From: Paul Brannan <pbrannan@...>
Date: 2002-04-19T23:50:01+09:00
Subject: Re: GC problem in C++ extension

On Fri, Apr 19, 2002 at 12:00:40PM +0900, Yukihiro Matsumoto wrote:
> Unfortunately, GC timing is not controllable, even if you call
> GC.start explicitly, partly due to the conservative nature of Ruby's
> GC.  All you can expect is that destructors are called before process
> termination.

Hmm... if I change line 327 of gc.c from:

  VALUE obj;

to:

  static VALUE obj;

then my program works.  It seems that the problem is rb_newobj() is
keeping obj on the stack, and then I'm never writing over that location.
If I add a new function to my test program:

  void clear_my_stack(void) {
      void * x = alloca(10000);
      memset(x, 0, 10000);
  }

and call it after the Data_Wrap_Struct, then everything gets cleaned up
properly (without the change to gc.c).

Paul