[BUG] cudf.to_datetime with arg with Z drops UTC offset incorrectly

[mroeschke]

Describe the bug
cudf.to_datetime with arg with Z drops UTC offset incorrectly

Steps/Code to reproduce bug

In [1]: import cudf

In [2]: cudf.to_datetime(["2019-01-01T00:00:00.000Z"])
Out[2]: DatetimeIndex(['2019-01-01'], dtype='datetime64[ns]')

Expected behavior
Since cudf does not support timezones currently, I would expect this to raise a NotImplementedError

It appears there was a commit that tried to address this, 9508210, but it looks to simply ignore "Z"

Environment overview (please complete the following information)

• Environment location: Bare-metal
• Method of cuDF install: conda
  • If method of install is [Docker], provide docker pull & docker run commands used

Environment details
Please run and paste the output of the cudf/print_env.sh script here, to gather any other relevant environment details

Additional context
Add any other context about the problem here.

[cwharris]

It appears there was a commit that tried to address this, 9508210, but it looks to simply ignore "Z"

This is exactly how it should work. Adding Z does not change the timezone. All time data should be assumed to be in UTC unless otherwise specified by a timezone offset. Throwing an exception is not appropriate and is a breaking change that is effecting nv-morpheus/morpheus as we try to update to newer versions of CUDF.

[bdice]

All time data should be assumed to be in UTC unless otherwise specified by a timezone offset.

Pandas differentiates between "naive" datetimes and timezone-aware datetimes. In pandas, parsing a Z will return a datetime64[ns, UTC] type instead of a datetime64[ns] type. cudf needs to align with pandas behavior.

We are temporarily reverting this change of behavior in #14701 when cuDF is used with pandas compatibility mode turned off. See my proposal for a deprecation step followed by a behavior change in cuDF here: #14701 (review)