The analysis of 55419 pointed out the conflict in a TARGET_EXPR having
both TREE_SIDE_EFFECTS and TREE_CONSTANT set. After reading that I
decided to stop setting TREE_CONSTANT on the TARGET_EXPR and do the
necessary work to adjust for that. And then was surprised to find that
no additional work was necessary.
Tested x86_64-pc-linux-gnu, applying to trunk and 4.7.
commit b0f1533a6e7981e5ae12809a60f7b4a96221ff13
Author: Jason Merrill <ja...@redhat.com>
Date: Fri Dec 7 10:42:01 2012 -0500
PR c++/55419
* tree.c (build_target_expr): Don't set TREE_CONSTANT.
diff --git a/gcc/cp/tree.c b/gcc/cp/tree.c
index ca82f75..963bb1b 100644
--- a/gcc/cp/tree.c
+++ b/gcc/cp/tree.c
@@ -329,8 +329,6 @@ build_target_expr (tree decl, tree value, tsubst_flags_t complain)
side-effects, then the optimizer should be able to get rid of
whatever code is generated anyhow. */
TREE_SIDE_EFFECTS (t) = 1;
- if (literal_type_p (type))
- TREE_CONSTANT (t) = TREE_CONSTANT (value);
return t;
}
diff --git a/gcc/testsuite/g++.dg/cpp0x/constexpr-initlist6.C b/gcc/testsuite/g++.dg/cpp0x/constexpr-initlist6.C
new file mode 100644
index 0000000..6b822a1
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp0x/constexpr-initlist6.C
@@ -0,0 +1,27 @@
+// PR c++/55419
+// { dg-options -std=c++11 }
+
+struct P
+{
+ P () = default;
+ explicit constexpr P (int x) : p (x) {}
+ int p;
+};
+
+struct Q
+{
+ constexpr Q () : q (0x7f) {}
+ int q;
+};
+
+struct R
+{
+ Q q;
+ P p;
+};
+
+void
+foo (R *x)
+{
+ *x = {};
+}
