Remove dependency on rtc::Thread and rtc_base from audio_mixer_unittests.

webrtc/modules/audio_mixer/audio_mixer_impl_unittest.cc

Index: webrtc/modules/audio_mixer/audio_mixer_impl_unittest.cc
diff --git a/webrtc/modules/audio_mixer/audio_mixer_impl_unittest.cc b/webrtc/modules/audio_mixer/audio_mixer_impl_unittest.cc
index 3ac61892b3094614c9ab2219e913200c407cc590..ac7783b1e8e6d725017cc28338597e91bdd5b31a 100644
--- a/webrtc/modules/audio_mixer/audio_mixer_impl_unittest.cc
+++ b/webrtc/modules/audio_mixer/audio_mixer_impl_unittest.cc
@@ -21,7 +21,8 @@
 #include "webrtc/modules/audio_mixer/default_output_rate_calculator.h"
 #include "webrtc/rtc_base/bind.h"
 #include "webrtc/rtc_base/checks.h"
-#include "webrtc/rtc_base/thread.h"
+#include "webrtc/rtc_base/event.h"
+#include "webrtc/rtc_base/task_queue.h"
 #include "webrtc/test/gmock.h"
 
 using testing::_;
@@ -373,23 +374,27 @@ TEST(AudioMixer, RampedOutSourcesShouldNotBeMarkedMixed) {
 // This test checks that the initialization and participant addition
 // can be done on a different thread.
 TEST(AudioMixer, ConstructFromOtherThread) {
-  std::unique_ptr<rtc::Thread> init_thread = rtc::Thread::Create();
-  std::unique_ptr<rtc::Thread> participant_thread = rtc::Thread::Create();
-  init_thread->Start();
-  const auto mixer = init_thread->Invoke<rtc::scoped_refptr<AudioMixer>>(
-      RTC_FROM_HERE,
-      // Since AudioMixerImpl::Create is overloaded, we have to
-      // specify the type of which version we want.
-      static_cast<rtc::scoped_refptr<AudioMixerImpl>(*)()>(
-          &AudioMixerImpl::Create));
+  rtc::TaskQueue init_queue("init");
+  rtc::scoped_refptr<AudioMixer> mixer;
+  rtc::Event event(false, false);
+  init_queue.PostTask([&mixer, &event]() {
+    mixer = AudioMixerImpl::Create();
+    event.Set();
+  });
+  event.Wait(rtc::Event::kForever);
+
   MockMixerAudioSource participant;
+  EXPECT_CALL(participant, PreferredSampleRate())

[minyue-webrtc, 2017/07/11 11:16:20]

would you explain why this additional expected call?

[tommi, 2017/07/11 11:58:36]

This call is and should always be made so I'm setting the expectation and
avoiding polluting the log.  That wasn't done previously.

There's an ON_CALL() handler set above, but that one only sets the default
handling and prints an error when it's called - which because it always happens,
we get trained to ignore (which is not a good thing).

Take for example a look at the "GMOCK WARNING" lines for the AudioMixer tests in
these runs:

https://isolateserver.appspot.com/browse?namespace=default-gzip&digest=5961e2...

https://isolateserver.appspot.com/browse?namespace=default-gzip&digest=f88549...

(etc - same amount of noise from most of the AudioMixer tests)

Then look at the output from the ConstructFromOtherThread test when it runs
normally (with this patch):

https://isolateserver.appspot.com/browse?namespace=default-gzip&digest=799c70...

[minyue-webrtc, 2017/07/11 13:03:22]

Acknowledged.

I also advice using StrictMock<MockMixerAudioSource> if possible

[tommi, 2017/07/11 13:13:29]

Yes that would be an improvement.  This CL is about removing use of rtc::Thread
though so I think that's a separate goal.
Alex - could you take a look at that at some point?

+      .WillRepeatedly(Return(kDefaultSampleRateHz));
 
   ResetFrame(participant.fake_frame());
 
-  participant_thread->Start();
-  EXPECT_TRUE(participant_thread->Invoke<int>(
-      RTC_FROM_HERE,
-      rtc::Bind(&AudioMixer::AddSource, mixer.get(), &participant)));
+  rtc::TaskQueue participant_queue("participant");
+  participant_queue.PostTask([&mixer, &event, &participant]() {
+    mixer->AddSource(&participant);
+    event.Set();
+  });
+  event.Wait(rtc::Event::kForever);
 
   EXPECT_CALL(participant, GetAudioFrameWithInfo(kDefaultSampleRateHz, _))
       .Times(Exactly(1));