From: Michael Bruschkewitz <brusch2@...>
Date: 2003-03-17T04:39:39+09:00
Subject: Re: method call on terminated object [W2k,1.6.8}

In article <200303141523.h2EFNbeU010813
@sharui.nakada.kanuma.tochigi.jp>, nobu.nokada@softhome.net says...
> Hi,
> 
> At Fri, 14 Mar 2003 23:21:26 +0900,
> Michael Bruschkewitz wrote:
> > Under which circumstances is an String "terminated"?
> Bug.
Neccessary feed for small birds.
> 
> > conn.read calls the following function from an own extension:		
> > 	VALUE TCP_Read(VALUE self, VALUE maxL) 	{
> > 		CRbTcp* p; Data_Get_Struct(self, CRbTcp, p);
> > 		int ml = NUM2INT(maxL);		
> > 		std::string data;
> > 		try { data=p->Read(ml); }
> > 		catch(CRbException & e) { 
> > 			ext_raise(self,e.str(),__FILE__,__LINE__); }
> > 		return rb_str_new(data.data(),data.size());
> > 	}
> 
> Perhaps, you should check if you mark ruby objects properly.

Just to be sure, for my problem:
class TC ; def initialize ; @s = '' ; end ; end
$tc = Testcase.new
# from here, tc.s should be valid (and never be terminated) until ...
$tc = nil
# ... here (after next GC)?

Is it possible $tc is not marked because it is assigned inside tc1.rb 
here:
	load 'tc1.rb' # tc1.rb: $tc = TC.new

If I do not hold any references to Ruby objects inside my C++-classes, 
do I still need to handle the mark-pass of GC?
(rather, C++ objects are held inside TC-member variables)

I will track down this bug tomorrow (it's now Sun 8:15pm here), any 
hints are appreciated.

Regards,
Michael B.