In this testcase we find ourselves in split_nonconstant_init:

       init = cp_fully_fold (init);
       code = push_stmt_list ();
       if (split_nonconstant_init_1 (dest, init))

where initially INIT was a CONSTRUCTOR, but cp_fully_fold returned
a TARGET_EXPR.  This confuses split_nonconstant_init_1 which only expects
a CONSTRUCTOR or a VECTOR_CST.  Jason suggested stripping TARGET_EXPRs /
VIEW_CONVERT_EXPRs, so here it is.

Bootstrapped/regtested on x86_64-linux, ok for trunk?

2018-03-01  Marek Polacek  <pola...@redhat.com>

        PR c++/84590
        * cp-gimplify.c (cp_fully_fold): Unwrap TARGET_EXPR or a CONSTRUCTOR
        wrapped in VIEW_CONVERT_EXPR.

        * c-c++-common/ubsan/shift-11.c: New test.

diff --git gcc/cp/cp-gimplify.c gcc/cp/cp-gimplify.c
index 55a9d278dbe..1fa9c466e6b 100644
--- gcc/cp/cp-gimplify.c
+++ gcc/cp/cp-gimplify.c
@@ -2038,7 +2038,18 @@ cp_fully_fold (tree x)
      have to call both.  */
   if (cxx_dialect >= cxx11)
     x = maybe_constant_value (x);
-  return cp_fold_rvalue (x);
+  x = cp_fold_rvalue (x);
+
+  /* Sometimes we are given a CONSTRUCTOR but cxx_eval_outermost_constant_expr
+     wraps it into a TARGET_EXPR; undo that here.  */
+  if (TREE_CODE (x) == TARGET_EXPR)
+    x = TARGET_EXPR_INITIAL (x);
+  else if (TREE_CODE (x) == VIEW_CONVERT_EXPR
+          && TREE_CODE (TREE_OPERAND (x, 0)) == CONSTRUCTOR
+          && TREE_TYPE (TREE_OPERAND (x, 0)) == TREE_TYPE (x))
+    x = TREE_OPERAND (x, 0);
+
+  return x;
 }
 
 /* c-common interface to cp_fold.  If IN_INIT, this is in a static initializer
diff --git gcc/testsuite/c-c++-common/ubsan/shift-11.c 
gcc/testsuite/c-c++-common/ubsan/shift-11.c
index e69de29bb2d..03a72e217a2 100644
--- gcc/testsuite/c-c++-common/ubsan/shift-11.c
+++ gcc/testsuite/c-c++-common/ubsan/shift-11.c
@@ -0,0 +1,13 @@
+/* PR c++/84590 */
+/* { dg-do compile } */
+/* { dg-options "-fsanitize=shift" } */
+
+struct S {
+  int b;
+};
+
+void
+fn (void)
+{
+  struct S c1 = { 1 << -1 }; /* { dg-warning "left shift" } */
+}

        Marek

Reply via email to