From: ts <decoux@...>
Date: 2004-10-19T23:14:12+09:00
Subject: Re: [BUG] segfault in ruby-1.8.2p2

>>>>> "t" == ts  <decoux@moulon.inra.fr> writes:

t> b.rb:3: [BUG] Segmentation fault
t> ruby 1.8.2 (2004-10-18) [i686-linux]

 Do someone has a C compiler which don't define __libc_stack_end ?

 This is the difference between 1.8 and 1.9

 Actually my conclusion is that you can't predict the order of the
 arguments on the stack. For example with 1.8.2-preview2

Breakpoint 2, ruby_exec () at eval.c:1456
1456            eval_node(ruby_top_self, ruby_eval_tree);
(gdb) p &tmp
$1 = (volatile struct RNode **) 0xbffff8ec
(gdb) p prot_tag
$2 = (struct tag *) 0xbffff8f0
(gdb) p prot_tag+1
$3 = (struct tag *) 0xbffff9ac
(gdb) p rb_gc_stack_start
$4 = (VALUE *) 0xbffff900
(gdb) 

 and ruby make the assumption that tmp is in the first position

  Init_stack((void*)&tmp);

 a simple fix like

int
ruby_exec()
{
    volatile NODE *tmp;

    Init_stack((void*)&tmp);
    return ruby_exec0();
}

 seems to correct the problem


Guy Decoux