[PATCH] Fix PR c++/65154 (ICE on braced initializer)

[PATCH] Fix PR c++/65154 (ICE on braced initializer)

[Mikhail Maltsev]

The following patch is supposed to fix the problem related to
initializing an array of aggregates with members, which have
user-defined default constructors.

Bootstrapped and regtested on x86_64-unknown-linux-gnu.

diff --git a/gcc/cp/init.c b/gcc/cp/init.c
index 0274663..0f53146 100644
--- a/gcc/cp/init.c
+++ b/gcc/cp/init.c
@@ -3716,11 +3716,9 @@ build_vec_init (tree base, tree maxindex, tree
init, {
 	  if (cxx_dialect >= cxx11 && AGGREGATE_TYPE_P (type))
 	    {
-	      if (BRACE_ENCLOSED_INITIALIZER_P (init)
-		  && CONSTRUCTOR_NELTS (init) == 0)
-		/* Reuse it.  */;
-	      else
-		init = build_constructor (init_list_type_node, NULL);
+	      /* We cannot reuse INIT here, because it must not be shared
+		 with member initializers.  */
+	      init = build_constructor (init_list_type_node, NULL);
 	      CONSTRUCTOR_IS_DIRECT_INIT (init) = true;
 	    }
 	  else
diff --git a/gcc/testsuite/g++.dg/init/array39.C
b/gcc/testsuite/g++.dg/init/array39.C new file mode 100644
index 0000000..2fd8937
--- /dev/null
+++ b/gcc/testsuite/g++.dg/init/array39.C
@@ -0,0 +1,46 @@
+// PR c++/65154
+// { dg-do run { target c++11 } }
+
+int cnt1 = 0,
+    cnt2 = 0;
+
+struct S_empty
+{
+    S_empty () {
+	cnt1++;
+    };
+};
+
+struct C1
+{
+  S_empty s;
+};
+
+struct S_init
+{
+  S_init () : i(42)
+  {
+    cnt2++;
+  };
+  int i;
+};
+
+struct C2
+{
+  S_init a, b;
+};
+
+int
+main ()
+{
+  C1 c1[5]{};
+  C2 c2[1]{};
+
+  if (c2[0].a.i != 42 || c2[0].b.i != 42)
+    return 1;
+
+  if (cnt1 != 5 || cnt2 != 2)
+    return 1;
+
+  return 0;
+}


Changelog record:

2015-03-02  Mikhail Maltsev  <maltsevm@gmail.com>

    	PR c++/65154
        * gcc/cp/init.c (build_vec_init): Fixed initializing aggregates
          with empty init list.
        * gcc/testsuite/g++.dg/init/array39.C: New test.


N.B.:
Consider the test case. The constructors of S_init and S_empty are
called from main (this can be seen in GIMPLE dump, right after
gimplification), while clang, for example, generates an explicit call to
constructors of C1 and C2. GCC does the same, if direct initializers
are removed. I don't know, if such behavior is OK or not.

-- 
Regards,
    Mikhail Maltsev

Re: [PATCH] Fix PR c++/65154 (ICE on braced initializer)

[Jason Merrill]

Applied, thanks.

In future please CC me on C++ patches so that I see them faster.  And 
feel free to ping as often as once a week.

Jason