public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [C++ PATCH] fix canonical type ICE
@ 2015-05-22  2:56 Nathan Sidwell
  2015-05-22  6:34 ` Jason Merrill
  0 siblings, 1 reply; 4+ messages in thread
From: Nathan Sidwell @ 2015-05-22  2:56 UTC (permalink / raw)
  To: Jason Merrill; +Cc: GCC Patches

[-- Attachment #1: Type: text/plain, Size: 1084 bytes --]

Jason,
this fixes 65936, where we have two identical types with mismatching 
TYPE_CANONICAL.  The problem comes from template decl creation vs template 
instantiation.

At the point we create 'const C<int>' in the operator function we have not 
applied the 'may_alias' attribute to 'const C<int>'.  This means that the 
following snippet in tree.c:7729 build_pointer_type_for_mode

   if (lookup_attribute ("may_alias", TYPE_ATTRIBUTES (to_type)))
     can_alias_all = true;

doesn't set 'can_alias_all'.

Later we complete C<int> and parse the static_cast in the operator function. 
Completing C<int> will have applied the may_alias attribute.  So this time 
can_alias_all gets set on pointers to C<int> variants.  This leads to two 
different pointer types differing by TYPE_REF_CAN_ALIAS_ALL, each having 
different CANONICAL_TYPEs.  However as TYPE_STRUCTURAL_EQUALITY_P is false, we 
have a problem. We blow up in comptypes verifying the canonical type machinery.

This patch applies the may_alias attribute to the instantiated decl.

built & tested on x86_64-linux, ok?

nathan

[-- Attachment #2: 65936.patch --]
[-- Type: text/x-patch, Size: 1372 bytes --]

2015-05-21  Nathan Sidwell  <nathan@acm.org>

	cp/
	PR c++/65936
	pt.c (lookup_template_class_1): Apply may_alias attribute here.

	testsuite/
	* g++.dg/template/pr65936.C: New.

Index: cp/pt.c
===================================================================
--- cp/pt.c	(revision 223503)
+++ cp/pt.c	(working copy)
@@ -7856,6 +7856,10 @@ lookup_template_class_1 (tree d1, tree a
 	    = CLASSTYPE_DECLARED_CLASS (template_type);
 	  SET_CLASSTYPE_IMPLICIT_INSTANTIATION (t);
 	  TYPE_FOR_JAVA (t) = TYPE_FOR_JAVA (template_type);
+	  if (lookup_attribute ("may_alias",
+				TYPE_ATTRIBUTES (template_type)))
+	    TYPE_ATTRIBUTES (t) = tree_cons (get_identifier ("may_alias"),
+					     NULL_TREE, NULL_TREE);
 
 	  /* A local class.  Make sure the decl gets registered properly.  */
 	  if (context == current_function_decl)
Index: testsuite/g++.dg/template/pr65936.C
===================================================================
--- testsuite/g++.dg/template/pr65936.C	(revision 0)
+++ testsuite/g++.dg/template/pr65936.C	(working copy)
@@ -0,0 +1,21 @@
+// checking ICE in canonical typing
+
+class A;
+
+template <typename> struct B
+{
+  typedef A type;
+};
+
+template <class T> class C
+  : public B<T>::type
+{
+} __attribute__ ((__may_alias__));
+
+class A
+{
+  operator const C<int> &()
+  {
+    return *static_cast<const C<int> *> (this);
+  }
+};

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [C++ PATCH] fix canonical type ICE
  2015-05-22  2:56 [C++ PATCH] fix canonical type ICE Nathan Sidwell
@ 2015-05-22  6:34 ` Jason Merrill
  2015-05-22 18:46   ` Nathan Sidwell
  0 siblings, 1 reply; 4+ messages in thread
From: Jason Merrill @ 2015-05-22  6:34 UTC (permalink / raw)
  To: Nathan Sidwell; +Cc: GCC Patches

How about adding may_alias support to the code a bit lower down that 
copies the abi_tag attribute?

Jason

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [C++ PATCH] fix canonical type ICE
  2015-05-22  6:34 ` Jason Merrill
@ 2015-05-22 18:46   ` Nathan Sidwell
  2015-05-23 22:45     ` Jason Merrill
  0 siblings, 1 reply; 4+ messages in thread
From: Nathan Sidwell @ 2015-05-22 18:46 UTC (permalink / raw)
  To: Jason Merrill; +Cc: GCC Patches

[-- Attachment #1: Type: text/plain, Size: 370 bytes --]

On 05/21/15 21:47, Jason Merrill wrote:
> How about adding may_alias support to the code a bit lower down that copies the
> abi_tag attribute?

Good idea.  This keeps the non-copy behavior when the attribute is the last on 
the source attribute list, and fixes up the case for when there are both 
may_alias and abi_tag.

booted and tested on x86_64-linux

ok?


nathan

[-- Attachment #2: 65936.patch --]
[-- Type: text/x-patch, Size: 1721 bytes --]

2015-05-22  Nathan Sidwell  <nathan@acm.org>

	PR c++/65936
	* pt.c (lookup_template_class_1): Copy may_alias attribute too.

	PR c++/65936
	* g++.dg/template/pr65936.C: New.

Index: cp/pt.c
===================================================================
--- cp/pt.c	(revision 223503)
+++ cp/pt.c	(working copy)
@@ -7905,15 +7905,22 @@ lookup_template_class_1 (tree d1, tree a
       if (OVERLOAD_TYPE_P (t)
 	  && !DECL_ALIAS_TEMPLATE_P (gen_tmpl))
 	{
-	  if (tree attributes
-	      = lookup_attribute ("abi_tag", TYPE_ATTRIBUTES (template_type)))
+	  static const char *tags[] = {"abi_tag", "may_alias"};
+
+	  for (unsigned ix = 0; ix != 2; ix++)
 	    {
-	      if (!TREE_CHAIN (attributes))
+	      tree attributes
+		= lookup_attribute (tags[ix], TYPE_ATTRIBUTES (template_type));
+
+	      if (!attributes)
+		;
+	      else if (!TREE_CHAIN (attributes) && !TYPE_ATTRIBUTES (t))
 		TYPE_ATTRIBUTES (t) = attributes;
 	      else
 		TYPE_ATTRIBUTES (t)
-		  = build_tree_list (TREE_PURPOSE (attributes),
-				     TREE_VALUE (attributes));
+		  = tree_cons (TREE_PURPOSE (attributes),
+			       TREE_VALUE (attributes),
+			       TYPE_ATTRIBUTES (t));
 	    }
 	}
 
Index: testsuite/g++.dg/template/pr65936.C
===================================================================
--- testsuite/g++.dg/template/pr65936.C	(revision 0)
+++ testsuite/g++.dg/template/pr65936.C	(working copy)
@@ -0,0 +1,21 @@
+// checking ICE in canonical typing
+
+class A;
+
+template <typename> struct B
+{
+  typedef A type;
+};
+
+template <class T> class C
+  : public B<T>::type
+{
+} __attribute__ ((__may_alias__));
+
+class A
+{
+  operator const C<int> &()
+  {
+    return *static_cast<const C<int> *> (this);
+  }
+};

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [C++ PATCH] fix canonical type ICE
  2015-05-22 18:46   ` Nathan Sidwell
@ 2015-05-23 22:45     ` Jason Merrill
  0 siblings, 0 replies; 4+ messages in thread
From: Jason Merrill @ 2015-05-23 22:45 UTC (permalink / raw)
  To: Nathan Sidwell; +Cc: GCC Patches

OK, thanks.

Jason

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2015-05-23 20:25 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-05-22  2:56 [C++ PATCH] fix canonical type ICE Nathan Sidwell
2015-05-22  6:34 ` Jason Merrill
2015-05-22 18:46   ` Nathan Sidwell
2015-05-23 22:45     ` Jason Merrill

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).