FakeVideoCaptureDevice: Regulate framerate without drifting

media/video/capture/fake_video_capture_device.cc

Index: media/video/capture/fake_video_capture_device.cc
diff --git a/media/video/capture/fake_video_capture_device.cc b/media/video/capture/fake_video_capture_device.cc
index 12a8a3054523ce9510f85653b19dabc7e8d32cff..902a8d07a0b7e004f857ab20fe445bab9c899a28 100644
--- a/media/video/capture/fake_video_capture_device.cc
+++ b/media/video/capture/fake_video_capture_device.cc
@@ -123,6 +123,7 @@ void FakeVideoCaptureDevice::AllocateAndStart(
 
 void FakeVideoCaptureDevice::StopAndDeAllocate() {
   DCHECK(thread_checker_.CalledOnValidThread());
+  next_frame_time_ = base::TimeTicks();
   client_.reset();
 }
 
@@ -202,8 +203,20 @@ void FakeVideoCaptureDevice::BeepAndScheduleNextCapture(
     FakeAudioInputStream::BeepOnce();
 
   // Reschedule next CaptureTask.
-  base::MessageLoop::current()->PostDelayedTask(FROM_HERE, next_capture,
-      base::TimeDelta::FromMilliseconds(kFakeCapturePeriodMs));
+  const base::TimeTicks current_time = base::TimeTicks::Now();
+  const base::TimeDelta frame_interval =
+      base::TimeDelta::FromMilliseconds(kFakeCapturePeriodMs);
+  if (next_frame_time_.is_null()) {

[mcasas, 2015/06/08 17:30:35]

I'm not super happy with l.209, it feels like we'll be testing
next_frame_time_.is_null() every frame, when it'll only
verify on first startup (and after StopAndDeallocate()).
Why don't you initialise it (with perhaps appropriate 
expected delay) in AllocateAndStart() ?

+    next_frame_time_ = current_time + frame_interval;
+  } else {
+    next_frame_time_ += frame_interval;
+    // Don't accumulate any debt if we are lagging behind - just post the next
+    // frame immediately and continue as normal.
+    if (next_frame_time_ < current_time)
+      next_frame_time_ = current_time;
+  }
+  base::MessageLoop::current()->PostDelayedTask(
+      FROM_HERE, next_capture, next_frame_time_ - current_time);
 }
 
 }  // namespace media