
   package Q is ...;
end P;

Re: generic parameter Copy for primitifs types.

[Marius Amado Alves]

> The access type is unnecessary.  The feature you want here has been
> available since Ada83:
>
> generic
>    type T is private;
> package P is
>    procedure Default_Copy (From : in T; To : out T);
>
>    generic
>       with procedure Copy (From : in T; To : out T) is Default_Copy;
>    package Q is ...;
> end P;

Of course. What was I thinking? It's right there on RM95 12.6(2). I 
hope the original poster is still listening, as surely this is the best 
idiom.

I know what I was thinking. I was under the influence of a recent case 
of (non-generic) subprograms with access-to-subprogram parameters with 
defaults. Here you do need access types in Ada 95. Will the anonymous 
access-to-subprogram types of Ada 2006 allow defaults?