Interfacing C++ classes, but what about the destructor?

Interfacing C++ classes, but what about the destructor?

[Andreas Almroth]

Hi all,

I have been searching google, and read the Gnat RM&UG, and I have some
working code interfacing C++ classes. No problem really so far, although
 it feels as if I'm wading in unchartered waters... Documentation seems
to be sparse, although the documentation that is there pretty much shows
you how to do it...

My question is though on the topic of C++ class destructors. When I
search, I find references to CPP_Destructor pragma in some
documentation, but when I compile the code, I get a warning stating that
the pragma is unknown. I have tried GNAT 3.15p, gcc 3.4.2 and gcc 4.0.1,
all the same.

As the pragma doesn't seem to exist, contrary to older documentation, I
assume it it has been removed due to it doesn't work, or was deemed
unneccesary.

Regardless of reasons, I 1) wonder the history behind this, 2) how to
call the destructor of the C++ class from Ada side, or why I shouldn't.

Well, any pointers is appreciated...

Regards,
/Andreas

Re: Interfacing C++ classes, but what about the destructor?

[Brian May]

>>>>> "Andreas" == Andreas Almroth <andreas_no_spam@almroth.com> writes:

    Andreas> Hi all, I have been searching google, and read the Gnat
    Andreas> RM&UG, and I have some working code interfacing C++
    Andreas> classes. No problem really so far, although it feels as
    Andreas> if I'm wading in unchartered waters... Documentation
    Andreas> seems to be sparse, although the documentation that is
    Andreas> there pretty much shows you how to do it...

I think you have already done way more then most of us - I seem to
remember the recommended approach is usually to write a C layer
between the Ada layer and C++ layer.

    Andreas> My question is though on the topic of C++ class
    Andreas> destructors. When I search, I find references to
    Andreas> CPP_Destructor pragma in some documentation, but when I
    Andreas> compile the code, I get a warning stating that the pragma
    Andreas> is unknown. I have tried GNAT 3.15p, gcc 3.4.2 and gcc
    Andreas> 4.0.1, all the same.

I can't see any documentation on the CPP_Destructor call here (version
3.4 info pages).

    Andreas> As the pragma doesn't seem to exist, contrary to older
    Andreas> documentation, I assume it it has been removed due to it
    Andreas> doesn't work, or was deemed unneccesary.

    Andreas> Regardless of reasons, I 1) wonder the history behind
    Andreas> this, 2) how to call the destructor of the C++ class from
    Andreas> Ada side, or why I shouldn't.

Sorry, I suspect you may not get much help here :-(.

The example in my copy of the UG "2.11.3 A Simple Example" appears to
use C bindings, not C++ bindings, even though the language is C++. A
good example that uses C++ bindings would be appreciated...

However, I would speculate that a virtual destructor (remember
destructor don't need to be virtual, but it seems like a good idea) is
treated like any other virtual function.

This might imply GNAT will not call the destructor automatically for
you.
-- 
Brian May <bam@snoopy.apana.org.au>

Re: Interfacing C++ classes, but what about the destructor?

[Andreas Almroth]

Brian May wrote:
>>>>>>"Andreas" == Andreas Almroth <andreas_no_spam@almroth.com> writes:
> 
> 
>     Andreas> Hi all, I have been searching google, and read the Gnat
>     Andreas> RM&UG, and I have some working code interfacing C++
>     Andreas> classes. No problem really so far, although it feels as
>     Andreas> if I'm wading in unchartered waters... Documentation
>     Andreas> seems to be sparse, although the documentation that is
>     Andreas> there pretty much shows you how to do it...
> 
> I think you have already done way more then most of us - I seem to
> remember the recommended approach is usually to write a C layer
> between the Ada layer and C++ layer.
> 

Yes, this is my understanding as well, after reading some more on the
topic today. The recommended approach is to use C layer to wrap methods
working on C++ classes or class methods. Which is what I have done as
well. And it works quite well, although, I must admit, I was a bit
surprised that passing an Ada object (not access to) to these functions
worked "out-of-the-box" as I declare a Object *p in the C functions.
But hey, as long as it works... All this pass-by-reference, pass-by-copy
is just confusing me anyway... :)

>     Andreas> My question is though on the topic of C++ class
>     Andreas> destructors. When I search, I find references to
>     Andreas> CPP_Destructor pragma in some documentation, but when I
>     Andreas> compile the code, I get a warning stating that the pragma
>     Andreas> is unknown. I have tried GNAT 3.15p, gcc 3.4.2 and gcc
>     Andreas> 4.0.1, all the same.
> 
> I can't see any documentation on the CPP_Destructor call here (version
> 3.4 info pages).

Older documentation for GNAT 3.10 does mention the pragma. OK, I know
that version is _quite_ old...

> 
>     Andreas> As the pragma doesn't seem to exist, contrary to older
>     Andreas> documentation, I assume it it has been removed due to it
>     Andreas> doesn't work, or was deemed unneccesary.
> 
>     Andreas> Regardless of reasons, I 1) wonder the history behind
>     Andreas> this, 2) how to call the destructor of the C++ class from
>     Andreas> Ada side, or why I shouldn't.
> 
> Sorry, I suspect you may not get much help here :-(.
> 
> The example in my copy of the UG "2.11.3 A Simple Example" appears to
> use C bindings, not C++ bindings, even though the language is C++. A
> good example that uses C++ bindings would be appreciated...
> 

The simple example I have read is using a limited record to mimic a C++
class which has inherited from its parent. This approach is easy when
you don't have a huge class framework, or so I imagine.

I'm using CPP_Class pragma in order to retain the tagged<=>class
relationship. This way, to my understanding, I should be able to follow
the class tree in C++ with corresponding Ada tagged types.

> However, I would speculate that a virtual destructor (remember
> destructor don't need to be virtual, but it seems like a good idea) is
> treated like any other virtual function.
> 
> This might imply GNAT will not call the destructor automatically for
> you.

Quite right. I have worked on this some more, using a C layer wrapper
for the destructor, importing its mangled name, and calling it manually
in my Ada code. So, the neat feature with CPP_Constructor pragma will
have to be handled manually for the destructor. I will do some work on
this and calling the C++ destructor from my Finalize of my controlled
Ada type. I foresee a controlled type might interfere with the clean
mapping between my Ada type and the C++ class though.
Anyway,I hope this will work. Obviously not as neat as automating it
using a pragma, but will have the effect that I would like to achieve.

Thanks for your reply.

Regards,
Andreas