Delete Timing class, timing.h, and update all users.

webrtc/base/timeutils.cc

Index: webrtc/base/timeutils.cc
diff --git a/webrtc/base/timeutils.cc b/webrtc/base/timeutils.cc
index 1be368eb5847a95cca75833ee0e4d0e57644afcd..71a715d7d39b4e1864385c42c4ce68d21c7b9092 100644
--- a/webrtc/base/timeutils.cc
+++ b/webrtc/base/timeutils.cc
@@ -23,6 +23,7 @@
 #endif
 #include <windows.h>
 #include <mmsystem.h>
+#include <sys/timeb.h>
 #endif
 
 #include "webrtc/base/checks.h"
@@ -185,4 +186,21 @@ int64_t TmToSeconds(const std::tm& tm) {
             (year - 1970) * 365 + day) * 24 + hour) * 60 + min) * 60 + sec;
 }
 
+double WallClockSeconds() {

[pthatcher1, 2016/08/31 00:10:17]

Why don't we call this TimeDouble and use TimeNanos to calculate it?  That would
work better with the rest of the file, especially SetClockForTesting.

And you wouldn't need to duplicate code here between POSIX and WIN.  And you
don't treat MAC the same way.

[nisse-webrtc, 2016/08/31 06:34:41]

Because TimeNanos and related functions return monotonic time with an arbitrary
epoch (clock_gettime(CLOCK_MONOTONIC, ...) on posix), while WallClockSeconds
uses gettimeofday, so it gets time relative to the unix epoch, and will jump
backwards or forwards if the system clock is modified using settimeofday.

So they're very different. "WallClock" in the name is intended to suggest this
difference. I'm open to renaming it or moving it elsewhere (before this cl it
was the static method rtc::Timing::WallTimeNow, defined in timing.cc).
As for Macs, my understanding is that macs are posix enough to have
gettimeofday, so they don't need their own case here. (Maybe they have
clock_gettime too? I really don't know why the current SystemTimeNanos needs to
use low-level mach calls, but I guess clock_gettime for some reason is worse).

[pthatcher1, 2016/08/31 18:19:46]

But why would anything want non-monotonic time?  It looks like only the
StatsCollector is using it.  Is that intentional?  Why would we want
non-monotonic timestamps in the stats?  It sounds like it would drive people
using the stats crazy to have non-monotonic values.  

[nisse-webrtc, 2016/09/01 09:05:37]

UTC has the advantage that you can compare the timestamps on stats from
different devices. And it's *usually* monotonic.

Ultimately, I think these stats are exposed in the javascript APIs, I'll have to
leave to hbos to answer what's really required there.

+#if defined(WEBRTC_POSIX)
+  struct timeval time;
+  gettimeofday(&time, NULL);
+  // Convert from second (1.0) and microsecond (1e-6).
+  return (static_cast<double>(time.tv_sec) +
+          static_cast<double>(time.tv_usec) * 1.0e-6);
+
+#elif defined(WEBRTC_WIN)
+  struct _timeb time;
+  _ftime(&time);
+  // Convert from second (1.0) and milliseconds (1e-3).
+  return (static_cast<double>(time.time) +
+          static_cast<double>(time.millitm) * 1.0e-3);
+#endif
+}
+
 } // namespace rtc