On Fri, Apr 08, 2011 at 01:58:14PM -0400, Jason Merrill wrote:
> On 04/05/2011 10:19 AM, Jakub Jelinek wrote:
> >i686-linux LTO bootstrap currently fails, because in one partition
> >we emit .Ldebug_info0 label twice. The problem is that
> >resolve_addr for call_site support attempts to force_decl_die external
> >function decls, and at least with LTO that in turn can attempt
> >to create new type DIEs, in this case an enumeration with context_die
> >being NULL. Unfortunately the code to add proper parents to limbo nodes
> >is done right before resolve_addr (and should be done there, so that
> >resolve_addr reaches all the needed DIEs).
> >
> >+/* Traverse the limbo die list, and add parent/child links. The only
> >+ dies without parents that should be here are concrete instances of
> >+ inline functions, and the comp_unit_die. We can ignore the
> >comp_unit_die.
> >+ For concrete instances, we can get the parent die from the abstract
> >+ instance. */
>
> Sounds like this comment needs to be updated if there can be types
> on the list as well.
On a closer look, this seems to be because LTO messes up types terribly,
struct cpp_options's lang field doesn't have enum c_lang type, but
enum prec whose TYPE_CONTEXT is c_parser_binary_expression
function from c_parser.c. So when trying to create DIE for cpp_options
and stuff in it we end up with the surprising limbo die.
Therefore, I'm withdrawing my patch and will look into this mess on Monday.
Jakub
