Fuchsia port of base/time, with some refactoring of POSIX time modules.

base/time/time_fuchsia.cc

+// Copyright 2017 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "base/time/time.h"
+
+#include <magenta/syscalls.h>
+
+namespace base {
+
+// Time -----------------------------------------------------------------------
+
+// static
+Time Time::Now() {
+  const mx_time_t nanos_since_unix_epoch = mx_time_get(MX_CLOCK_UTC);
+  CHECK(nanos_since_unix_epoch != 0);

[Lei Zhang, 2017/05/17 08:34:17]

We handle gettimeofday() failures in time_now_posix.cc. Should we be consistent
one way or the other?

[miu, 2017/05/18 02:52:38]

Done (changes to time_now_posix.cc, to not tolerate failures). Notes on why I
changed things the way I did:

1. Given recent discussion on chromium-dev@, the CHECK() is much more the
preferred style over "if (some expression) { NOTREACHED(); }".

2. I re-examined the POSIX documentation for clock_gettime() and gettimeofday().
Based on the description of the return values and all the possible error codes,
only a bug in our code (e.g., bad pointer to struct) should ever trip an error.
Also, we check at compile-time that the clock ID argument is supported by the
platform. Thus, CHECK() should be appropriate.

+  const mx_time_t usec_since_unix_epoch =

[Lei Zhang, 2017/05/17 08:34:17]

What is mx_time_t underneath? Can we get away with making this int64_t to avoid
the static_casts?

[miu, 2017/05/18 02:52:38]

Good point. I had looked into that, and it's uint64_t.

I am being very pedantic about where the unsigned math is being done versus the
signed math because it scares me to depend on the subtle implicit conversion
rules of the C++ language. ;-) (It has bitten me before!) For example, OTOH, I
don't know what gets implicitly converted when an unsigned 64-bit int is divided
by a signed 64-bit int. The numerator? The denominator? Also, is the result of
the expression an unsigned or signed type?

[Lei Zhang, 2017/05/18 19:06:12]

Ya, I can't remember all the rules off the top of my head either. Let's back up
our assumptions with tests.

BTW, do you want to add a helper function, since we do the same divison by
|kNanosecondsPerMicrosecond| + casts in 4 different places?

[miu, 2017/05/18 20:39:04]

Done.

+      nanos_since_unix_epoch /
+      static_cast<mx_time_t>(kNanosecondsPerMicrosecond);
+  // The following expression will overflow in the year 289938 A.D.:
+  return Time(static_cast<int64_t>(usec_since_unix_epoch) +
+              kTimeTToMicrosecondsOffset);
+}
+
+// static
+Time Time::NowFromSystemTime() {
+  return Now();
+}
+
+// TimeTicks ------------------------------------------------------------------
+
+// static
+TimeTicks TimeTicks::Now() {
+  const mx_time_t nanos_since_boot = mx_time_get(MX_CLOCK_MONOTONIC);
+  CHECK(nanos_since_boot != 0);
+  const mx_time_t usec_since_boot =
+      nanos_since_boot / static_cast<mx_time_t>(kNanosecondsPerMicrosecond);
+  return TimeTicks(static_cast<int64_t>(usec_since_boot));
+}
+
+// static
+TimeTicks::Clock TimeTicks::GetClock() {
+  return Clock::FUCHSIA_MX_CLOCK_MONOTONIC;
+}
+
+// static
+bool TimeTicks::IsHighResolution() {
+  return true;
+}
+
+// static
+bool TimeTicks::IsConsistentAcrossProcesses() {
+  return true;
+}
+
+// static
+TimeTicks TimeTicks::FromMXTime(mx_time_t nanos_since_boot) {
+  const mx_time_t usec_since_boot =
+      nanos_since_boot / static_cast<mx_time_t>(kNanosecondsPerMicrosecond);
+  return TimeTicks(static_cast<int64_t>(usec_since_boot));
+}
+
+// static
+ThreadTicks ThreadTicks::Now() {
+  const mx_time_t nanos_since_thread_started = mx_time_get(MX_CLOCK_THREAD);
+  CHECK(nanos_since_thread_started != 0);
+  const mx_time_t usec_since_thread_started =
+      nanos_since_thread_started /
+      static_cast<mx_time_t>(kNanosecondsPerMicrosecond);
+  return ThreadTicks(static_cast<int64_t>(usec_since_thread_started));
+}
+
+}  // namespace base