Re: [announcement] SYSAPI and SYSSVC for Windows

From: Dmitry A. Kazakov (mailbox_at_dmitry-kazakov.de)
Date: 12/21/03 

Date: Sun, 21 Dec 2003 14:42:09 +0100

Ekkehard Morgenstern wrote:

> "Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> wrote:
>> The problem is to tell the customers, that *their* applications should be
>> designed so that *our* DLLs and hardware drivers would not stray into an
>> unpredictable state.
>
> What's the problem communicating this to the customer?

We are not MS, to tell customers, that they are wrong. (:-))

>> > C++ is well-defined also, btw. There's already a number of ISO
>> > standards for C and C++,
>>
>> A *number of* standards, fascinating! (:-))
>
> For Ada, there's also an '83 and a '95 rationale. ;)

Ada 95 practically includes Ada 83, with very minor exceptions. Rationale is
not the standard, it is rather comments to the standard. BTW, there is also
AARM - Annotated Ada Reference Manual.

>> Come on! C++ does not have:
>>
>> 1. multiple dispatch;
>
> True, except for constructors and destructors. They're always called for
> all derived classes of an object.

This is not multiple dispatch. MD is when, for example:

type Matrix is tagged ...;
function "+" (X, Y : Matrix) return Matrix;
   -- This is dispatching in X, Y and the result

In C++ only one (hidden) parameter can be dispatching. All others are
contravariant, which leads to numerous nasty pitfalls

>> 3. dispatch on access types;
>
> False. Method overloading based on parameters of different types, even
> pointer types, is supported.

Overloading, not overriding. In C++:

class X {...};
void f(X * Ptr); // This is not a member of X!

Formally f is not dispatching in Ptr. It is class-wide. Compare it with Ada;

type X is tagged ...
procedure F1 (Object : access X); -- This is a "member"
procedure F2 (Object : access X'Class); -- This is not

type Y is new X with ...
procedure F1 (Object : access Y); -- This overrides F1
procedure F2 (Object : access Y'Class); -- This overloads F2

>> 4. differentiation between class-wide and specific types;
>
> Class-wide types in Ada roughly correspond to base classes in C++.

No. A class-wide type is a closure of the derived types. C++ tries to
equalize specific and class-wide types. The price is that C++ never become
a fully OO langauge, while Ada can do it easily. It possible to make all
types in Ada "tagged", even Boolean.

>> 5. functions returning class-wide objects on the stack;
>
> True, but you can return a reference to an object. (I'm not sure if
> passing a reference to a temporary object would work)

This won't help, because of 4. The same type name is sometimes denotes
class-wide, sometimes specific in C++. In declarations it is always
specific, so you cannot do:

declare
   X : Object'Class := Get_From_File; -- Any descendant of Object

To achive same effect, you have to use the heap:

   X_Ptr * Object = Get_From_File (); // internally calls new
   ...
   free (X_Ptr); // do not forget to release it

>> 7. formal derived type parameter for generics;
>
> False. Of course you can use derived classes or other types in templates.

Georg Bauhaus has answered this

>> Why? The following is OK:
>>
>> type T is limited private;
>> private
>> type T is limited record
>> I : aliased Integer;
>
> Yes, but I is an Integer. Discrete types can always be aliased.
>
> In my case, it was an array of aliased types, and I wanted to create an
> access to an array element.

>> "Aliased" is the keyword for that. Objects of some types are always
>> aliased:
>>
>> type T is tagged limited ...;
>
> so I could've just used a tagged limited record?
>
> Like this:
>
> type T is tagged limited
> record
> A : My_Array_Type;
> end record;
>
> procedure F ( O : in out T ) is
> Ptr : My_Array_Cell_Ptr;
> begin
> Ptr := O.A(1)'Access;
> end;
>
> Right?

No. It is unrelated. The thing you are getting access of has to be aliased.
So it is the array elements which has to be, for example:

   type T is limited private;
private
   type Integer_Array is array (Integer range <>) of aliased Integer;
   type T is limited record
      A : Integer_Array (1..3);
   end record;

   procedure F (O : in out T ) is
   begin
       ... O.A(1)'Access; -- This is OK, A(i) are aliased
   end;

If the array elements be tagged, then you would need not write "aliased" in
the array declaration. 

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