Make basic_string::operator=() tail call properly

[mvels]

Summary

We discovered that the compiler may chose not to inline the operator=, which leads to an expensive extra stack frame. This change makes __assign_no_alias always tail called.

libcxx/include/__string

#define _LIBCPP_STRING_UNSTABLE_EXTERN_TEMPLATE_LIST(_Func, _CharType) \
   _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>::size_type basic_string<_CharType>::copy(value_type*, size_type, size_type) const) \
   _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>::basic_string(basic_string const&, size_type, size_type, std::allocator<_CharType> const&)) \
   _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>::size_type basic_string<_CharType>::find(value_type, size_type) const) \
   _Func(_LIBCPP_FUNC_VIS void basic_string<_CharType>::__init(size_type, value_type)) \
   _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>& basic_string<_CharType>::insert(size_type, value_type const*)) \
   _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>::size_type basic_string<_CharType>::find_last_of(value_type const*, size_type, size_type) const) \
   _Func(_LIBCPP_FUNC_VIS void basic_string<_CharType>::__grow_by(size_type, size_type, size_type, size_type, size_type, size_type)) \
   _Func(_LIBCPP_FUNC_VIS void basic_string<_CharType>::__grow_by_and_replace(size_type, size_type, size_type, size_type, size_type, size_type, value_type const*)) \
-  _Func(_LIBCPP_FUNC_VIS void basic_string<_CharType>::__assign_no_alias<false>(value_type const*, size_type)) \
-  _Func(_LIBCPP_FUNC_VIS void basic_string<_CharType>::__assign_no_alias<true>(value_type const*, size_type)) \
+  _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>& basic_string<_CharType>::__assign_no_alias<false>(value_type const*, size_type)) \
+  _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>& basic_string<_CharType>::__assign_no_alias<true>(value_type const*, size_type)) \
   _Func(_LIBCPP_FUNC_VIS void basic_string<_CharType>::push_back(value_type)) \
   _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>& basic_string<_CharType>::append(size_type, value_type)) \
   _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>::size_type basic_string<_CharType>::rfind(value_type, size_type) const) \
   _Func(_LIBCPP_FUNC_VIS const basic_string<_CharType>::size_type basic_string<_CharType>::npos) \
   _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>& basic_string<_CharType>::assign(size_type, value_type)) \
   _Func(_LIBCPP_FUNC_VIS void basic_string<_CharType>::__erase_external_with_move(size_type, size_type)) \
   _Func(_LIBCPP_FUNC_VIS basic_string<_CharType>& basic_string<_CharType>::append(basic_string const&, size_type, size_type)) \
   _Func(_LIBCPP_FUNC_VIS int basic_string<_CharType>::compare(value_type const*) const) \

libcxx/include/string

#endif  // _LIBCPP_ABI_ALTERNATE_STRING_LAYOUT
     void __grow_by_and_replace(size_type __old_cap, size_type __delta_cap, size_type __old_sz,
                                size_type __n_copy,  size_type __n_del,
                                size_type __n_add, const value_type* __p_new_stuff);
 
     // __assign_no_alias is invoked for assignment operations where we
     // have proof that the input does not alias the current instance.
     // For example, operator=(basic_string) performs a 'self' check.
     template <bool __is_short>
-    void __assign_no_alias(const value_type* __s, size_type __n);
+    basic_string& __assign_no_alias(const value_type* __s, size_type __n);
 
     _LIBCPP_INLINE_VISIBILITY
     void __erase_to_end(size_type __pos);
 
     // __erase_external_with_move is invoked for erase() invocations where
     // `n ~= npos`, likely requiring memory moves on the string data.
     void __erase_external_with_move(size_type __pos, size_type __n);
 
basic_string<_CharT, _Traits, _Allocator>::__grow_by(size_type __old_cap, size_type __delta_cap, size_type __old_sz,
     __set_long_pointer(__p);
     __set_long_cap(__cap+1);
 }
 
 // assign
 
 template <class _CharT, class _Traits, class _Allocator>
 template <bool __is_short>
-void basic_string<_CharT, _Traits, _Allocator>::__assign_no_alias(
+basic_string<_CharT, _Traits, _Allocator>&
+basic_string<_CharT, _Traits, _Allocator>::__assign_no_alias(
     const value_type* __s, size_type __n) {
   size_type __cap = __is_short ? __min_cap : __get_long_cap();
   if (__n < __cap) {
     pointer __p = __is_short ? __get_short_pointer() : __get_long_pointer();
     __is_short ? __set_short_size(__n) : __set_long_size(__n);
     traits_type::copy(_VSTD::__to_address(__p), __s, __n);
     traits_type::assign(__p[__n], value_type());
     __invalidate_iterators_past(__n);
   } else {
     size_type __sz = __is_short ? __get_short_size() : __get_long_size();
     __grow_by_and_replace(__cap - 1, __n - __cap + 1, __sz, 0, __sz, __n, __s);
   }
+  return *this;
 }
 
 template <class _CharT, class _Traits, class _Allocator>
 basic_string<_CharT, _Traits, _Allocator>&
 basic_string<_CharT, _Traits, _Allocator>::assign(const value_type* __s, size_type __n)
 {
     _LIBCPP_ASSERT(__n == 0 || __s != nullptr, "string::assign received nullptr");
     size_type __cap = capacity();

 basic_string<_CharT, _Traits, _Allocator>::operator=(const basic_string& __str)
 {
   if (this != &__str) {
     __copy_assign_alloc(__str);
     if (!__is_long()) {
       if (!__str.__is_long()) {
         __r_.first().__r = __str.__r_.first().__r;
       } else {
-        __assign_no_alias<true>(__str.data(), __str.size());
+        return __assign_no_alias<true>(__str.data(), __str.size());
       }
     } else {
-      __assign_no_alias<false>(__str.data(), __str.size());
+      return __assign_no_alias<false>(__str.data(), __str.size());
     }
   }
   return *this;
 }
 
 #ifndef _LIBCPP_CXX03_LANG
 
 template <class _CharT, class _Traits, class _Allocator>