API: Default to stdlib timezone objects instead of pytz

[jbrockmendel]

i.e. when we get pd.Timestamp.now("UTC") we should return a Timestamp with datetime.timezone.utc rather than pytz.UTC. Similarly when we parse an ISO8601 datetime we should use tzinfo of timezone(timedelta(seconds=val*60)) instead of pytz.FixedOffset(val)

This isn't that hard to implement, but doing it breaks a couple dozen tests where we are currently checking for pytz objects. This would technically be an API change, so putting it up for discussion before implementing it.

[jreback]

i don’t think we easily change this as it’s a failrly large api change

we would need to have an option for this instead

[mroeschke]

Might be a reasonable version 2.0 break. But +1 to moving the default timezone to the stdlib version

[jorisvandenbossche]

Since there is also the discussion of using zoneinfo (in Python 3.9 stdblib, backport https://github.com/pganssle/zoneinfo), which is also a breaking change, those should potentially be considered together?

(zoneinfo actually also has a UTC timezone)

[mroeschke]

+1 on incorporating zoneinfo as well. timezone object from the current stdlib otherwise are pretty bare.

[jbrockmendel]

Thoughts on doing this without deprecation in 2.0?

[mroeschke]

Thoughts on doing this without deprecation in 2.0?

+1. Might be even better if we make ZoneInfo the default timezone object though if 2.0 still supports < 3.9 backports.zoneinfo would need to be taken as a dependency I think.