Add explicit TimeDelta constexpr copy constructor.

base/time/time_unittest.cc

Index: base/time/time_unittest.cc
diff --git a/base/time/time_unittest.cc b/base/time/time_unittest.cc
index 4f47d56522b0d30b883f86b236110ca9722750f5..8e880edf5182105b8f72766847554dc3da2d1730 100644
--- a/base/time/time_unittest.cc
+++ b/base/time/time_unittest.cc
@@ -21,6 +21,11 @@ namespace base {
 
 namespace {
 
+// For TimeDelta::ConstexprInitialization
+enum { kExpectedDeltaInMilliseconds = 10 };

[miu, 2016/08/11 21:48:56]

Is it necessary for this to be an enum to trigger the compiler error? Otherwise,
nit: Please make this constexpr int64_t kExpected... = 10;

[liberato (no reviews please), 2016/08/11 22:10:35]

it's not necessary.

originally, i had it as constexpr.  it seemed a bit like a red-herring, given
the subject of the test.  i'll switch it back.

+constexpr TimeDelta kConstexprTimeDelta =
+    TimeDelta::FromMilliseconds(kExpectedDeltaInMilliseconds);
+
 TEST(TimeTestOutOfBounds, FromExplodedOutOfBoundsTime) {
   // FromUTCExploded must set time to Time(0) and failure, if the day is set to
   // 31 on a 28-30 day month. Test |exploded| returns Time(0) on 31st of
@@ -1093,6 +1098,11 @@ TEST(TimeDelta, Overflows) {
   EXPECT_EQ(one_second, (ticks_now + one_second) - ticks_now);
 }
 
+TEST(TimeDelta, ConstexprInitialization) {

[miu, 2016/08/11 21:48:56]

Please put these changes in time_win_unittest.cc instead, since they're
addressing a Windows compiler-specific bug.

[liberato (no reviews please), 2016/08/11 22:10:35]

Done.

+  // Make sure that TimeDelta works around crbug.com/635974
+  EXPECT_EQ(kExpectedDeltaInMilliseconds, kConstexprTimeDelta.InMilliseconds());
+}
+
 TEST(TimeDeltaLogging, DCheckEqCompiles) {
   DCHECK_EQ(TimeDelta(), TimeDelta());
 }