Confusing datetime.datetime

[Damjan]

I've been struggling with an app that uses 
Postgresql/Psycopg2/SQLAlchemy  and I've come to this confusing 
behaviour of datetime.datetime.


First of all, the "Seconds since Epoch" timestamps are always in UTC, so 
shouldn't change with timezones. So I'd expect that a round trip of a 
timestamp through datetime.datetime, shouldn't change it.


Now, all is good when I use a naive datetime.datetime


-- TZ=UTC python
 >>> from datetime import datetime
 >>> dt = datetime.fromtimestamp(1341446400)
 >>> dt
datetime.datetime(2012, 7, 5, 0, 0)
 >>> dt.strftime('%s')
'1341446400'


-- TZ=Asia/Tokyo python
 >>> from datetime import datetime
 >>> dt = datetime.fromtimestamp(1341446400)
 >>> dt
datetime.datetime(2012, 7, 5, 9, 0)
 >>> dt.strftime('%s')
'1341446400'



But when I use an timezone aware datetime.datetime objects, the 
timestamp roundtrip is destroyed. I get 2 different timestamps.
Am I missing something here, I've been reading the datetime 
documentation several times, but I can't understand what is the intended 
behaviour.


-- TZ=UTC python
 >>> from datetime import datetime
 >>> import pytz
 >>> tz = pytz.timezone('Europe/Skopje')
 >>> dt = datetime.fromtimestamp(1341446400, tz)
 >>> dt
datetime.datetime(2012, 7, 5, 2, 0, tzinfo=<DstTzInfo 'Europe/Skopje' 
CEST+2:00:00 DST>)
 >>> dt.strftime('%s')
'1341453600'


-- TZ=Asia/Tokyo python
 >>> from datetime import datetime
 >>> import pytz
 >>> tz = pytz.timezone('Europe/Skopje')
 >>> dt = datetime.fromtimestamp(1341446400, tz)
 >>> dt
datetime.datetime(2012, 7, 5, 2, 0, tzinfo=<DstTzInfo 'Europe/Skopje' 
CEST+2:00:00 DST>)
 >>> dt.strftime('%s')
'1341421200'



Python 2.7.3, pytz 2012c

-- 
damjan