Gnat: broken dispatch tables for interface types ?

Gnat: broken dispatch tables for interface types ?

[rodakay]

Hi all,

   Has anyone had any success with interface types ?

   I recently added a function to an interface type. Now when I invoke a function of the interface type, a different function is called. Which function is actually called seems to depend on the order of declaration of the subprograms for the interface type.

   So it seems that the dispatch table for the interface is broken in Gnat ?

   I've tried building with both GPL12 and latest GCC from svn, with the same result.



regards,
Rod.

Re: Gnat: broken dispatch tables for interface types ?

[Dmitry A. Kazakov]

On Sat, 7 Jul 2012 20:02:36 -0700 (PDT), rodakay@internode.on.net wrote:

>  Has anyone had any success with interface types ?

Yes, of course. We are heavily using them in our projects.
 
> I recently added a function to an interface type. Now when I invoke a
> function of the interface type, a different function is called.

You should post a small example illustrating the case.

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

Re: Gnat: broken dispatch tables for interface types ?

[Jeffrey Carter]

> On Sat, 7 Jul 2012 20:02:36 -0700 (PDT), rodakay@internode.on.net wrote:
>
>   Has anyone had any success with interface types ?

According to Randy Brukardt, ARG member and editor of the ARM, "Interfaces are 
worthless."

-- 
Jeff Carter
"This school was here before you came,
and it'll be here before you go."
Horse Feathers
48



--- Posted via news://freenews.netfront.net/ - Complaints to news@netfront.net ---

Re: Gnat: broken dispatch tables for interface types ?

[Dmitry A. Kazakov]

On Sun, 08 Jul 2012 00:38:05 -0700, Jeffrey Carter wrote:

>> On Sat, 7 Jul 2012 20:02:36 -0700 (PDT), rodakay@internode.on.net wrote:
>>
>>   Has anyone had any success with interface types ?
> 
> According to Randy Brukardt, ARG member and editor of the ARM, "Interfaces are 
> worthless."

Yes, if he meant that interface is nothing but abstract tagged null record
plus some silly restrictions.

No, if he meant an ability to codify the interface of a type, multiple
inheritance, interface inheritance, generic programming.

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

Re: Gnat: broken dispatch tables for interface types ?

[rodakay]

On Sunday, 8 July 2012 16:53:30 UTC+10, Dmitry A. Kazakov  wrote:
> 
> Yes, of course. We are heavily using them in our projects.

   Thank you Dmitry, that's most encouraging. It did seem unlikely that I could be the only person using interface types :). (Probably, I am the only person using them poorly, though ;) ).


>  
> You should post a small example illustrating the case.

   Yes, indeed. I will try to set up a minimal test case in the next day or so.

Re: Gnat: broken dispatch tables for interface types ?

[rodakay]

Here is the 'cut down' code which shows the problem ...

http://www.assembla.com/spaces/openada/documents/bf0foSYzir4y2zacwqjQXA/download/bf0foSYzir4y2zacwqjQXA

Re: Gnat: broken dispatch tables for interface types ?

[Dmitry A. Kazakov]

On Sun, 8 Jul 2012 23:53:42 -0700 (PDT), rodakay@internode.on.net wrote:

> Here is the 'cut down' code which shows the problem ...

I see nothing wrong with dispatching tables. Though the code much too large
and complicated for a small example.

One thing though is that you should not derive new tagged types in bodies.
 
-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de

Re: Gnat: broken dispatch tables for interface types ?

[Rod Kay]

> One thing though is that you should not derive new tagged types in bodies.

So it seems. Moving the derivation to a spec and problem disappears. Many thanks ! 

Re: Gnat: broken dispatch tables for interface types ?

[Simon Wright]

"Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:

> One thing though is that you should not derive new tagged types in
> bodies.

As a matter of interest, why not?

Re: Gnat: broken dispatch tables for interface types ?

[Dmitry A. Kazakov]

On Mon, 09 Jul 2012 10:37:11 +0100, Simon Wright wrote:

> "Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:
> 
>> One thing though is that you should not derive new tagged types in
>> bodies.
> 
> As a matter of interest, why not?

It does not work well. I never analyzed the RM for the cases when the type
of an object could die before the object itself. But I certainly prefer not
to tease GNAT with these.

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

Re: Gnat: broken dispatch tables for interface types ?

[Simon Wright]

"Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:

> On Mon, 09 Jul 2012 10:37:11 +0100, Simon Wright wrote:
>
>> "Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:
>> 
>>> One thing though is that you should not derive new tagged types in
>>> bodies.
>> 
>> As a matter of interest, why not?
>
> It does not work well. I never analyzed the RM for the cases when the type
> of an object could die before the object itself. But I certainly prefer not
> to tease GNAT with these.

Perhaps you meant in subprogram bodies? I don't see how the life of a
type declared in a library-level package body can die.

Re: Gnat: broken dispatch tables for interface types ?

[Dmitry A. Kazakov]

On Mon, 09 Jul 2012 18:32:54 +0100, Simon Wright wrote:

> "Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:
> 
>> On Mon, 09 Jul 2012 10:37:11 +0100, Simon Wright wrote:
>>
>>> "Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:
>>> 
>>>> One thing though is that you should not derive new tagged types in
>>>> bodies.
>>> 
>>> As a matter of interest, why not?
>>
>> It does not work well. I never analyzed the RM for the cases when the type
>> of an object could die before the object itself. But I certainly prefer not
>> to tease GNAT with these.
> 
> Perhaps you meant in subprogram bodies? I don't see how the life of a
> type declared in a library-level package body can die.

Upon package finalization and elaboration when things get used prematurely.
It may work or fail in both directions.

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

Re: Gnat: broken dispatch tables for interface types ?

[Robert A Duff]

"Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:

> It does not work well. I never analyzed the RM for the cases when the type
> of an object could die before the object itself.

That can't happen.

If you don't believe me, try to construct an example.  If you succeed
then either the language or the compiler is broken.

- Bob

Re: Gnat: broken dispatch tables for interface types ?

[Vadim Godunko]

On Sunday, July 8, 2012 7:02:36 AM UTC+4, (unknown) wrote:
> 
>    So it seems that the dispatch table for the interface is broken in Gnat ?
> 
You can try to do type conversion before call, sometimes it helps to compiler to use correct dispatching table.

Re: Gnat: broken dispatch tables for interface types ?

[Rod Kay]

Thanks Vadim, I'll try it out.

Re: Gnat: broken dispatch tables for interface types ?

[sbelmont700]

On Saturday, July 7, 2012 11:02:36 PM UTC-4, (unknown) wrote:
>    So it seems that the dispatch table for the interface is broken in Gnat ?

I encountered a similar sounding bug (there is a thread about it somewhere), where GNAT would mangle the dispatch table when allocating (i.e. 'new') limited, controlled, classwide types.  IIRC, the only workaround I found was to allocate it as the concrete type and then convert it to the classwide type manually, but there were still cases in which things were tricky (i.e. functions returning classwide types had to be modified).  Originally it was for GNAT GPL 2011, but it appears as though it remains unfixed in GNAT 2012 (you would think it would be a dealbreaker for GNAT Pro users, but apparently not...)

-sb

Re: Gnat: broken dispatch tables for interface types ?

[Rod Kay]

   I seem to have found the cause of my problem.


   In one case, when an Observer registered itself with a Subject, I'd used the ".all'access" notation to convert from a derived Observer 'access' to the base Observer "access" ...

   Self.Keyboard.register_Observer (Self.all'access, ...);



   When I do the conversion explicitly, all is fine ...

   Self.Keyboard.add (lace.event.Observer.view (Self), ...);



   Thank you Vadim, for the tip.



   Also, it seems I managed to introduce a different error in my (not so) cut-down version of the code (ie deriving types within a body, rather than spec), which Dmitri pointed out. Would two bug reports be in order then ?


   Finally, thanks again to all for the prompt, helpful replies.

Re: Gnat: broken dispatch tables for interface types ?

[Simon Wright]

Rod Kay <rodakay@internode.on.net> writes:

> Also, it seems I managed to introduce a different error in my (not
> so) cut-down version of the code (ie deriving types within a body,
> rather than spec), which Dmitri pointed out. Would two bug reports be
> in order then ?

Yes. They may well be caused by different compiler issues.