[Zope] Re: DateTime strftime problem

[Florent Guillaume]

The DateTime implementation in Zope is notoriously undertested and 
underspecified when it comes to time zones. Until recently strftime was 
quite buggy too, and as you saw it has been recoded in terms of the python 
datetime implementation now that it exists. It's very possible that a bug 
was introduced at that point, it would be useful to check with Zope 2.7's 
DateTime. Or the bug has always been there.

It think it's a good thing if DateTime can behave more regularly, which 
means be more in line with python's datetime, in corner cases.

Please submit a patch to the collector. It probably will be included in 2.9 
but not 2.8 which is strictly in maintenance mode, unless you convince us 
that it's very unlikely that code would change behavior as a result.

Florent

Jürgen Herrmann wrote:
> [ Jürgen Herrmann wrote:]
> 
>>i looked at the source of DateTime::strftime(), surpirse, surprise :)
>>strftime uses python's datetime class and it's strftime method!
>>but no care is taken at this time for timezone information, so i
>>decided to code a tzinfo subclass for datetime that can represent
>>fixed offset from gmt (no dst) and hand one such instance to
>>datetime.fromtimestamp(). seems like this code is working correctly now.
>>i'll run it against some tests tomorrow and report back on the results.
>>
>>regards, juergen herrmann
> 
> 
> if i run the DateTime testsuite against my patched version:
> ======================================================================
> ERROR: Checks strftime in dates <= 1900 or >= 2038
> ----------------------------------------------------------------------
> Traceback (most recent call last):
>   File "/usr/lib/python2.4/unittest.py", line 260, in run
>     testMethod()
>   File "/home/bliss/zope/lib/python/DateTime/tests/testDateTime.py", line
> 366, in testStrftimeFarDates
>     self.assertEqual(dt.strftime('%d/%m/%Y'), '30/01/1900')
>   File "/home/bliss/zope/lib/python/DateTime/DateTime.py", line 1542, in
> strftime
>     ds = datetime.fromtimestamp(self._t+offsetsecs, tzi).strftime(format)
> ValueError: timestamp out of range for platform time_t
> 
> ======================================================================
> ERROR: Checks time zone in dates <= 1900 or >= 2038
> ----------------------------------------------------------------------
> Traceback (most recent call last):
>   File "/usr/lib/python2.4/unittest.py", line 260, in run
>     testMethod()
>   File "/home/bliss/zope/lib/python/DateTime/tests/testDateTime.py", line
> 374, in testZoneInFarDates
>     self.assertEqual(dt1.strftime('%d/%m/%Y %H:%M'),
> dt2.strftime('%d/%m/%Y %H:%M'))
>   File "/home/bliss/zope/lib/python/DateTime/DateTime.py", line 1542, in
> strftime
>     ds = datetime.fromtimestamp(self._t+offsetsecs, tzi).strftime(format)
> ValueError: timestamp out of range for platform time_t
> 
> ======================================================================
> FAIL: strftime timezone testing
> ----------------------------------------------------------------------
> Traceback (most recent call last):
>   File "/usr/lib/python2.4/unittest.py", line 260, in run
>     testMethod()
>   File "/home/bliss/zope/lib/python/DateTime/tests/testDateTime.py", line
> 361, in testStrftimeTZhandling
>     self.assertEqual(dt_string, dt_localstring)
>   File "/usr/lib/python2.4/unittest.py", line 333, in failUnlessEqual
>     raise self.failureException, \
> AssertionError: '2003-11-19 17:32 -0215' != '2003-11-19 21:47 GMT+0100'
> 
> ----------------------------------------------------------------------
> Ran 32 tests in 13.056s
> 
> th first two are obviously side effects from using a timestamp for
> the instantiation of my datetime object, could be fixed, not the focus
> for now.
> 
> the third one brings me back to my initial question:
> what is this code snippet supposed to return?
> 
>>>>d = DateTime('2005/04/03 02:01 UTC')
>>>>d.toZone('GMT+1').strftime('%Y/%m/%d %H:%M %Z')
> 
> '2005/04/03 03:01 GMT+0100'
> 
> is this correct?
> 
> if so, the i would tend to say, the testcase was written to pass with
> wrong strftime() behaviour. (*duck*)
> 
> regards, juergen herrmann
> 
> ps: tomorrow has been shifted to today due to a bug in my brain's
> datetime implementation, so i ran the tests today :)
> 

-- 
Florent Guillaume, Nuxeo (Paris, France)   CTO, Director of R&D
+33 1 40 33 71 59   http://nuxeo.com   fg at nuxeo.com