From patchwork Wed Jul  7 15:10:34 2010
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Subject: Fix pr middle-end/41355
X-Patchwork-Submitter: Duncan Sands <baldrick@free.fr>
X-Patchwork-Id: 58123
Message-Id: <4C3498EA.1090803@free.fr>
To: gcc-patches <gcc-patches@gcc.gnu.org>
Date: Wed, 07 Jul 2010 17:10:34 +0200
From: Duncan Sands <baldrick@free.fr>
List-Id: <gcc-patches.gcc.gnu.org>

When materializing a cloned function with arguments removed,
to TYPE_POINTER_TO field of the new type (which has fewer
arguments) was left pointing to the original type.  Bootstraps
and passes regression testing (tested on the 4.5 branch since
4.6 seems ill right now).  Jan Hubicka says "it seems fine to me,
but I can not approve it".  OK to apply to mainline and 4.5?

Changelog entry:

  2010-07-07  Duncan Sands  <baldrick@free.fr>

  	PR middle-end/41355
  	* tree.c (build_function_type_skip_args): Copy the original type using
          build_distinct_type_copy rather than copy_node.

Index: tree.c
===================================================================
--- tree.c	(revision 161914)
+++ tree.c	(working copy)
@@ -7350,7 +7350,7 @@
    if (TREE_CODE (orig_type) != METHOD_TYPE
        || !bitmap_bit_p (args_to_skip, 0))
      {
-      new_type = copy_node (orig_type);
+      new_type = build_distinct_type_copy (orig_type);
        TYPE_ARG_TYPES (new_type) = new_reversed;
      }
    else