Re: exposing do_octave_atexit in the API

[c.]

On 27 Feb 2014, at 08:07, c. <address@hidden> wrote:
> On 11 Feb 2014, at 16:37, c. <address@hidden> wrote:
> 
>> Hi,
>> 
>> I'm embedding the interpreter in an MPI application,
>> so I need to call both clean_up_and_exit (0) and MPI_Finalize () at the and 
>> of my code.
>> 
>> Unfortunately whatever order I choose causes a crash when the application 
>> tries to exit.
>> 
>> By applying the patch below to toplev.{cc,h} I can di
>> 
>> 
>> do_octave_atexit ();
>> MPI_Finalize ();
>> return (0);
>> 
>> 
>> and have my application shut down gracefully.
>> Does anyone know whether this patch is harmful in any way?
>> Would it be OK to apply it in the repository? 
>> Is there a better way to achieve the same?
>> 
>> Thanks,
>> c.
>> 
>> 
>> diff --git a/libinterp/corefcn/toplev.cc b/libinterp/corefcn/toplev.cc
>> --- a/libinterp/corefcn/toplev.cc
>> +++ b/libinterp/corefcn/toplev.cc
>> @@ -662,7 +662,7 @@
>> 
>> static std::list<std::string> octave_atexit_functions;
>> 
>> -static void
>> +void
>> do_octave_atexit (void)
>> {
>>  static bool deja_vu = false;
>> diff --git a/libinterp/corefcn/toplev.h b/libinterp/corefcn/toplev.h
>> --- a/libinterp/corefcn/toplev.h
>> +++ b/libinterp/corefcn/toplev.h
>> @@ -485,4 +485,7 @@
>>    } \
>>  while (0)
>> 
>> +void
>> +do_octave_atexit (void);
>> +
>> #endif
>> 
> 


So, 

thanks to a suggestion  I got from jwe in IRC,
I used a cleaner solution that does not
require patching Octave, i.e. I defined 

void octave_mpi_exit (int)
{
  MPI_Finalize ();
}


and then used the following at the endo of my program:

 if (rank == 0)
    {
      octave_exit = octave_mpi_exit;
      clean_up_and_exit (0);
    }
  else
    MPI_Finalize ();

  return 0; 
}

There is currently an example in the source tree and in the manual
about embedding the interpreter, would it be useful if I submitted
a changeset to add an example about ebedding in a parallel application?

thanks,
c.