diff --git a/libcxx/include/__config b/libcxx/include/__config
--- a/libcxx/include/__config
+++ b/libcxx/include/__config
@@ -102,6 +102,10 @@
 #  define _LIBCPP_ABI_OPTIMIZED_FUNCTION
 // All the regex constants must be distinct and nonzero.
 #  define _LIBCPP_ABI_REGEX_CONSTANTS_NONZERO
+// Optimize the copy constructor to inline SSO copies.
+// This removes the copy constructor from the ABI, and depends on
+// the presence of __init_long() in the dylib / lib.
+#  define _LIBCPP_ABI_OPTIMIZE_COPY_CTOR
 #elif _LIBCPP_ABI_VERSION == 1
 #  if !defined(_LIBCPP_OBJECT_FORMAT_COFF)
 // Enable compiling copies of now inline methods into the dylib to support
diff --git a/libcxx/include/memory b/libcxx/include/memory
--- a/libcxx/include/memory
+++ b/libcxx/include/memory
@@ -2167,6 +2167,9 @@
 };
 #endif
 
+// Tag used to default initialize one or both of the pair's elements.
+struct __default_value_tag {};
+
 template <class _Tp, int _Idx,
           bool _CanBeEmptyBase =
               is_empty<_Tp>::value && !__libcpp_is_final<_Tp>::value>
@@ -2177,6 +2180,7 @@
 
 #ifndef _LIBCPP_CXX03_LANG
   _LIBCPP_INLINE_VISIBILITY constexpr __compressed_pair_elem() : __value_() {}
+  _LIBCPP_INLINE_VISIBILITY constexpr __compressed_pair_elem(__default_value_tag) {}
 
   template <class _Up, class = typename enable_if<
       !is_same<__compressed_pair_elem, typename decay<_Up>::type>::value
diff --git a/libcxx/include/string b/libcxx/include/string
--- a/libcxx/include/string
+++ b/libcxx/include/string
@@ -796,7 +796,13 @@
         _NOEXCEPT;
 #endif
 
+#ifdef _LIBCPP_ABI_OPTIMIZE_COPY_CTOR
+    _LIBCPP_HIDE_FROM_ABI
+#endif
     basic_string(const basic_string& __str);
+#ifdef _LIBCPP_ABI_OPTIMIZE_COPY_CTOR
+    _LIBCPP_HIDE_FROM_ABI
+#endif
     basic_string(const basic_string& __str, const allocator_type& __a);
 
 #ifndef _LIBCPP_CXX03_LANG
@@ -1543,6 +1549,11 @@
     inline
     void __init(size_type __n, value_type __c);
 
+#ifndef _LIBCPP_ABI_OPTIMIZE_COPY_CTOR
+    _LIBCPP_HIDE_FROM_ABI
+#endif
+    void __init_long(const basic_string& __str);
+
     template <class _InputIterator>
     inline
     typename enable_if
@@ -1793,6 +1804,23 @@
     traits_type::assign(__p[__sz], value_type());
 }
 
+template <class _CharT, class _Traits, class _Allocator>
+#ifdef _LIBCPP_ABI_OPTIMIZE_COPY_CTOR
+inline
+#endif
+void basic_string<_CharT, _Traits, _Allocator>::__init_long(
+    const basic_string& __str) {
+  const value_type* __s = _VSTD::__to_raw_pointer(__str.__get_long_pointer());
+  size_type __sz = __str.__get_long_size();
+  size_type __cap = __recommend(__sz);
+  pointer __p = __alloc_traits::allocate(__alloc(), __cap + 1);
+  __set_long_pointer(__p);
+  __set_long_cap(__cap + 1);
+  __set_long_size(__sz);
+  traits_type::copy(_VSTD::__to_raw_pointer(__p), __s, __sz);
+  traits_type::assign(__p[__sz], value_type());
+}
+
 template <class _CharT, class _Traits, class _Allocator>
 template <class>
 basic_string<_CharT, _Traits, _Allocator>::basic_string(const _CharT* __s, const _Allocator& __a)
@@ -1829,27 +1857,33 @@
 }
 
 template <class _CharT, class _Traits, class _Allocator>
+#ifdef _LIBCPP_ABI_OPTIMIZE_COPY_CTOR
+inline
+#endif
 basic_string<_CharT, _Traits, _Allocator>::basic_string(const basic_string& __str)
-    : __r_(__second_tag(), __alloc_traits::select_on_container_copy_construction(__str.__alloc()))
+    : __r_(__default_value_tag(), __alloc_traits::select_on_container_copy_construction(__str.__alloc()))
 {
     if (!__str.__is_long())
         __r_.first().__r = __str.__r_.first().__r;
     else
-        __init(_VSTD::__to_raw_pointer(__str.__get_long_pointer()), __str.__get_long_size());
+        __init_long(__str);
 #if _LIBCPP_DEBUG_LEVEL >= 2
     __get_db()->__insert_c(this);
 #endif
 }
 
 template <class _CharT, class _Traits, class _Allocator>
+#ifdef _LIBCPP_ABI_OPTIMIZE_COPY_CTOR
+inline
+#endif
 basic_string<_CharT, _Traits, _Allocator>::basic_string(
     const basic_string& __str, const allocator_type& __a)
-    : __r_(__second_tag(), __a)
+    : __r_(__default_value_tag(), __a)
 {
     if (!__str.__is_long())
         __r_.first().__r = __str.__r_.first().__r;
     else
-        __init(_VSTD::__to_raw_pointer(__str.__get_long_pointer()), __str.__get_long_size());
+        __init_long(__str);
 #if _LIBCPP_DEBUG_LEVEL >= 2
     __get_db()->__insert_c(this);
 #endif