pandas: BUG: read_csv returns object dtype for dates in empty frame

Code Sample, a copy-pastable example if possible

>>> import io
>>> import pandas as pd

>>> csv = lambda: io.StringIO('date')
>>> df = pd.read_csv(csv(), dtype=int)
>>> df.dtypes

date    int64
dtype: object

>>> df = pd.read_csv(csv(), parse_dates=['date'])
>>> df.dtypes

date      object
dtype: object

>>> df.date.dt.tz_localize('UTC')

Traceback (most recent call last):
  File "<ipython-input-49-067c7ce27232>", line 1, in <module>
    df.date = df.date.dt.tz_localize('UTC')
  File "~/anaconda/lib/python3.5/site-packages/pandas/core/generic.py", line 2740, in __getattr__
    return object.__getattribute__(self, name)
  File "~/anaconda/lib/python3.5/site-packages/pandas/core/base.py", line 241, in __get__
    return self.construct_accessor(instance)
  File "~/anaconda/lib/python3.5/site-packages/pandas/core/series.py", line 2743, in _make_dt_accessor
    raise AttributeError("Can only use .dt accessor with datetimelike "
AttributeError: Can only use .dt accessor with datetimelike values

>>> parser = lambda x: pd.to_datetime(x, utc=True)
>>> df = pd.read_csv(csv(), parse_dates=['date'], date_parser=parser)
>>> df.dtypes

date      object
dtype: object

>>> df.date.dt.tz_convert('EST')

Traceback (most recent call last):
  File "<ipython-input-129-3d458af8513c>", line 1, in <module>
    df.date.dt.tz_convert('EST')
  File "~/anaconda/lib/python3.5/site-packages/pandas/core/generic.py", line 2740, in __getattr__
    return object.__getattribute__(self, name)
  File "~/anaconda/lib/python3.5/site-packages/pandas/core/base.py", line 241, in __get__
    return self.construct_accessor(instance)
  File "~/anaconda/lib/python3.5/site-packages/pandas/core/series.py", line 2743, in _make_dt_accessor
    raise AttributeError("Can only use .dt accessor with datetimelike "
AttributeError: Can only use .dt accessor with datetimelike values

Problem description

When reading CSVs with no data rows, read_csv() returns the dtype object for dates, which can raise errors on later manipulation. This is contrary to the general behaviour of read_csv(), which otherwise correctly sets dtypes for empty frames when those dtypes are explicitly passed.

I don’t think it would be hard to return the correct dtype here? If date_parser is not set, we know the dtype is datetime64[ns]; otherwise, we can call the parser with empty data, and use the returned dtype.

Note that e.g. read_csv(..., dtype='datetime64[ns]') is not a solution, as this throws an error when the csv is non-empty.

Expected Output

date    int64
dtype: object

date      datetime64[ns]
dtype: object

date      datetime64[ns, UTC]
dtype: object

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None python: 3.5.2.final.0 python-bits: 64 OS: Darwin OS-release: 16.4.0 machine: x86_64 processor: i386 byteorder: little LC_ALL: None LANG: None LOCALE: None.None

pandas: 0.19.2 nose: 1.3.7 pip: 9.0.1 setuptools: 27.2.0 Cython: 0.25.2 numpy: 1.11.3 scipy: 0.18.1 statsmodels: 0.6.1 xarray: 0.9.1 IPython: 4.2.0 sphinx: 1.5.1 patsy: 0.4.1 dateutil: 2.6.0 pytz: 2016.10 blosc: None bottleneck: 1.2.0 tables: 3.3.0 numexpr: 2.6.1 matplotlib: 1.5.3 openpyxl: 2.4.0 xlrd: 1.0.0 xlwt: 1.1.2 xlsxwriter: 0.9.6 lxml: 3.7.2 bs4: 4.5.3 html5lib: None httplib2: None apiclient: None sqlalchemy: 1.1.4 pymysql: None psycopg2: None jinja2: 2.8.1 boto: 2.45.0 pandas_datareader: None