DEPR/TEST: fix numpy tz-aware parsing warnings

[jreback]

/home/travis/build/pydata/pandas/pandas/tseries/tests/test_period.py:420: DeprecationWarning: parsing timezone aware datetimes is deprecated; this will raise an error in the future

full build here
note this is only on the numpy-dev builds (last 2)

@shoyer recent change to show a deprecation warning on passing tz-aware strings to numpy here is now causing loads of tests to show these.

so in order to keep backward compat I suppose we need a wrapper function

In [1]: np.__version__
Out[1]: '1.12.0.dev0+a38942a'

In [2]: np.datetime64('2007-01-01 09:00:00.001Z')
/Users/jreback/miniconda/envs/numpy_dev/bin/ipython:1: DeprecationWarning: parsing timezone aware datetimes is deprecated; this will raise an error in the future
  #!/bin/bash /Users/jreback/miniconda/envs/numpy_dev/bin/python.app
Out[2]: numpy.datetime64('2007-01-01T09:00:00.001')

In [3]: np.datetime64('2007-01-01 09:00:00.001')
Out[3]: numpy.datetime64('2007-01-01T09:00:00.001')

In [4]: np.datetime64('2007-01-01 09:00:00.001').item()
Out[4]: datetime.datetime(2007, 1, 1, 9, 0, 0, 1000)

current numpy

In [2]: np.__version__
Out[2]: '1.10.2'

In [3]: np.datetime64('2007-01-01 09:00:00.001Z')
Out[3]: numpy.datetime64('2007-01-01T04:00:00.001-0500')

In [5]: np.datetime64('2007-01-01 09:00:00.001Z').item()
Out[5]: datetime.datetime(2007, 1, 1, 9, 0, 0, 1000)

so maybe

np_datetime64_compat(....) that can then introspect our defined _np_version_under1p11 (define in pandas/__init__.py) and add a 'Z' if needed (pre-suppose that we change the input formats to
2007-01-01 09:00:00.001') for clarity.

[shoyer]

Hmm. I wonder if we should adjust numpy to not issue a deprecation warning when the "Z" specifier is provided? Folks will be using that for quite a while for backwards compat, and it's actually 100% forwards compat in the typical case where you're time zone naive as UTC only...

[jreback]

it IS technically wrong I think though. a 'Z' implies by definition a tz-aware date (that is UTC). Since numpy doesn't support aware, then I guess it is clear what it is.

But you are right, it is no harm. OTOH, if you don't issue a DeprecationWarning then no-one will ever fix it.