[prev in list] [next in list] [prev in thread] [next in thread] 

List:       gcc-patches
Subject:    Re: [PATCH c/4.2] Fix PR c/27273
From:       Dirk Mueller <dmuell () gmx ! net>
Date:       2006-05-19 7:41:22
Message-ID: 200605190941.23059.dmuell () gmx ! net
[Download RAW message or body]

On Thursday, 18. May 2006 02:19, Roger Sayle wrote:

> However, in the meantime, I believe the less intrusive patch is
> to check for CONSTANT_CLASS_P(expr) immediately before the test
> of TREE_CONSTANT_OVERFLOW.

Hmm. I didn't do that because it introduces a warning in the testcase that 
hasn't been there before (with older gcc's), and I can't decide if that 
should be a bug or a feature. Having thought more about it, I tend to believe 
in a feature. 

New patch, bootstrapped, regtested on i686-suse-linux. Ok?

:ADDPATCH c:

2006-05-19  Dirk Mueller  <dmueller@suse.de>

	PR c/27273
	* c-common.c (convert_and_check): only track TREE_CONSTANT_OVERFLOW
	for CONSTANT_CLASS_P expressions.

	* gcc.dg/overflow-warn-5.c: New testcase.

--- gcc/c-common.c	(revision 113851)
+++ gcc/c-common.c	(working copy)
@@ -1083,7 +1083,8 @@ convert_and_check (tree type, tree expr)
 
 	  /* Do not diagnose overflow in a constant expression merely
 	     because a conversion overflowed.  */
-	  TREE_CONSTANT_OVERFLOW (t) = TREE_CONSTANT_OVERFLOW (expr);
+	  TREE_CONSTANT_OVERFLOW (t) = CONSTANT_CLASS_P (expr) 
+	                               && TREE_CONSTANT_OVERFLOW (expr);
 
 	  /* No warning for converting 0x80000000 to int.  */
 	  if (!(TYPE_UNSIGNED (type) < TYPE_UNSIGNED (TREE_TYPE (expr))
--- gcc/testsuite/gcc.dg/overflow-warn-5.c	(revision 0)
+++ gcc/testsuite/gcc.dg/overflow-warn-5.c	(revision 0)
@@ -0,0 +1,7 @@
+/* PR c/27273 */
+/* { dg-do compile } */
+/* { dg-options "-Woverflow" } */
+
+unsigned char rx_async(unsigned char p) {
+    return p & 512; /* { dg-warning "overflow in implicit constant 
conversion" } */
+}


[prev in list] [next in list] [prev in thread] [next in thread]