From: nobu.nokada@...
Date: 2004-10-18T18:11:59+09:00
Subject: Re: calling define_finalizer from within C?

Hi,

At Mon, 18 Oct 2004 15:54:24 +0900,
Andres Salomon wrote in [ruby-talk:116931]:
> 
> Hi,
> 
> I've got the following bit of C code;
>    static VALUE do_destroy(...) { ...; return Qnil; }
> 
>    cObjectSpace = rb_const_get(rb_cObject, rb_intern("ObjectSpace"));
>    rb_include_module(cFoo, cObjectSpace);
>    
>    /* ... somewhere inside constructor... */
>    p = rb_proc_new(do_destroy, obj);
>    fprintf(stderr, "created new proc; class %s\n", rb_obj_classname(p));
>    rb_funcall(obj, rb_intern("define_finalizer"), 1, p);
> 
> Basically, I need to define a finalizer from within a constructor, but I'm
> running into the following problems:
> 
> 1) define_finalizer throws an exception: 
> ArgumentError: tried to create Proc object without a block
> I'm not sure what would cause that.  The proc object should have a block..

Use rb_iterate().

  static VALUE mObjectSpace;

  static void
  define_final(VALUE obj)
  {
      rb_funcall(mObjectSpace, rb_intern("define_finalizer"), 1, p);
  }

  static VALUE
  do_initialize(VALUE obj)
  {
      rb_iterate(define_final, obj, do_destroy, Qnil);
      return Qnil;
  }

  void
  Init_foo()
  {
      mObjectSpace = rb_const_get(rb_cObject, rb_intern("ObjectSpace"));
  }

> 2) At least in irb, calling define_finalizer within a class, and passing
> it self as the first arg, causes nothing to happen.  Is define_finalizer
> the proper way to do this sort of thing, or is there a better way?  Note
> that I don't want to use Data_Make_Struct's free callback, as I need this
> to be overridable by users of the class.

Do you mean like this?

  ObjectSpace.define_finalizer(self) do |id|
    # ...
  end

If so, the object is referred as self from the finalizer so
that it never get freed.

Since not sure what you want really, I cannot tell what is
"better".  At least, finalizers are different from destructors.

-- 
Nobu Nakada