NonExistentTimeError after Resample

[jackalack]

Code Sample

import pandas as pd
import numpy as np
import pytz 
index=pd.date_range('2017-03-12', '2017-03-12 1:45:00', freq='15T')
my_series = pd.Series(np.zeros(len(index)), index=index)
my_series = my_series.tz_localize('US/Pacific')
my_series.resample('900S').mean()

Some Version Information:

Python: 3.6.2
pandas: 0.22.0
pytz: 2017.3
numpy: 1.13.0

Problem description

The above code returns the error message NonExistentTimeError: 2017-03-12 02:00:00; however, the input series only contains timestamps up to 2017-03-12 01:45:00. Does anyone understand why this error is occurring? If so, how can I address the problem?

[gfyoung]

@jackalack : AFAICT, this error is occurring because it's a bug 😄 PR to patch is welcome!

[markjszy]

Had a quick look into this and can reproduce consistently. Exception is thrown by pandas/_libs/tslibs/conversion.pyx#tz_localize_to_utc

I isolated the values from the example above, and found some others that also result in an exception like this. Time zone and last_dst values don't seem to make a difference.

from pandas._libs.lib import Timestamp

fresult = 1489276800000000000
lresult = 1489284000000000000
last_tzinfo = 'US/Pacific'
last_dst = False

# this is fine
Timestamp(fresult).tz_localize(last_tzinfo, ambiguous=last_dst)

# this is not
Timestamp(lresult).tz_localize(last_tzinfo, ambiguous=last_dst)

# other problematic examples
Timestamp(1489285000000000000).tz_localize(last_tzinfo, ambiguous=last_dst)
Timestamp(1489286000000000000).tz_localize(last_tzinfo, ambiguous=last_dst)