test_forecast failures with pandas 1.0.0

[cwhanse]

Pandas has released v1.0.0. With pandas 1.0.0, test_forecast.test_datetime fails, as #872 shows in the test log. It is not clear if the test itself is the problem: these lines do not set a timezone.

    start = datetime.now()
    end = start + timedelta(days=1)

Should start and end be timezone-aware? The intent of the test is not clear to me.

To Reproduce
Steps to reproduce the behavior:

1. Install pandas 1.0.0

import pandas as pd
from datetime import datetime, timedelta
from pvlib.forecast import NAM

_latitude = 32.2
_longitude = -110.9
_tz = 'US/Arizona'
_start = pd.Timestamp.now(tz=_tz)
_end = _start + pd.Timedelta(days=1)

amodel = NAM()
start = datetime.now()
end = start + timedelta(days=1)
try:
    amodel.get_processed_data(_latitude, _longitude, start, end)
    print("Success with {}".format(pd.__version__))
except:
    print("Failed with {}".format(pd.__version__))

Now run with pandas < 1.0.0 (e.g., 0.22.0)

[wholmgren]

The test ensures that the function works if you use a python datetime object instead of pandas datetime object. We should check that start/end are localized and, if not, raise TypeError before issuing the query.

[cwhanse]

Want to correct this in conjunction with #878? Or I can do it. The test failures will repeat for every PR unless we pin pandas <= 1.0.0.

[wholmgren]

Yes, I should have said I'll provide a fix as soon as I get out of conda hell. Apparently I should not have run pip install pandas -U in an environment with conda-provided numpy/netcdf/hdf5. Yay python packaging.