[patch] Fix PR bootstrap/69513

[Eric Botcazou]

Hi,

this is the failure of the LTO bootstrap with profile-directed optimization 
when compiling gnat1 with debug info.  The issue is that flush_limbo_die_list 
doesn't actually fully flush the limbo list because it calls get_context_die, 
which calls force_decl_die, which can put new DIEs on the limbo list in LTO
mode when nested functions are put in a different partition than that of their 
parent function.

Richard, this is the version of the patch that you said you would approve.
Tested on x86_64-suse-linux, is it OK for GCC 6 or must it wait until GCC 7?


2016-03-15  Eric Botcazou  <ebotca...@adacore.com>

        PR bootstrap/69513
        * dwarf2out.c (flush_limbo_die_list): Really flush the limbo list.

-- 
Eric Botcazou
Index: dwarf2out.c
===================================================================
--- dwarf2out.c	(revision 234171)
+++ dwarf2out.c	(working copy)
@@ -27281,12 +27281,15 @@ optimize_location_lists (dw_die_ref die)
 static void
 flush_limbo_die_list (void)
 {
-  limbo_die_node *node, *next_node;
+  limbo_die_node *node;
 
-  for (node = limbo_die_list; node; node = next_node)
+  /* get_context_die calls force_decl_die, which can put new DIEs on the
+     limbo list in LTO mode when nested functions are put in a different
+     partition than that of their parent function.  */
+  while ((node = limbo_die_list))
     {
       dw_die_ref die = node->die;
-      next_node = node->next;
+      limbo_die_list = node->next;
 
       if (die->die_parent == NULL)
 	{
@@ -27324,8 +27327,6 @@ flush_limbo_die_list (void)
 	    }
 	}
     }
-
-  limbo_die_list = NULL;
 }
 
 /* Output stuff that dwarf requires at the end of every file,