Add UMA for High Resolution Timer Usage

base/time/time_win_unittest.cc

Index: base/time/time_win_unittest.cc
diff --git a/base/time/time_win_unittest.cc b/base/time/time_win_unittest.cc
index 17c722300d681d67d0010503bace12773b236585..b81701455ee6a5a17cf03363f6f8d2a408aded10 100644
--- a/base/time/time_win_unittest.cc
+++ b/base/time/time_win_unittest.cc
@@ -300,4 +300,40 @@ TEST(TimeDelta, ConstexprInitialization) {
   EXPECT_EQ(kExpectedDeltaInMilliseconds, kConstexprTimeDelta.InMilliseconds());
 }
 
+TEST(HighResolutionTimer, GetUsage) {
+  EXPECT_EQ(0.0, Time::GetHighResolutionTimerUsage());
+
+  Time::ResetHighResolutionTimerUsage();
+  EXPECT_EQ(0.0, Time::GetHighResolutionTimerUsage());

[gab, 2017/06/28 16:39:33]

Why is this guaranteed? Can't Now() tick between Reset() and Get()?

[stanisc, 2017/06/28 22:28:35]

Since the high resolution timer isn't activated yet it doesn't matter how much
timer elapses since the reset. I'll change the test comments to make that more
obvious.

+
+  Sleep(10);
+  // 0% usage since the timer isn't activated.
+  EXPECT_EQ(0.0, Time::GetHighResolutionTimerUsage());
+
+  Time::ActivateHighResolutionTimer(true);
+  Time::ResetHighResolutionTimerUsage();
+
+  Sleep(20);
+  // 100% usage since the timer has been activated entire time.
+  EXPECT_EQ(100.0, Time::GetHighResolutionTimerUsage());
+
+  Time::ActivateHighResolutionTimer(false);
+  Sleep(20);
+  double usage1 = Time::GetHighResolutionTimerUsage();
+  // usage1 should be about 50%.
+  EXPECT_LT(usage1, 100.0);
+  EXPECT_GT(usage1, 0.0);

[gab, 2017/06/28 16:39:33]

Precise enough to use EXPECT_NEAR? (and below)

[stanisc, 2017/06/28 22:28:35]

In my experience anything involving time checks is very unreliable in tests that
run in swarming environment. That Sleep(20) could sleep for an arbitrary amount
on time so the actual usage could be pretty far from 50%. All I can tell for
sure that it must be greater than 0% and less than 100%.

I considered adding a way to override Now() function similar to 
TimeTicks::SetMockTickFunction(). That would be the most reliable way to deal
with time. But that seemed fragile because the test infrastructure likely uses
TimeTicks::Now too.

+
+  Time::ActivateHighResolutionTimer(true);
+  Sleep(10);
+  Time::ActivateHighResolutionTimer(false);
+  double usage2 = Time::GetHighResolutionTimerUsage();
+  // usage2 should be about 60%.
+  EXPECT_LT(usage2, 100.0);
+  EXPECT_GT(usage2, usage1);
+
+  Time::ResetHighResolutionTimerUsage();
+  EXPECT_EQ(0.0, Time::GetHighResolutionTimerUsage());
+}
+
 }  // namespace base