[nylug-talk] Python time zones

[Peter C. Norton]

I'll try to duplicate this at home when I have a few minutes, but I do
believe that the pytz package, which includes the tzinfo definitions
for most of the world, has a normalize() feature that can be used to
DTRT when performing operations around clock-changing weirdness.

-Peter

On Tue, Mar 11, 2008 at 04:31:22PM -0400, Chris Knadle wrote:
>    As previously discussed, I had a chance to talk to my friend dealing with 
> the Python timezone issues.
> 
> On Wednesday 05 March 2008, Peter C. Norton wrote:
> > On Wed, Mar 05, 2008 at 11:54:15AM -0500, Chris Knadle wrote:
> > >    Just something I wanted to mention concerning Python time zones since
> > > the subject was brought up.  A friend has run into some interesting
> > > difficulties dealing with timezones in Python relating to DST: the
> > > standard way Python subtracts time is to check if the timezones match,
> > > and if they don't time is converted to UTC first then subtracted, but if
> > > the timezones do match then the subtraction is done directly.  [Seemingly
> > > makes sense at first.]  This has an interesting consequence, though,
> > > because it means a change to DST isn't taken into account -- because a
> > > switch to DST is only taken into account during a conversion to UTC.
> >
> > The tz object in a datetime object has the offset from UTC encoded in
> > it, so if the objects were created timezone-aware, they shouldn't
> > compare as being equal. I think there's a subtlety to this
> > that hasn't come out in the story.
> 
>    It sounds like there are a couple of subtleties.  Disclaimer: I'm not much 
> of a Python programmer myself thusfar, so I'm relaying info here.
> 
>    First, Python itself only implements a general TZinfo class, so the actual 
> code to implement this class is implemented in an external library.  Two 
> current implementations are in the 'dateutils' and 'pytz' libraries.
> 
>    Second, whether a timezone is in or out of DST has to do with information 
> within a TZinfo object.  Thus, changing to DST or out of DST does not change 
> which TZinfo object a DateTime object references.
> 
>    When two DateTime objects are to be subtracted, if they have the same 
> TZinfo object (I.E. they're in the same geographical region), then they are 
> not first converted to UTC before the subtraction.  This means that DST
> *IS NOT* taken into account when subtracting two DateTime objects if they have 
> the same TZinfo object.
> 
> 
>    The following program illustrates this issue.
> 
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
> 
> #!/usr/bin/env python
> # encoding: utf-8
> 
> from dateutil.tz import gettz
> from datetime import datetime, timedelta, tzinfo
> from pytz import timezone
> import copy
> 
> NYC = gettz('America/New_York')
> UTC = gettz('UTC')
> 
> a = datetime(2008, 3, 9, 0, 0, 0, tzinfo=NYC)
> b = datetime(2008, 3, 9, 3, 0, 0, tzinfo=NYC)
> 
> print 'Time a is %s' % a.isoformat()
> print 'Time b is %s' % b.isoformat()
> print
> print 'Time a as UTC is %s' % a.astimezone(UTC).isoformat()
> print 'Time b as UTC is %s' % b.astimezone(UTC).isoformat()
> print
> print 'Therefore, the time delta when subtracting a from b should be %s' % \
>          (b.astimezone(UTC) - a.astimezone(UTC))
> print
> print 'But b - a is %s, which is incorrect.' % (b-a)
> print
> print 'However, if a different tzinfo object is used, the subtraction is done correctly.'
> 
> # This makes a functionally identical, but distinct tzinfo object
> alt_NYC = copy.deepcopy(NYC)
> 
> c = datetime(2008, 3, 9, 3, 0, 0, tzinfo=alt_NYC)
> 
> print 'Time c is %s, which is the same as b above.' % c.isoformat()
> print
> print 'Yet, c - a is %s, which is correct, and different from b - a!' % (c-a)
> 
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
> 
>    -- Chris
> 
> -- 
> 
> Chris Knadle
> Chris.Knadle at coredump.us
> 
> _____________________________________________________________________________
> Hire expert Linux talent by posting jobs here :: http://jobs.nylug.org
> The nylug-talk mailing list is at nylug-talk at nylug.org
> The list archive is at http://nylug.org/pipermail/nylug-talk
> To subscribe or unsubscribe: http://nylug.org/mailman/listinfo/nylug-talk

-- 
The 5 year plan:
In five years we'll make up another plan.
Or just re-use this one.