Re: Which compiler is correct?

[bobduff@world.std.com (Robert A Duff)]

In article <JSA.97Jun20185414@alexandria.organon.com>,
Jon S Anthony <jsa@alexandria.organon.com> wrote:
>In article <ud205z1ifw.fsf@tanana.llnl.gov> woodruff@tanana.llnl.gov (John Woodruff) writes:
...
>> package Parent.Child is
>>    type Good_Obj is new Parent.Obj with private ;
>>    type Bad_Obj  is new Parent.Obj with null record ;
>
>NOTE: at this point
>
>a) Parent's i component is _not_ visible and so not available to this
>view of the children.

Right.

>b) Bad_Obj is _fully_ declared at this point and so will never get the
>implicit declaration of Parent's i component.

No, that's not quite right.

>>    procedure Do_Nothing ;  -- just to require body

By the way, "pragma Elaborate_Body;" is a good way to require a body,
when it's not otherwise required.

>> private

At *this* point, an "additional characteristic" (see 7.3.1(3-4)) of
Parent.Obj (namely, the existence of I) becomes visible, and since this
is within the immediate scope of Bad_Obj, it gets an I, too.  By magic.

Note: There's an AI that says this happens only *immediately* within the
immediate scope of Bad_Obj, to match Ada 83.  (The wording of RM95
accidentally lets it happen *anywhere* within the immediate scope of
Bad_Obj, which is wrong.)  That's irrelevant to this example.

>>    type Good_Obj is new Parent.Obj with null record ;
>
>At this point, Parent.I is visible and the full declaration of
>Good_Obj implicitly declares this component.
>
>
>> end ;
>> 
>> One of the compilers (compiler "r") is able to compile the body without
>> error; the other compiler reports an error in the body when I try to
>> reference the component of the parent type.
>> 
>> package body Parent.Child is
>>    good : Good_Obj ;
>>    Bad  : Bad_Obj ;
>>    procedure Do_Nothing is
>>    begin
>>       null;
>>    end ;
>> begin
>>    Good.I := 2 ;
>>    Bad.I := 2 ;  -- compiler "g" reports error
>> end ;
>
>So, Bad.I is not directly available and compiler "g", I believe, is
>correct.

No.

>> selector".  I am unable to think of a reason why the body of the child
>> package can reference the parent's private declaration of Obj for one,
>> but not for the other derived types.
>
>I think this all falls out of the rules in 7.3.1, and in particular
>pay close attention to what is said in (3).

Closer attention than that.  ;-)

- Bob