base: Make it possible to replace the MessageLoop's task runner

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 74bfdb6f4ef6c2d16f66891c8189f8a5ffb76613..e3d548be07ed8562efb39e3e3ca3bff8a0a0a339 100644
--- a/base/message_loop/message_loop.cc
+++ b/base/message_loop/message_loop.cc
@@ -367,6 +367,7 @@ bool MessageLoop::IsIdleForTesting() {
 
 //------------------------------------------------------------------------------
 
+// static
 scoped_ptr<MessageLoop> MessageLoop::CreateUnbound(
     Type type, MessagePumpFactoryCallback pump_factory) {
   return make_scoped_ptr(new MessageLoop(type, pump_factory));
@@ -403,7 +404,23 @@ void MessageLoop::BindToCurrentThread() {
 
   incoming_task_queue_->StartScheduling();
   task_runner_->BindToCurrentThread();
-  thread_task_runner_handle_.reset(new ThreadTaskRunnerHandle(task_runner_));
+  SwapTaskRunner(task_runner_);

[danakj, 2015/06/29 19:07:23]

Why do we want to support both:

1. SwapTaskRunner
2. BindToCurrentThread

and

1. BindToCurrentThread
2. SwapTaskRunner

?

[danakj, 2015/06/29 19:07:23]

can we DCHECK that the return from this is null?

can you .Pass() the task_runner_?

[alex clarke (OOO till 29th), 2015/06/30 10:51:43]

I don't know, and Sami is currently OOO so I can't ask him.

[alex clarke (OOO till 29th), 2015/06/30 10:51:43]

Done.

+}
+
+scoped_refptr<BindableSingleThreadTaskRunner> MessageLoop::SwapTaskRunner(
+    scoped_refptr<BindableSingleThreadTaskRunner> task_runner) {
+  // If this message loop was already bound to a thread, ensure the task runner
+  // belongs to the same thread and update the thread task runner handle.
+  // Otherwise the task runner will be bound later in BindToCurrentThread().
+  if (pump_) {
+    DCHECK_EQ(this, current());
+    DCHECK(task_runner->BelongsToCurrentThread());
+    // This reset() is needed or ThreadTaskRunnerHandle::Get() will crash.

[danakj, 2015/06/29 19:07:23]

We're not calling Get() here.. can you expand the explanation in this comment?

[alex clarke (OOO till 29th), 2015/06/30 10:51:43]

It's there because in the original review kinuko asked why there are two resets.
 This does look weird, and it's because reset() doesn't behave the way we would
expect.

+    thread_task_runner_handle_.reset();
+    thread_task_runner_handle_.reset(new ThreadTaskRunnerHandle(task_runner));
+  }
+  task_runner_.swap(task_runner);
+  return task_runner;
 }
 
 void MessageLoop::RunHandler() {