C++ PATCH for c++/45315 (ICE on new T())

C++ PATCH for c++/45315 (ICE on new T())

[Jason Merrill]

[-- Attachment #1: Type: text/plain, Size: 255 bytes --]

The hoops that we jump through to implement value-initialization of 
aggregates don't work in templates, so we shouldn't try to do it at all; 
we're just going to throw away the result of build_new_1 anyway.

Tested x86_64-pc-linux-gnu, applied to trunk.

[-- Attachment #2: 45315.patch --]
[-- Type: text/x-patch, Size: 2291 bytes --]

commit 2bec61b98748c07fe7e985d787cf92b9a7faa4e5
Author: Jason Merrill <jason@redhat.com>
Date:   Wed Aug 18 16:00:41 2010 -0400

    	PR c++/45315
    	* init.c (build_new_1): Don't use build_value_init in a template.
    	(build_value_init): Make sure we don't.

diff --git a/gcc/cp/init.c b/gcc/cp/init.c
index 8555fad..189bcbe 100644
--- a/gcc/cp/init.c
+++ b/gcc/cp/init.c
@@ -295,6 +295,9 @@ build_value_init (tree type, tsubst_flags_t complain)
      zero-initializing the object and then calling the default
      constructor.  */
 
+  /* The AGGR_INIT_EXPR tweaking below breaks in templates.  */
+  gcc_assert (!processing_template_decl);
+
   if (CLASS_TYPE_P (type))
     {
       if (type_has_user_provided_constructor (type))
@@ -2310,7 +2313,8 @@ build_new_1 (VEC(tree,gc) **placement, tree type, tree nelts,
 	{
 	  init_expr = cp_build_indirect_ref (data_addr, RO_NULL, complain);
 
-	  if (TYPE_NEEDS_CONSTRUCTING (type) && !explicit_value_init_p)
+	  if (TYPE_NEEDS_CONSTRUCTING (type)
+	      && (!explicit_value_init_p || processing_template_decl))
 	    {
 	      init_expr = build_special_member_call (init_expr,
 						     complete_ctor_identifier,
@@ -2320,11 +2324,17 @@ build_new_1 (VEC(tree,gc) **placement, tree type, tree nelts,
 	    }
 	  else if (explicit_value_init_p)
 	    {
-	      /* Something like `new int()'.  */
-	      tree val = build_value_init (type, complain);
-	      if (val == error_mark_node)
-		return error_mark_node;
-	      init_expr = build2 (INIT_EXPR, type, init_expr, val);
+	      if (processing_template_decl)
+		/* Don't worry about it, we'll handle this properly at
+		   instantiation time.  */;
+	      else
+		{
+		  /* Something like `new int()'.  */
+		  tree val = build_value_init (type, complain);
+		  if (val == error_mark_node)
+		    return error_mark_node;
+		  init_expr = build2 (INIT_EXPR, type, init_expr, val);
+		}
 	    }
 	  else
 	    {
diff --git a/gcc/testsuite/g++.dg/init/value8.C b/gcc/testsuite/g++.dg/init/value8.C
new file mode 100644
index 0000000..0a9b90b
--- /dev/null
+++ b/gcc/testsuite/g++.dg/init/value8.C
@@ -0,0 +1,19 @@
+// PR c++/45315
+
+struct A
+{
+  A ();
+};
+
+template < int > struct B : A
+{
+  void foo ()
+  {
+    new B < 0 > ();
+  }
+};
+
+int main()
+{
+  B<1>().foo();
+}

Re: C++ PATCH for c++/45315 (ICE on new T())

[H.J. Lu]

On Thu, Aug 19, 2010 at 9:54 AM, Jason Merrill <jason@redhat.com> wrote:
> The hoops that we jump through to implement value-initialization of
> aggregates don't work in templates, so we shouldn't try to do it at all;
> we're just going to throw away the result of build_new_1 anyway.
>
> Tested x86_64-pc-linux-gnu, applied to trunk.
>

This caused:

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46658


-- 
H.J.