message.header.date should specify a timezone about jupyter_client

[Carreau]

Hum, the alternative is to use datetime.utcnow() I guess.

But good point.

[jcb91]

Actually, datetime.utcnow() still doesn't return a timezone-aware datetime object (I've no idea why, really, UTC seems simple enough that it could be part of standard library, but hey ho). To remove dependency on pytz, something like this SO answer, which defines a tzinfo class for UTC, might be a better solution...

[jcb91]

see, for reference, the difference in the two print statements from this snippet:

from datetime import datetime, timedelta, tzinfo

TDZERO = timedelta(0)

class UTC(tzinfo):
    """A UTC tzinfo class"""

    def utcoffset(self, dt):
        return TDZERO

    def tzname(self, dt):
        return 'UTC'

    def dst(self, dt):
        return TDZERO

utc = UTC()

print('utcnow', datetime.utcnow().isoformat())
print('tz=utc', datetime.now(tz=utc).isoformat())

gives

utcnow 2016-03-21T19:03:04.625622
tz=utc 2016-03-21T19:03:04.625970+00:00

[Carreau]

Actually, datetime.utcnow() still doesn't return a timezone-aware datetime object (I've no idea why, really, UTC seems simple enough that it could be part of standard library, but hey ho)

Sorry I was unclear in my statement, we could normalize the spec saying that the date time should be UTC if the current spec is not precise enough.
Of course for backward compatibility we should encourage people to have timezone, but I think in the long term, having UTC everywhere (execpt on user facing UI) make sens.

[jcb91]

Right, gotcha. Still, I think it's worth adding the timezone to the string values as well, rather than just altering the spec to allow people to assume they're in UTC - after all Explicit is better than implicit 😉

[takluyver]

We already have three copies of the machinery to produce tz-aware UTC timestamps - there's one in jupyter_client, but it's currently in the tests (test_jsonutil to be precise). I'm pretty sure that UTC timestamps are the way to go, not localised timezone-aware timestamps.

[jcb91]

Oh absolutely, utc is the way forward, I just meant the offset (of zero, naturally) should get included in the json-ified text version, as with the utcnow in test_jsonutil.

[Carreau]

Oh absolutely, utc is the way forward, I just meant the offset (of zero, naturally) should get included in the json-ified text version, as with the utcnow in test_jsonutil.

Do I smell a Pull Request coming ?

[jcb91]

Haha 😆 I'm happy to submit one, I was a little apprehensive of adding yet another mechanism to produce tz-aware timestamps! If you guys know where you'd like to put it, I'll be happy to fix it up...

[minrk]

We can have nice tz-aware utc timestamps with the stdlib in Python 3 with:

datetime.datetime.now(datetime.timezone.utc)

Python 2, of course, needs an backport of the trivial UTC timezone, which we have implemented here (and elsewhere). We can use utc for that, I suppose.

[minrk]

master (will be 5.0) uses timezone-aware datetimes.