Fix a "benign" data race during thread construction

base/message_loop/message_loop.cc

Index: base/message_loop/message_loop.cc
diff --git a/base/message_loop/message_loop.cc b/base/message_loop/message_loop.cc
index 4fecbc5a60383ac5c5a6cd84b92edac99afa6223..d4f9c546b58ae3e3ceabcd7e6b723ab85de81502 100644
--- a/base/message_loop/message_loop.cc
+++ b/base/message_loop/message_loop.cc
@@ -407,7 +407,8 @@ void MessageLoop::BindToCurrentThread() {
 
   incoming_task_queue_->StartScheduling();
   unbound_task_runner_->BindToCurrentThread();
-  SetTaskRunner(unbound_task_runner_.Pass());
+  unbound_task_runner_ = nullptr;

[danakj, 2015/07/10 18:32:38]

I'm feeling like this unbound_task_runner_ is providing less value and more
confusion over time.

What is is doing more than a bool bound_to_thread_ that only exists when
DCHECK_IS_ON()?

[Sami, 2015/07/13 10:43:31]

Yes, it's not all that useful but I can't think of a simpler way to do this. A
bool won't cut it because we still need an object on which we can call
BindToCurrentThread(). If we changed the type of task_runner_ to something that
has BindToCurrentThread(), then we'd need to leak that interface to the outside
world in SetTaskRunner.

The only other way I can think of is to keep task_runner_ a
SingleThreadTaskRunner and then do a downcast in BindToCurrentThread and add the
debug-only boolean. WDYT?

[danakj, 2015/07/13 18:33:47]

Ah, I see. LGTM then.

+  SetThreadTaskRunnerHandle();
 }
 
 void MessageLoop::SetTaskRunner(
@@ -416,6 +417,11 @@ void MessageLoop::SetTaskRunner(
   DCHECK(task_runner->BelongsToCurrentThread());
   DCHECK(!unbound_task_runner_);
   task_runner_ = task_runner.Pass();
+  SetThreadTaskRunnerHandle();
+}
+
+void MessageLoop::SetThreadTaskRunnerHandle() {
+  DCHECK_EQ(this, current());
   // Clear the previous thread task runner first because only one can exist at
   // a time.
   thread_task_runner_handle_.reset();