Re: [cross-binutils-2.15.91.0.2] assertion fail linker.c:619

[Alan Modra <amodra at bigpond dot net dot au>]

On Tue, Aug 17, 2004 at 03:16:10PM -0700, H. J. Lu wrote:
> The problem is bfd_link_add_undef assumes h->und_next == NULL. But it
> may not be true in all cases, depending on how lucky/unlucky you are.
> This patch seems to fix the testcase. But I am not sure if it is 100%
> correct.

I don't think it's right.  und_next should be NULL at the tail of the
list.  You need to fix some place where an undefined symbol finds
a definition.

Hmm, it so happens that the ELF linker doesn't traverse the list (except
for ppc64 with a recent patch of mine), so und_next is effectively just
a flag.  Neither are entries removed from the list (except again for
ppc64).  So I'm not certain what we should do here.  Obviously, I need
to ensure that the ppc64 list traversal doesn't fall off the end of the
list, or loop..

> --- linker.c.undef	2004-08-13 08:00:10.000000000 -0700
> +++ linker.c	2004-08-17 15:14:00.595475841 -0700
> @@ -616,7 +616,7 @@ void
>  bfd_link_add_undef (struct bfd_link_hash_table *table,
>  		    struct bfd_link_hash_entry *h)
>  {
> -  BFD_ASSERT (h->und_next == NULL);
> +  BFD_ASSERT (h->und_next == NULL || h->und_next == table->undefs_tail);
>    if (table->undefs_tail != NULL)
>      table->undefs_tail->und_next = h;
>    if (table->undefs == NULL)

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre