Test failure in datefmt module on OSX

[Ryan J Ollos]

======================================================================
FAIL: test_pytz_choibalsan (trac.util.tests.datefmt.LocalTimezoneTestCase)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/rjollos/Documents/Workspace/trac-dev/teo-rjollos.git/trac/util/tests/datefmt.py", line 1723, in test_pytz_choibalsan
    self._compare_pytz(tz, '1983-10-01T00:00')  # -1:00 (DST end)
  File "/Users/rjollos/Documents/Workspace/trac-dev/teo-rjollos.git/trac/util/tests/datefmt.py", line 1711, in _compare_pytz
    self._compare_pytz_localize_and_normalize(tz, dt_naive)
  File "/Users/rjollos/Documents/Workspace/trac-dev/teo-rjollos.git/trac/util/tests/datefmt.py", line 1688, in _compare_pytz_localize_and_normalize
    '%r != %r (%r)' % (dt_tz, dt_localtz, dt_naive))
AssertionError: datetime.datetime(1983, 9, 30, 22, 0, tzinfo=<DstTzInfo 'Asia/Choibalsan' CHOST+10:00:00 DST>) != datetime.datetime(1983, 9, 30, 22, 0, tzinfo=<LocalTimezone "UTC+09:00" 9:00:00>) (datetime.datetime(1983, 9, 30, 22, 0))

----------------------------------------------------------------------

[Ryan J Ollos]

Could be unrelated to OSX, rather due to using a newer version of pytz. I haven't tested on any other platforms yet.

>>> import pytz
>>> pytz.__version__
'2015.2'

The failing assertions are trunk/trac/util/tests/datefmt.py@12788:1723-1725#L1713​, confirmed by commenting out the failing assertions so that other assertions in test case would be executed.

[Ryan J Ollos]

I haven't been able to reproduce on Windows 7 with pytz 2015.2, but the issue persists on OSX.

[Ryan J Ollos]

After seeing this issue for months, I can no longer reproduce with either pytz 2015.2 or 2015.4. Maybe there was an update to OSX that fixed the issue? I'll reopen if the issue occurs again.

[Jun Omae]

Reproduced with Mac OS X build environment on Travis CI, while trying #12208. Perhaps, mktime in libc on Mac OS X doesn't have inconsistency…. I'll investigate it.

[Jun Omae]

After the patch, unit and functional tests pass on Mac OS X. See ​https://travis-ci.org/jun66j5/trac/builds/118786133.

trac/util/tests/datefmt.py

@@ -1817 +1817 @@
             self._compare_pytz(tz, '1982-04-01T00:00')  # No DST
             self._compare_pytz(tz, '1983-04-01T00:00')  # +2:00 (TZ change)
             self._compare_pytz(tz, '1983-04-01T02:00')  #       (TZ change)
-            self._compare_pytz(tz, '1983-10-01T00:00')  # -1:00 (DST end)
+            self._compare_pytz(tz, '1983-10-01T00:00',  # -1:00 (DST end)
+                               localize=False)
             self._compare_pytz(tz, '2006-03-25T02:00')  # +1:00 (DST start)
-            self._compare_pytz(tz, '2006-09-30T02:00')  # -1:00 (DST end)
+            self._compare_pytz(tz, '2006-09-30T02:00',  # -1:00 (DST end)
+                               localize=False)
             self._compare_pytz(tz, '2007-07-01T00:00')  # No DST in 2007
             self._compare_pytz(tz, '2008-03-30T23:00',  #       (TZ change)
                                localize=False)

[Ryan J Ollos]

Travis CI is ​currently using OSX 10.9.5. I don't recall the version of OSX that I was running when I discovered the issue, but I'm currently running 10.11.4 and I don't see the issue. I guess it's probably difficult to determine the exact version of OSX in which the issue is fixed.

[Jun Omae]

Thanks. I'm using OSX 10.10.5 (version of libSystem.dylib is 1213.0.0). This issue cannot be reproduced on my environment. The test cases isn't important. It's no problem to skip the cases.

$ sw_vers -productVersion
10.10.5
$ ls -l /usr/lib/libc.*
lrwxr-xr-x  1 root  wheel  15 Jun 26  2015 /usr/lib/libc.dylib -> libSystem.dylib
$ otool -L /usr/lib/libc.dylib | grep libSystem
        /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 1213.0.0)

[Jun Omae]

Committed in [14678] and merged to trunk in [14679].