_PyTime_Divide() rounding was wrong: copy code from Python default which has
now much better unit tests.

Overflow test in test_FromSecondsObject() fails on FreeBSD 10.0 buildbot which
uses clang. clang implements more aggressive optimization which gives
different result than GCC on undefined behaviours.

Check if a multiplication will overflow, instead of checking if a
multiplicatin had overflowed, to avoid undefined behaviour.

Add also debug information if the test on overflow fails.

Filter also values for check_float_rounding().

Filter values which would overflow when converted to a C time_t type.

* Filter values which would overflow on conversion to the C long type
  (for timeval.tv_sec).
* Adjust also the message of OverflowError on PyTime conversions
* test_time: add debug information if a timestamp conversion fails

Drop all hardcoded tests. Instead, reimplement each function in Python, usually
using decimal.Decimal for the rounding mode.

Add much more values to the dataset. Test various timestamp units from
picroseconds to seconds, in integer and float.

Enhance also _PyTime_AsSecondsDouble().

datetime.datetime now round microseconds to nearest with ties going to nearest
even integer (ROUND_HALF_EVEN), as round(float), instead of rounding towards
-Infinity (ROUND_FLOOR).

pytime API: replace _PyTime_ROUND_HALF_UP with _PyTime_ROUND_HALF_EVEN. Fix
also _PyTime_Divide() for negative numbers.

_PyTime_AsTimeval_impl() now reuses _PyTime_Divide() instead of reimplementing
rounding modes.

datetime.datetime.fromtimestamp() and datetime.datetime.utcfromtimestamp().
microseconds sign should be kept before rounding.

_PyTime_ObjectToTimespec()

rounding

used.

Fix also _Py_InitializeEx_Private(): initialize time before initializing
import, import_init() uses the _PyTime API (for thread locks).

_PyTime_AsMicroseconds() rounding.

Add also unit tests.

Use _PyTime_ROUND_FLOOR and _PyTime_ROUND_CEILING instead.

Add also more tests for ROUNd_FLOOR.

- _PyTime_ObjectToTime_t()
- _PyTime_ObjectToTimespec()
- _PyTime_ObjectToTimeval()

module. time.clock_settime() now uses this rounding method instead of
_PyTime_ROUND_DOWN to handle correctly dates before 1970.

* _PyTime_AsTimeval() now ensures that tv_usec is always positive
* _PyTime_AsTimespec() now ensures that tv_nsec is always positive
* _PyTime_AsTimeval() now returns an integer on overflow instead of raising an
  exception

* Add _PyTime_AsTimespec()
* Add unit tests for _PyTime_AsTimespec()

* Add _PyTime_FromNanoseconds()
* Add _PyTime_AsSecondsDouble()
* Add unit tests for _PyTime_AsSecondsDouble()

* Rename _PyTime_FromObject() to _PyTime_FromSecondsObject()
* Add _PyTime_AsNanosecondsObject() and _testcapi.pytime_fromsecondsobject()
* Add unit tests

is used as a sentinel () to detect if -1 is really an error or a valid
timestamp. On AIX, tm_wday is unchanged even on success and so cannot be used
as a sentinel.

outsize range [1902; 2037].

aways from zero, instead of rounding towards zero.

It should make test_asyncio more reliable, especially test_timeout_rounding() test.

time.get_clock_info('monotonic').resolution == 0.015600099999999999

This is just a workaround pending a real resolution to #20101.

busy/slow buildbot

time.strptime().  Initial patch by Claudiu Popa.