Delete SignalThread class.

webrtc/base/messagequeue.cc

Index: webrtc/base/messagequeue.cc
diff --git a/webrtc/base/messagequeue.cc b/webrtc/base/messagequeue.cc
index cafb70bd002faaf3241db5f9f4633b70767cb887..95d2045a4a00c8247c55a05355e57d93da37a3f7 100644
--- a/webrtc/base/messagequeue.cc
+++ b/webrtc/base/messagequeue.cc
@@ -44,6 +44,18 @@ class SCOPED_LOCKABLE DebugNonReentrantCritScope {
 
   RTC_DISALLOW_COPY_AND_ASSIGN(DebugNonReentrantCritScope);
 };
+
+class FunctorPostMessageHandler : public MessageHandler {
+ public:
+  void OnMessage(Message* msg) override;
+};
+
+void FunctorPostMessageHandler::OnMessage(Message* msg) {
+  CallableData* data = static_cast<CallableData*>(msg->pdata);
+  data->call();
+  delete data;
+}

[kwiberg-webrtc, 2017/06/26 13:07:48]

Define this function inline? There's no downside to that when you define a class
in a .cc file.

[nisse-webrtc, 2017/06/27 08:27:17]

Done.

+
 }  // namespace
 
 //------------------------------------------------------------------
@@ -529,4 +541,12 @@ void MessageQueue::Dispatch(Message *pmsg) {
   }
 }
 
+void MessageQueue::PostFunctorInternal(const Location& posted_from,
+                                       CallableData* message_data)
+{

[tommi, 2017/06/27 12:01:14]

can you run git cl format?

[nisse-webrtc, 2017/06/28 11:47:54]

Already done, in ps#14.

+  // FunctorPostMessageHandler keeps no state, create a static object.
+  RTC_DEFINE_STATIC_LOCAL(FunctorPostMessageHandler, handler, ());

[kwiberg-webrtc, 2017/06/26 13:07:48]

Why not just

  FunctorPostMessageHandler handler;

? That class has a trivial default constructor and a trivial destructor, since
it has no members.

[nisse-webrtc, 2017/06/27 08:27:17]

It has to outlive this scope, since it's accessed (vtable lookup for OnMessage)
by the time the target thread gets to process the posted message.

But if you say a plain "static" definition works fine, I'll do that. The vtable
is initialized properly at link time, then?

[kwiberg-webrtc, 2017/06/27 09:16:28]

Ah, right, it has a vtable. But yes, as of C++11, function-local static
variables are threadsafely initialized (with a mutex if necessary, but as you
point out vtables might be initialized at link time), and vtables shouldn't have
to cause any work in the destructor, so a plain "static" definition works fine.
(RTC_DEFINE_STATIC_LOCAL relies on the same mechanisms for synchronizing the
creation of the object (but because of the heap allocation, we know there's
runtime initialization that needs to be mutexed). The only difference is that it
leaks the object into the heap, so that its destructor is never run, but that
should make no difference in this case.)

[nisse-webrtc, 2017/06/28 11:47:54]

Good. But it still has to be in a local scope? (If we implement some cancel
mechanism as I outlined as option (1), it would help to move it into a global
unnamed namespace, but I'm not sure if that would involved constructor magic
before main(). 

+  Post(posted_from, &handler, 0, message_data);
+}
+
 }  // namespace rtc