Change flat_tree::unsafe_emplace to actually emplace

base/containers/flat_tree.h

Index: base/containers/flat_tree.h
diff --git a/base/containers/flat_tree.h b/base/containers/flat_tree.h
index 57335d8882b0268fb6d607595b4bec9579f27c55..fc9482df8d476d514ae32f62f44fbf57444ad159 100644
--- a/base/containers/flat_tree.h
+++ b/base/containers/flat_tree.h
@@ -248,7 +248,7 @@ class flat_tree {
 
  protected:
   // Emplaces a new item into the tree that is known not to be in it. This
-  // is for implementing map [] and at().
+  // is for implementing map operator[].
   template <class... Args>
   iterator unsafe_emplace(const_iterator position, Args&&... args);
 
@@ -683,8 +683,8 @@ auto flat_tree<Key, Value, GetKeyFromValue, KeyCompare>::unsafe_emplace(
     const_iterator position,
     Args&&... args) -> iterator {
   // We have to cast away const because of crbug.com/677044.
-  return impl_.body_.insert(const_cast_it(position),
-                            value_type(std::forward<Args>(args)...));
+  return impl_.body_.emplace(const_cast_it(position),
+                             std::forward<Args>(args)...);

[jdoerrie, 2017/04/04 14:14:13]

There was some discussion in the emplace thread claiming that sometimes
explicitly constructing the type is clearer and worth more than the small
performance cost of doing an extra move
(https://groups.google.com/a/chromium.org/forum/#!topic/cxx/m3cblzEta7A).

However, I think given that this is library code and this particular method is
not supposed to be called directly by customers we should try to be efficient as
possible.

[dyaroshev, 2017/04/04 15:48:21]

This wasn't the intention (at least from my point of view).

 }
 
 // For containers like sets, the key is the same as the value. This implements