Makes GetBuildTime behave identically on all build types.

base/build_time.h

Index: base/build_time.h
diff --git a/base/build_time.h b/base/build_time.h
index 4f0abc381336641425dab76d4f1533a0041b22bb..8014cf1d242815b93dddddecc1c791186de16cdb 100644
--- a/base/build_time.h
+++ b/base/build_time.h
@@ -10,17 +10,13 @@
 
 namespace base {
 
-// GetBuildTime returns the time at which the current binary was built.
+// GetBuildTime returns the time at which the current binary was built,
+// rounded down to midnight at the start of the day in UTC.
 //
-// This uses the __DATE__ and __TIME__ macros, which don't trigger a rebuild
-// when they change. However, official builds will always be rebuilt from
-// scratch.
+// This uses a generated file, which doesn't trigger a rebuild when the time
+// changes. However, official builds will always be built from scratch.
 //
-// Also, since __TIME__ doesn't include a timezone, this value should only be
-// considered accurate to a day.
-//
-// NOTE: This function is disabled except for the official builds, by default
-// the date returned is "Sep 02 2008 08:00:00 PST".
+// This value should only be considered accurate to within a day.

[lgarron, 2016/02/02 04:03:11]

Unless it's off by a month. ;-)

Would you mind documenting and enforcing a guarantee that the returned time is
in the past, relative to the computer running the build? IsUserClockInThePast()
it requires that the build time is in the past, otherwise you will get strange
results [1]. It seems that the CT code would also prefer to avoid a build date
in the past.
(If "in the past" is too strong, "at most one day ahead" is probably fine.)

In particular, could you add a unit test with a check that `GetBuildTime() <
__DATE__`?
The test will require a recompile every time, but it's trading that recompile
against security-sensitive correctness on the build speedup.

[1] In particular, if the build date is at least 2 days the future, an expired
certificate will show a "bad clock" warning even if your clock is fine.
https://code.google.com/p/chromium/issues/detail?id=414843#c16

[Zachary Forman, 2016/02/02 06:30:38]

Have done!

[lgarron, 2016/02/02 19:38:30]

Thanks!

I'm not a real owner, but LGTM for this test and for
components/ssl_errors/error_classification.cc

 Time BASE_EXPORT GetBuildTime();
 
 }  // namespace base