Recreate WebrtcVideoSendStream if screen content setting is changed.

webrtc/media/engine/fakewebrtcvideoengine.h

Index: webrtc/media/engine/fakewebrtcvideoengine.h
diff --git a/webrtc/media/engine/fakewebrtcvideoengine.h b/webrtc/media/engine/fakewebrtcvideoengine.h
index 497d70133ad20881081063be7ab5f6f192d18197..2dce2c503fcf67e284ef0d3d84027a6dc7ffd74a 100644
--- a/webrtc/media/engine/fakewebrtcvideoengine.h
+++ b/webrtc/media/engine/fakewebrtcvideoengine.h
@@ -196,10 +196,13 @@ class FakeWebRtcVideoEncoderFactory : public WebRtcVideoEncoderFactory {
   }
 
   bool WaitForCreatedVideoEncoders(int num_encoders) {
-    while (created_video_encoder_event_.Wait(kEventTimeoutMs)) {
+    int64_t start_offset_ms = rtc::TimeMillis();
+    int64_t wait_time = kEventTimeoutMs;
+    do {
       if (GetNumCreatedEncoders() >= num_encoders)
         return true;
-    }
+      wait_time = kEventTimeoutMs - (rtc::TimeMillis() - start_offset_ms);
+    } while (wait_time > 0 && created_video_encoder_event_.Wait(wait_time));

[Taylor Brandstetter, 2017/02/23 00:46:49]

Why are start_offset_ms/wait_time needed?

[sprang_webrtc, 2017/02/23 15:10:14]

It was racy, so if |num_encoders| reached X before a call to
WaitForCreatedVideoEncoders(X), it would block for kEventTimeoutMs and return
false.

This update is also a little stricter and wait for a total of kEventTimeoutMs,
instead of resetting the timer for each update of the count until it reaches the
desired value.

Turns out this was actually needed for my test though, so if you feel it's
unnecessary I can remove it.

[Taylor Brandstetter, 2017/02/24 00:13:55]

Ah, I see. I think just changing it to a "do while" is sufficient though; the
stricter waiting seems unnecessary. I don't feel strongly about it though.

     return false;
   }