visibility of private incomplete types

visibility of private incomplete types

[Georg Bauhaus]

hello,

is GNAT correct in compiling the following program
without complaints? (There is a complaint if the
function in the body of P is uncommented.)

package p is

   package inner is
   private
      type TI;
   end inner;

private
   type TI_Ptr is access constant inner.TI;
end p;

package body p is

   -----------
   -- inner --
   -----------

   package body inner is

      type TI is null record;

   end inner;

--   function test return TI_Ptr is
--   begin
--      return new inner.TI;
--   end test;

end p;


with p; 
procedure p_test is 
begin null; end; 

Re: visibility of private incomplete types

[Martin Dowie]

Georg Bauhaus wrote:
> is GNAT correct in compiling the following program
> without complaints? (There is a complaint if the
> function in the body of P is uncommented.)

No - according to ObjectAda v 7.2.2

Re: visibility of private incomplete types

[Randy Brukardt]

"Martin Dowie" <martin.dowie@baesystems.com> wrote in message
news:4181108a_1@baen1673807.greenlnk.net...
> Georg Bauhaus wrote:
> > is GNAT correct in compiling the following program
> > without complaints? (There is a complaint if the
> > function in the body of P is uncommented.)
>
> No - according to ObjectAda v 7.2.2

Right, because the outer private part doesn't have visibility on the private
part of Inner. Only the body of Inner should be able to see TI.

              Randy.

Re: visibility of private incomplete types

[Nick Roberts]

Randy Brukardt wrote:

>>>is GNAT correct in compiling the following program
>>>without complaints? (There is a complaint if the
>>>function in the body of P is uncommented.)
>>
>>No - according to ObjectAda v 7.2.2
> 
> Right, because the outer private part doesn't have visibility on the
> private part of Inner. Only the body of Inner should be able to see
> TI.

Indeed.

Also, would I be right in saying that, in Ada 95, it is not permitted
for a type declared (by an incomplete type declaration) in the private
part of a package to be completed in the package's body?

Is this going to be permitted in Ada 2005?

-- 
Nick Roberts

Re: visibility of private incomplete types

[Frank J. Lhota]

"Nick Roberts" <nick.roberts@acm.org> wrote in message 
news:2uf1loF27r1iqU1@uni-berlin.de...
> Randy Brukardt wrote:
> Also, would I be right in saying that, in Ada 95, it is not permitted
> for a type declared (by an incomplete type declaration) in the private
> part of a package to be completed in the package's body?

No, this was permitted as early as Ada 83, and I used this technique many 
times. For example, assume that we are writing a package that implements 
widgets. Widgets will be implemented as pointers to a record, but of course 
we will hide this implementation detail using a private type. We can also 
hide the structure of this record type as follows:

    package Widgets is

        type Widget_Type is limited private;

        procedure Open( Widget : in out Widget_Type; Name : in String );
        procedure Close( Widget : in out Widget_Type );
        ...

    private

        type Widget_Data;
        type Widget_Type is access Widget_Data;

    end Widget;

This type of package specification in legal in both Ada 83 and Ada 95 (and 
presumably in Ada 200x), for Widget_Data is never used in a way that 
requires a full declaration. Of course, Widget_Data must be fully declared 
in the package body.

Re: visibility of private incomplete types

[Nick Roberts]

Frank J. Lhota wrote:

>>Also, would I be right in saying that, in Ada 95, it is not permitted
>>for a type declared (by an incomplete type declaration) in the private
>>part of a package to be completed in the package's body?
> 
> No, this was permitted as early as Ada 83, and I used this technique many 
> times. For example, assume that we are writing a package that implements 
> widgets. Widgets will be implemented as pointers to a record, but of course 
> we will hide this implementation detail using a private type. We can also 
> hide the structure of this record type as follows:
> 
>     package Widgets is
> 
>         type Widget_Type is limited private;
> 
>         procedure Open( Widget : in out Widget_Type; Name : in String );
>         procedure Close( Widget : in out Widget_Type );
>         ...
> 
>     private
> 
>         type Widget_Data;
>         type Widget_Type is access Widget_Data;
> 
>     end Widget;
> 
> This type of package specification in legal in both Ada 83 and Ada 95 (and 
> presumably in Ada 200x), for Widget_Data is never used in a way that 
> requires a full declaration. Of course, Widget_Data must be fully declared 
> in the package body.

I have to laugh. I now recall that I have used this technique /myself/
in the past. What a memory! RM95 3.10.1 (3). Sorry.

-- 
Nick Roberts

Re: visibility of private incomplete types

[Dr. Adrian Wrigley]

> Frank J. Lhota wrote:
...example omitted
>> This type of package specification in legal in both Ada 83 and Ada 95
>> (and presumably in Ada 200x), for Widget_Data is never used in a way
>> that requires a full declaration. Of course, Widget_Data must be fully
>> declared in the package body.

At last!  An answer to a long-standing annoyance I found with the language...
how to put the full type declarations (implementations) into the package
body, where they belong!

I had thought it was meant to be possible, but had never seen exactly how.
(presumably this one way to get physical isolation between different
components in a large system, so changes to data types don't trigger
recompilation of everything)

time to tidy up some old code of mine...
Thanks for pointing this out, guys!
-- 
Adrian
(maybe I'd better also (re-)read the ARM and those books gathering dust...)

Re: visibility of private incomplete types

[Lionel Draghi]

Dr. Adrian Wrigley a ï¿œcrit :
...
> At last!  An answer to a long-standing annoyance I found with the language...
> how to put the full type declarations (implementations) into the package
> body, where they belong!
...
 > (maybe I'd better also (re-)read the ARM and those books gathering
 > dust...)

Wouldn't it be a good idea to start a wiki on Ada idioms and patterns to 
gather those small to mid sized tips?

-- 
Lionel Draghi

Re: visibility of private incomplete types

[Alexander E. Kopilovich]

Lionel Draghi wrote:

>Wouldn't it be a good idea to start a wiki on Ada idioms and patterns to 
>gather those small to mid sized tips?

This is certainly the most heretical idea I have seen in comp.lang.ada for
several years! -;)




Alexander Kopilovich                      aek@vib.usr.pu.ru
Saint-Petersburg
Russia

Re: visibility of private incomplete types

[Martin Krischik]

Lionel Draghi wrote:

> Dr. Adrian Wrigley a ï¿œcrit :
> ...
>> At last!  An answer to a long-standing annoyance I found with the
>> language... how to put the full type declarations (implementations) into
>> the package body, where they belong!
> ...
>  > (maybe I'd better also (re-)read the ARM and those books gathering
>  > dust...)
> 
> Wouldn't it be a good idea to start a wiki on Ada idioms and patterns to
> gather those small to mid sized tips?

Well, what are you waiting for:

http://en.wikibooks.org/w/wiki.phtml?title=Programming:Ada:Tips&action=edit

The good thing about wiki is that you don't need to wait for somebody else. 

Martin
-- 
mailto://krischik@users.sourceforge.net
http://www.ada.krischik.com

Re: visibility of private incomplete types

[Georg Bauhaus]

Randy Brukardt <randy@rrsoftware.com> wrote:
: "Martin Dowie" <martin.dowie@baesystems.com> wrote in message
: news:4181108a_1@baen1673807.greenlnk.net...
:> Georg Bauhaus wrote:
:> > is GNAT correct in compiling the following program
:> > without complaints? (There is a complaint if the
:> > function in the body of P is uncommented.)
:>
:> No - according to ObjectAda v 7.2.2
: 
: Right, because the outer private part doesn't have visibility on the private
: part of Inner. Only the body of Inner should be able to see TI.

Thanks.
 
(PR #18227)