Perform probing on network route change.

webrtc/modules/congestion_controller/probe_controller.cc

Index: webrtc/modules/congestion_controller/probe_controller.cc
diff --git a/webrtc/modules/congestion_controller/probe_controller.cc b/webrtc/modules/congestion_controller/probe_controller.cc
index 72d0e9ed0dbdb54b0b9698e86dcd5816527f6507..d3e43ef006c9bf10ff1f115d422433631c6bd1a8 100644
--- a/webrtc/modules/congestion_controller/probe_controller.cc
+++ b/webrtc/modules/congestion_controller/probe_controller.cc
@@ -50,16 +50,9 @@ constexpr int kRepeatedProbeMinPercentage = 70;
 ProbeController::ProbeController(PacedSender* pacer, Clock* clock)
     : pacer_(pacer),
       clock_(clock),
-      network_state_(kNetworkUp),
-      state_(State::kInit),
-      min_bitrate_to_probe_further_bps_(kExponentialProbingDisabled),
-      time_last_probing_initiated_ms_(0),
-      estimated_bitrate_bps_(0),
-      start_bitrate_bps_(0),
-      max_bitrate_bps_(0),
-      last_alr_probing_time_(clock_->TimeInMilliseconds()),
-      enable_periodic_alr_probing_(false),
-      mid_call_probing_waiting_for_result_(false) {}
+      enable_periodic_alr_probing_(false) {
+        Reset();

[stefan-webrtc, 2017/03/07 12:16:55]

I think indentation is wrong here.

+      }
 
 void ProbeController::SetBitrates(int64_t min_bitrate_bps,
                                   int64_t start_bitrate_bps,
@@ -186,6 +179,19 @@ void ProbeController::EnablePeriodicAlrProbing(bool enable) {
   enable_periodic_alr_probing_ = enable;
 }
 
+void ProbeController::Reset() {
+  rtc::CritScope cs(&critsect_);
+  network_state_ = kNetworkUp;
+  state_ = State::kInit;
+  min_bitrate_to_probe_further_bps_ = kExponentialProbingDisabled;
+  time_last_probing_initiated_ms_ = 0;
+  estimated_bitrate_bps_ = 0;
+  start_bitrate_bps_ = 0;
+  max_bitrate_bps_ = 0;
+  last_alr_probing_time_ = clock_->TimeInMilliseconds();
+  mid_call_probing_waiting_for_result_ = false;

[stefan-webrtc, 2017/03/07 12:16:55]

Can we do something like this instead:
*this = ProbeController(pacer_, clock_);

and keep the initializer list? That reduces the risk of this becoming out of
sync with the members, I think...

[philipel, 2017/03/07 13:09:13]

I can see two problems with that. First, what will happen to the pointer holding
the ProbeController? You might rewrite your local this pointer but the
CongestionController won't know about that. Second, what happens when we destroy
a lock other threads might potentially be waiting to acquire?

[stefan-webrtc, 2017/03/07 13:11:25]

Agree, that seems very risky. :)

+}
+
 void ProbeController::Process() {
   rtc::CritScope cs(&critsect_);