Re: on Ada abtract data type vs. OOP.

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

On 2022-09-09 11:32, Nasser M. Abbasi wrote:
> Lets say one does not want to do inheritance. (I think it causes
> more problems than it solves actually).

You cannot in Ada where one can always inherit from any type in some way 
or another:

- tagged extension: type S is new T with ...;
- subtype constraining: subtype S is T;
- cloing: type S is new T;

> The only difference I see between Ada's ADT and OOP, is that
> in Ada, the data itself is separated from the package and must
> be passed to the package methods at each call. The data lives
> in the client side.

This applies to all types and has nothing to do with ADT or OO. A state 
can be either localized in an object (good design) or kept outside it in 
global variables (bad design).

> While in OOP, the data lives inside the object. (these are called
> data memebers).

Just like in any type, not even abstract one. E.g.

    type X is range 1..100;

The "data" live in each instance of X.

> Let look at this simple example from

[...] Stack example

> In OOP, the stack itself, would live inside the "module" which will
> be the class/object in that case.

No, OOP example of stack is exactly the one you cited. There is a type 
Stack and operations of. A non-OO/ADT stack would be:

generic -- No parameters!!
package Generic_Integer_Stack is
    procedure Push (Val : Integer);
    procedure Pop (Val : out Integer);

You get a stack instance this way:

    package Integer_Stack is new Generic_Integer_Stack;

> So the main difference I see, is that in Ada ADT, the data (the stack
> in this example) lives on the client side, and the package
> just has the methods.

The sentence does not make sense to me. OO is ADT. I am not sure which 
issue you have problem with:

- No local states
- Stateful vs stateless
- Interface vs implementation inheritance

> For me, this actually better than OOP. Having methods separated
> from data is a good thing.

You presented a perfectly OO design of a stack type. A better one would be:

package Integer_Stacks is
    type Stack is tagged limited private;
    procedure Push (S : in out Stack; Val : Integer);
    procedure Pop (S : in out Stack; Val : out Integer);

Here:

- limited because we do not want copy or compare stacks
- tagged because we might want to reuse the type implementation.

For example:

with Integer_Stacks;  use Integer_Stack;

package Integer_Signaled_Stacks is
    type Signaled_Stack is new Stack with private;
    procedure Wait_For_Not_Empty
             (Stack : in out Signaled_Stack; Timeout : Duration);
private
    overriding
       procedure Push (S : in out Stack; Val : Integer);
    overriding
       procedure Pop (S : in out Stack; Val : out Integer);

Here the stack maintains a lock to make it task safe and provides event 
to wait for non-empty stack.

> Is there is something I am overlooking other than this? Again,
> assuming one does not want to do inheritance?

You cannot. Unless you use an extremely primitive language like C some 
form of inheritance is always there.

> It seems to me that Ada ADT provides all the benefits of OOP and more,
> as it does not mix data and methods inside one container.

Can you explain what do you mean under mixing data with methods? Ada is 
very limited in terms of using subprograms as data. Basically you need 
to resort to pointers or generic formals. You certainly meant something 
else.

> What do other think about this subject?

- Ada type system needs an overhaul.
- People confuse OOP with OOA&D. OOP is merely a better ADT.
- Ada 83 was object-based and its ADT was quite weak.
- Ada 95 fixed that, but stopped at single inheritance and dispatch and 
C++-esque idea of having types (AKA classes) and not so much types (AKA 
everything else we do not know how to deal right).
- Ada 2005 added castrated Java-esque multiple dispatch

Nothing happened to the type system since.

> Do you think it is
> better to do it as OOP, to have the data inside the object,
> or like with ADT, where the data instances are on the client side?

It is always preferable not to have global variables.

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