Re: [Swig-devel] CVS version of swig does not unload modules correctly.

[Marcelo M. <mm...@ac...>]

Well, I added a method to destroy the module to pyrun.swg:

  SWIG_Python_DestroyModule()

and it is destroying the python part, ie, the ty->cliendatas properly,
which seems to be the origin of the reported error. For the other
part that you mention, I have no idea what are you talking about :),
so, I leave it to you for whenever you have time. But I don't
know if is really needed anyway.

Marcelo



John Lenz wrote:

>On Mon, November 7, 2005 5:07 am, Marcelo Matus said:
>  
>
>>Better yet, submit the bug and a send us a patch :).
>>
>>Yes, swig never unload the modules, probably because nobody asked
>>for that before. So, if you, or somebody else, know how to unload modules,
>>please send us a patch.
>>
>>I guess, it could be important for debugging porpuses, as you are doing.
>>We usually use valgrind and mainly ignores all the messages comming
>>from the 'unloading' part.
>>
>>    
>>
>
>On a somewhat unrelated note, the new runtime type system I wrote I kept
>unloading modules in mind.  I never wrote the specific code, but since the
>swig_module stores the original arrays, you can go through every type and
>every cast in the sytem, and check if it is in the original array from the
>module you are unloading.  If so, you remove it.  If it is a
>swig_type_info, you just pick one of the modules represented in the list
>of casts that will be still around, go to that modules initial
>swig_type_info, and insert that.
>
>So assuming the rest of the unloading works (unregistering functions, etc)
>it wouldn't be too hard to unload the module from the runtime system.  I
>haven't written the code because no one has expressed any need for it,
>but....
>
>John
>
>  
>