Re: OO in Ada

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

Jim Rogers wrote:

> Dmitry A.Kazakov wrote:
> 
>> Jim Rogers wrote:
>> 
>>>I disagree here. Ada does not give the illusion of multiple dispatch.
>>>
>> Would "limited multiple dispatch" sound better?
>> 
>>>An Ada subprogram can only be primitive to one type. That characteristic
>>>is determined by the place in the source code where the subprogram
>>>interface is defined.
>> 
>> This is wrong. Technically you could do it, because type declaration
>> areas [before freezing point] may overlap:
> 
> I must again diagree. You are wrong. Section 3.9 paragraph 12 of the Ada
> Reference Manual states:
> 
> "A given subprogram shall not be a dispatching operation of two or more
> distinct tagged types."

You missed the point. I didn't challenged this rule. I said that it is 
wrong to think that this rule might be a consequence of the rules 
controlling the place in the source code where a primitive operation is 
allowed to be declared. I showed that this is wrong. For two different 
types there can be a place where one could define a primitive operation for 
any of them:
 
>> type A is tagged ...
>> type B is tagged ...
>> procedure Multiple_Dispatch (X : A; Y : B); -- Alas, compilation error

> There is a compilation error. Look at what it says. Gnat 3.14p states
> "operation can be dispatching in only one type."

Yes, but not because of "a primitive operation is declared too late". 
That's the point.

> This does not provide any for of multiple dispatch, limited or otherwise.

This does not, because it is illegal. But Ada 95 does have multiple 
dispatch, because of:

type A is tagged ...
procedure Multiple_Dispatch (X, Y : A);

This is a case of multiple dispatch because more than one parameter is 
dispatching. It is a limited multiple dispatch because:

1. In the dispatch table [note, table, not a vector like in C++! *] of 
Multiple_Dispatch only diagonal elements can be overridden. Non-diagonal 
elements are final and predefined as raising Constraint_Error.

2. The types associated with the rows and columns of the table have to have 
a common ancestor.

(*) Of course, you can implement this table using a vector.

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