Discount for inactivity timeout for recording session length

chrome/browser/metrics/desktop_engagement/desktop_engagement_service.cc

Index: chrome/browser/metrics/desktop_engagement/desktop_engagement_service.cc
diff --git a/chrome/browser/metrics/desktop_engagement/desktop_engagement_service.cc b/chrome/browser/metrics/desktop_engagement/desktop_engagement_service.cc
index 92e6c59601e2a4dcaed014cd8a6ef4c6081c6b84..b64ebddc13b028f712169397fd12b0f2265262eb 100644
--- a/chrome/browser/metrics/desktop_engagement/desktop_engagement_service.cc
+++ b/chrome/browser/metrics/desktop_engagement/desktop_engagement_service.cc
@@ -117,6 +117,11 @@ void DesktopEngagementService::EndSession() {
   in_session_ = false;
 
   base::TimeDelta delta = base::TimeTicks::Now() - session_start_;
+
+  // If timer is not running then session ended because of inactivity.
+  if (!timer_.IsRunning())

[Alexei Svitkine (slow), 2016/09/08 21:20:14]

Is there a way we can have a test for this?

I'm just worried about some case where this invariant could change in the
future.

Also, please update histograms.xml description of the metric noting this M55+
logic change on desktop,

[gayane -on leave until 09-2017, 2016/09/09 18:11:03]

Done.

+    delta -= inactivity_timeout_;
+
   DVLOG(4) << "Logging session length of " << delta.InSeconds() << " seconds.";
 
   // Note: This metric is recorded separately for Android in