From: Andrew Walrond <andrew@...>
Date: 2004-09-17T20:34:28+09:00
Subject: Re: Valgrind analysis of [BUG] unknown node type 0

On Friday 17 Sep 2004 12:01, ts wrote:
>
> '.local i' : the address will be allocated in bss section, and it will
> zeroed at run-time.
>

Ok; I concur. Infact running the 'fixed' code through Valgrind does not remove 
the errors.

The function causing valgrind to shout


==28001== Conditional jump or move depends on uninitialised value(s)
==28001==    at 0x806FF18: is_pointer_to_heap (gc.c:591)
==28001==    by 0x806FEE1: mark_locations_array (gc.c:609)
==28001==    by 0x80710F8: rb_gc (gc.c:1328)
==28001==    by 0x806FBAC: rb_newobj (gc.c:376)

is

static inline int
is_pointer_to_heap(ptr)
    void *ptr;
{
    register RVALUE *p = RANY(ptr);
    register RVALUE *heap_org;
    register long i;

    if (p < lomem || p > himem) return Qfalse;

    /* check if p looks like a pointer */
    for (i=0; i < heaps_used; i++) {
  heap_org = heaps[i].slot;
  if (heap_org <= p && p < heap_org + heaps[i].limit &&
      ((((char*)p)-((char*)heap_org))%sizeof(RVALUE)) == 0)
      return Qtrue;
    }
    return Qfalse;
}

So if lomem and himem are not the culprits, then it must be 'p'

I'll try following this back up the stack with gdb, to see where it leads...

Andrew