diff mbox

[C++,Patch/RFC] PR 64877

Message ID 54D0F3DE.7070909@oracle.com
State New
Headers show

Commit Message

Paolo Carlini Feb. 3, 2015, 4:14 p.m. UTC 
Hi,

On 02/03/2015 03:19 PM, Jason Merrill wrote:
> On 02/03/2015 05:45 AM, Paolo Carlini wrote:
>> +          if (TREE_CODE (pfn0) != ADDR_EXPR
>> +          || !decl_with_nonnull_addr_p (TREE_OPERAND (pfn0, 0)))
>
> I don't like duplicating the logic for when we might know the pfn is 
> non-null; that seems fragile.
I agree, seems fragile, on the other hand would be rather precise, IMHO. 
We could at least have a function wrapping the above and a comment in 
both places. Anyway...
>   I'd rather go with the tf_none approach, or otherwise suppress the 
> warning.
There are so many ways to fix this, and I'm a bit confused by the range 
of warning suppression mechanisms we have got at this point (eg, 
TREE_NO_WARNING, c_inhibit_evaluation_warnings, the new warning_sentinel 
in pt.c, tf_none, etc.. ;) But this case should be at least relatively 
safe because the expression is internally generated, right? Fiddling 
with tf_none, as mentioned by Manuel, should indeed work, but then, how 
far we would go with that? Only the 'e2 = cp_build_binary_op (location, 
EQ_EXPR, pfn0, ...)' lines or more of the calls for synthesized 
expressions? Would it be Ok for 5.0 to just use TREE_NO_WARNING like in 
the below?

Thanks,
Paolo.

/////////////////////

Comments

Jason Merrill Feb. 3, 2015, 4:26 p.m. UTC | #1 
On 02/03/2015 11:14 AM, Paolo Carlini wrote:
> +	  /* Avoid -Waddress warnings (c++/64877).  */
> +	  TREE_NO_WARNING (pfn0) = 1;

I'd check for ADDR_EXPR before doing this; OK with that change.

Jason
diff mbox

Patch

Index: cp/typeck.c
===================================================================
--- cp/typeck.c	(revision 220371)
+++ cp/typeck.c	(working copy)
@@ -4415,7 +4415,8 @@  cp_build_binary_op (location_t location,
 	      && decl_with_nonnull_addr_p (TREE_OPERAND (op0, 0)))
 	    {
 	      if ((complain & tf_warning)
-		  && c_inhibit_evaluation_warnings == 0)
+		  && c_inhibit_evaluation_warnings == 0
+		  && !TREE_NO_WARNING (op0))
 		warning (OPT_Waddress, "the address of %qD will never be NULL",
 			 TREE_OPERAND (op0, 0));
 	    }
@@ -4436,7 +4437,8 @@  cp_build_binary_op (location_t location,
 	      && decl_with_nonnull_addr_p (TREE_OPERAND (op1, 0)))
 	    {
 	      if ((complain & tf_warning)
-		  && c_inhibit_evaluation_warnings == 0)
+		  && c_inhibit_evaluation_warnings == 0
+		  && !TREE_NO_WARNING (op1))
 		warning (OPT_Waddress, "the address of %qD will never be NULL",
 			 TREE_OPERAND (op1, 0));
 	    }
@@ -4537,6 +4539,8 @@  cp_build_binary_op (location_t location,
 	    op1 = save_expr (op1);
 
 	  pfn0 = pfn_from_ptrmemfunc (op0);
+	  /* Avoid -Waddress warnings (c++/64877).  */
+	  TREE_NO_WARNING (pfn0) = 1;
 	  pfn1 = pfn_from_ptrmemfunc (op1);
 	  delta0 = delta_from_ptrmemfunc (op0);
 	  delta1 = delta_from_ptrmemfunc (op1);
Index: testsuite/g++.dg/warn/Waddress-2.C
===================================================================
--- testsuite/g++.dg/warn/Waddress-2.C	(revision 0)
+++ testsuite/g++.dg/warn/Waddress-2.C	(working copy)
@@ -0,0 +1,24 @@ 
+// PR c++/64877
+// { dg-options "-Waddress" }
+
+template<class Derived>
+struct S
+{
+  void m() {
+  }
+
+  S()
+  {
+    if (&S<Derived>::Unwrap != &Derived::Unwrap)
+      m();
+  }
+
+  void Unwrap() {
+  }
+};
+
+struct T : public S<T>
+{
+};
+
+T t;