This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH][C++] Fix PR67333
- From: Mikhail Maltsev <maltsevm at gmail dot com>
- To: Jason Merrill <jason at redhat dot com>, gcc-patches <gcc-patches at gnu dot org>
- Date: Sat, 17 Oct 2015 21:57:33 +0300
- Subject: Re: [PATCH][C++] Fix PR67333
- Authentication-results: sourceware.org; auth=none
- References: <55DF5684 dot 8050607 at gmail dot com> <5613D0CA dot 1070706 at redhat dot com>
On 10/06/2015 04:46 PM, Jason Merrill wrote:
> Hi, sorry for the slow response. Please feel free to ping once a week.
>
> On 08/27/2015 02:27 PM, Mikhail Maltsev wrote:
>> + if (TREE_THIS_VOLATILE (t) && (!DECL_P (t) || want_rval))
>
> Why the !DECL_P check? Pulling the value out of a volatile INDIRECT_REF is just
> as problematic as from a variable.
Hmm... my intent was to check that we are not dealing with an expression, i.e.
that should have been 'EXPR_P' rather than '!DECL_P'.
I changed the condition to 'TREE_THIS_VOLATILE (t) && (EXPR_P (t) || want_rval)'
and also added one more test (the one with 'test_ref' function). The updated
patch passes bootstrap & regtest on x86_64-pc-linux-gnu.
But now I wonder, if the combination 'TREE_THIS_VOLATILE (t) && EXPR_P (t) &&
!want_rval' is possible at all and should it be rejected?
--
Regards,
Mikhail Maltsev
diff --git a/gcc/cp/constexpr.c b/gcc/cp/constexpr.c
index ec9710c..b23d52a 100644
--- a/gcc/cp/constexpr.c
+++ b/gcc/cp/constexpr.c
@@ -4006,7 +4006,7 @@ potential_constant_expression_1 (tree t, bool want_rval, bool strict,
return false;
if (t == NULL_TREE)
return true;
- if (TREE_THIS_VOLATILE (t))
+ if (TREE_THIS_VOLATILE (t) && (EXPR_P (t) || want_rval))
{
if (flags & tf_error)
error ("expression %qE has side-effects", t);
diff --git a/gcc/testsuite/g++.dg/cpp0x/constexpr-67333.C b/gcc/testsuite/g++.dg/cpp0x/constexpr-67333.C
new file mode 100644
index 0000000..885ece6
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp0x/constexpr-67333.C
@@ -0,0 +1,29 @@
+// PR c++/67333
+// { dg-do compile { target c++11 } }
+
+template <int N>
+struct integral_constant
+{
+ static constexpr int value = N;
+};
+
+template <typename T, int S>
+constexpr int lengthof (const volatile T (&)[S])
+{
+ return S;
+}
+
+template <typename T, int S>
+constexpr int valueof (const volatile T (&s)[S]) // { dg-error "has side-effects" }
+{
+ return s[0];
+}
+
+int main ()
+{
+ volatile int meow[4] {};
+ integral_constant<lengthof (meow)>::value; // OK
+ integral_constant<valueof (meow)>::value; // { dg-error "in a constant expression" }
+ return 0;
+}
+
diff --git a/gcc/testsuite/g++.dg/cpp1y/constexpr-67333.C b/gcc/testsuite/g++.dg/cpp1y/constexpr-67333.C
new file mode 100644
index 0000000..333047a
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp1y/constexpr-67333.C
@@ -0,0 +1,36 @@
+// PR c++/67333
+// { dg-do compile { target c++14 } }
+
+template <int N>
+struct integral_constant
+{
+ static constexpr int value = N;
+};
+
+constexpr int decl (int x)
+{
+ volatile int v = x;
+ return x;
+}
+
+constexpr int use (int x)
+{
+ volatile int v = x;
+ return v;
+} // { dg-error "has side-effects" }
+
+constexpr int test_ref (volatile int &x)
+{
+ volatile int &vol_ref = x;
+ volatile int *vol_ptr = &x;
+ volatile int &vol_deref = *vol_ptr;
+ return 0;
+}
+
+int main()
+{
+ volatile int x = 0;
+ constexpr int t = test_ref (x);
+ integral_constant<decl (2)>::value; // OK
+ integral_constant<use (2)>::value; // { dg-error "in a constant expression" }
+}