Add thread checking to RunLoop, deprecate MessageLoopRunner.

base/run_loop.h

Index: base/run_loop.h
diff --git a/base/run_loop.h b/base/run_loop.h
index 635018f434ec1a69c331366d80506ab2dc6a2ce0..077d097ba9a2a35c6ec51d017f588b9f195e48c0 100644
--- a/base/run_loop.h
+++ b/base/run_loop.h
@@ -10,6 +10,7 @@
 #include "base/macros.h"
 #include "base/memory/weak_ptr.h"
 #include "base/message_loop/message_loop.h"
+#include "base/threading/thread_checker.h"
 #include "build/build_config.h"
 
 namespace base {
@@ -105,6 +106,8 @@ class BASE_EXPORT RunLoop {
   // that we should quit Run once it becomes idle.
   bool quit_when_idle_received_;
 
+  base::ThreadChecker thread_checker_;

[Nico, 2016/12/02 01:10:35]

should this be in an #if DCHECK_IS_ON() block?

[Alexander Semashko, 2016/12/02 10:47:34]

It won't compile this way: code inside DCHECK's still need to be correct.
Also, it makes little of a difference, since ThreadChecker is an empty class
when DCHECKs are off.

But there is a difference, because even empty objects have non-zero size. So
ThreadChecker can add 8 bytes (4 in 32-bit builds) to the size of objects that
use it, including ones in production code. This is not the case with
NonThreadSafe which is used as a base class instead of a member. But comments in
thread_checker.h say that ThreadChecker should be used wherever possible.
Why is it so? Shouldn't the use of NonThreadSafe be encouraged because of that?

I think this broader question should be resolved first. Then, if there is a
change in policy, we can update RunLoop as well as other classes. For now, imo,
it does not need to be special-cased, because it is definitely not used a lot in
production.

+
   // WeakPtrFactory for QuitClosure safety.
   base::WeakPtrFactory<RunLoop> weak_factory_;