[Bug ipa/68220] [5/6 Regression] Devirtualization ICE in record_target_from_binfo, at ipa-devirt.c:2389

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68220

--- Comment #5 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
I am testing the following. This bug crept in when adding the dynamic types:
Index: ipa-polymorphic-call.c
===================================================================
--- ipa-polymorphic-call.c      (revision 229820)
+++ ipa-polymorphic-call.c      (working copy)
@@ -154,6 +154,8 @@ ipa_polymorphic_call_context::restrict_t
           && tree_to_shwi (TYPE_SIZE (outer_type)) >= 0
           && tree_to_shwi (TYPE_SIZE (outer_type)) <= offset)
    {
+     bool der = maybe_derived_type; /* clear_outer_type will reset it.  */
+     bool dyn = dynamic;
      clear_outer_type (otr_type);
      type = otr_type;
      cur_offset = 0;
@@ -162,7 +164,7 @@ ipa_polymorphic_call_context::restrict_t
        For dynamic types, we really do not have information about
        size of the memory location.  It is possible that completely
        different type is stored after outer_type.  */
-     if (!maybe_derived_type && !dynamic)
+     if (!der && !dyn)
        {
         clear_speculation ();
         invalid = true;
@@ -467,6 +469,7 @@ contains_type_p (tree outer_type, HOST_W
   context.offset = offset;
   context.outer_type = TYPE_MAIN_VARIANT (outer_type);
   context.maybe_derived_type = false;
+  context.dynamic = false;
   return context.restrict_to_inner_class (otr_type, consider_placement_new,
consider_bases);
 }