From: emilie3012@... (Steve Hart)
Date: 2003-04-10T09:01:55+09:00
Subject: Re: embedding ruby with swig

"Simon Strandgaard" <0bz63fz3m1qt3001@sneakemail.com> wrote in message news:<pan.2003.04.09.11.06.37.642046@sneakemail.com>...
> On Tue, 08 Apr 2003 19:56:33 +0000, Steve Hart wrote:
> 
> > The scripts are loaded into a stack within my c++ code
> > I have a c++ class 'GMScript' and have used swig to wrap into ruby
> > I use a ruby script that inherits this class
> > i.e.
> > class Test1 < Gm::GMScript
> > 
> >         def initialize(sg,p1,p2)
> > 
> >                 super(sg,p1,p2)
> > 
> >                 @object1 = Gm::GMObjectNode.new("object1")
> >         end
> >         .
> >         .
> >         .
> > end
> 
> I had some trouble with extending a SWIG-class, it resulted 
> in a exception: nameerror. 
> You are using inheiritance, no problems here.
> 
>  
> > in c++ I call at start
> > 
> >         ruby_init();
> >         ruby_init_loadpath();
> >         ruby_script("embedded");
> >         rb_require("gm.so");            - swigged up wrapper lib
> 
> Look fine.
> 
>  
> > then each time a script load is requested i do the following (protect
> > code removed for clarity)
> > 
> >         rb_load_protect(rb_str_new2("Test1.rb"),Qfalse,&state);
> > 
> > then
> >         oClass = rb_const_get(rb_cObject, rb_intern("Test1"));
> >         scriptObject = rb_funcall(oClass, rb_intern("new"), 
> >                                   3,sgValue,pathName,className);
> >         Data_Get_Struct(scriptObject,GMScript,scriptPtr);
> >         scriptPtr->obj = scriptObject;
> 
> This code could be the problem. 
> * bootstrapping of your swig class.
> * you are not registering scriptObject, so GC won't know that its
>   in usage and thus thinks it must get killed. 
>   
> How does your SWIG interface file look like ???
> If it includes a .H file how does it look like ???
> 
> 
> > The problem is that as I load more scripts previous scripts get freed
> > of by the GC. In addition, the GMObjectNode's created within the
> > Scripts are also freed. 
> 
> Sounds like you object is holding ruby-childen. 
> You will need to do rb_mark on thise.
> 
> 
> > What I want to do is only free off the Script
> > when the user deletes it
> > from the script stack and decide at this time if the GMObjectNode
> > should
> > also be deleted. I assumed that since the scriptObject is stored
> > within the C code the GC would not free these off (or is it because
> > the scriptObject is stored in GMScript superclass itself. 
> 
> Wrong assumption. It depends on wheter or not you have registered
> your scriptObject.
> 
> 
> > I cannot
> > mark the object since this is under the control of Swig. I have tried
> > 
> >         rb_global_variable(&scriptObject)
> > 
> > but this does not prevent the freeing and in addition crashes the
> > program with an error:
> 
> How did you "try" it ???
> 
> 
> Read my text on embedding, look at the tar.gz file :-)
> http://metaeditor.sourceforge.net/embed/

Thanks for reply - i looked at your tgz last night (as well as vim and
mod_ruby.)
Thats how I've got this far!
Sorry for denseness but what does bootstrapping swig class mean and
why is it bad. ?
I have had a chance to look into problem further and found swig now
supports
user defined mark and free functions. I have defined these in the
interface file
but still no joy.

This is a cut down version of my code:

gm.i:
^^^^

%module gm

%{
#include "GMScript.h"
%}

%{
static void GMScript_freefunc(void *ptr)
{
        cerr << "FREE\n";
        GMScript *script = (GMScript *) ptr;
        delete script;
}

%}

%{
static void GMScript_markfunc(void *ptr)
{
        cerr << "MARK\n";
        GMScript *script = (GMScript *) ptr;
        rb_gc_mark(reinterpret_cast<VALUE>((script->scriptObj)));
}
%}

%markfunc GMScript      "GMScript_markfunc"
%freefunc GMScript      "GMScript_freefunc"

%include GMScript.h


GMScript.h
^^^^^^^^^^

#ifndef _GMSCRIPT_H
#define _GMSCRIPT_H

#include <iostream>
using namespace std;

#include <ruby.h>

class GMScript
{
public:

        GMScript();
        ~GMScript();

        VALUE           scriptObj;
};

#endif


GMScript.cpp
^^^^^^^^^^^^

#include "GMScript.h"

/*****************************************/
GMScript::GMScript()
{
        cerr << "GMScript::GMScript\n";
}

/*****************************************/
GMScript::~GMScript()
{
        cerr << "GMScript::~GMScript\n";
}


main.c
^^^^^^

int main()
{
        GMScript                *ptr = NULL;
        int                     state = 0;
        VALUE                   vTclass;
        VALUE                   tObj;
        vector<GMScript *>      stack;

        ruby_init();
        ruby_script("embedded");
        ruby_init_loadpath();

        rb_require("gm.so");

        for(;;)
        {
                rb_load_protect(rb_str_new2("Class.rb"),0,&state);
                vTclass = rb_const_get(rb_cObject,
rb_intern("MyClass"));
                tObj = rb_funcall(vTclass, rb_intern("new"), 0);

                Data_Get_Struct(tObj,GMScript,ptr);
                ptr->scriptObj = tObj;

                // save on stack
                stack.push_back(ptr);

                // see what it does
                rb_gc();

                sleep(1);
                cerr << "\n";
        }

        return 0;
}

and I get the following out put

GMScript::GMScript
FREE
GMScript::~GMScript

GMScript::GMScript
FREE
GMScript::~GMScript

GMScript::GMScript
FREE
GMScript::~GMScript

GMScript::GMScript
MARK

GMScript::GMScript
FREE
GMScript::~GMScript
FREE
GMScript::~GMScript

GMScript::GMScript
MARK

GMScript::GMScript
MARK
FREE
GMScript::~GMScript
..
..
..

So, the mark is called - but so is free. I did find that if I use a
ruby array instead of vector, push tObj and call rb_global_variable on
this array the free never gets called. This leads me to ask if the
scriptObj returned from rb_funcall is the correct value to mark.
If not, then how do I obtain the correct corresponding ruby VALUE for
ptr . In fact, I need to do this to mark children held by
GMScript::MyClass?
Do I need to do both mark AND rb_global_variable?

Cheers in advance
Steve