Bug in dateconversion?

[jbusecke]

I noticed an undesired behavior in xarray when using xarray.open_dataset:

running the following in version 0.8.2-90-g2c7730d:

import xarray as xr
fid = 'dt_global_allsat_msla_uv_20140101_20140829.nc'
ds = xr.open_dataset(fid)
ds.time

gives

<xarray.DataArray 'time' (time: 1)>
array(['2013-12-31T19:00:00.000000000-0500'], dtype='datetime64[ns]')
Coordinates:
  * time     (time) datetime64[ns] 2014-01-01
Attributes:
    long_name: Time
    standard_name: time
    axis: T

Note the hour is 19, I also encountered files with 20. Since the time in the .nc file is given in 'days since', the expected output would be 00.

Indeed when running version 0.8.2 the output is:

<xarray.DataArray 'time' (time: 1)>
array(['2014-01-01T00:00:00.000000000'], dtype='datetime64[ns]')
Coordinates:
  * time     (time) datetime64[ns] 2014-01-01
Attributes:
    long_name: Time
    standard_name: time
    axis: T

Any idea what could cause this?

Sample file used in the example can be found here: https://ufile.io/077da

[fmaussion]

I can't reproduce this. Which numpy version do you have installed? (see https://docs.scipy.org/doc/numpy-dev/reference/arrays.datetime.html#changes-with-numpy-1-11)

[shoyer]

I'm pretty sure this is indeed a timezone display issue fixed by NumPy 1.11.

[jhamman]

Agreed: https://github.com/pydata/xarray/issues/439

[jbusecke]

Fantastic! Thanks very much.