Fix datetime.timedelta casting bug in coding.times.infer_datetime_units (#2128)

* Fix #2127

* Fix typo in time-series.rst

* Use pd.to_timedelta to convert to np.timedelta64 objects

* Install cftime through netcdf4 through pip

* box=False

Author: spencerkclark

ci/requirements-py27-windows.yml

@@ -8,7 +8,6 @@ dependencies:
   - h5py
   - h5netcdf
   - matplotlib
-  - netcdf4
   - pathlib2
   - pytest
   - flake8
@@ -21,4 +20,4 @@ dependencies:
   - rasterio
   - zarr
   - pip:
-    - cftime
+    - netcdf4

doc/time-series.rst

@@ -73,7 +73,7 @@ native representation of dates to those that fall between the years 1678 and
 returned as arrays of ``cftime.datetime`` objects and a ``CFTimeIndex``
 can be used for indexing.  The ``CFTimeIndex`` enables only a subset of
 the indexing functionality of a ``pandas.DatetimeIndex`` and is only enabled
-when using standalone version of ``cftime`` (not the version packaged with
+when using the standalone version of ``cftime`` (not the version packaged with
 earlier versions ``netCDF4``).  See :ref:`CFTimeIndex` for more information.
 
 Datetime indexing

xarray/coding/times.py

@@ -253,7 +253,11 @@ def infer_datetime_units(dates):
     else:
         reference_date = dates[0] if len(dates) > 0 else '1970-01-01'
         reference_date = format_cftime_datetime(reference_date)
-    unique_timedeltas = np.unique(np.diff(dates)).astype('timedelta64[ns]')
+    unique_timedeltas = np.unique(np.diff(dates))
+    if unique_timedeltas.dtype == np.dtype('O'):
+        # Convert to np.timedelta64 objects using pandas to work around a
+        # NumPy casting bug: https://github.com/numpy/numpy/issues/11096
+        unique_timedeltas = pd.to_timedelta(unique_timedeltas, box=False)
     units = _infer_time_units_from_diff(unique_timedeltas)
     return '%s since %s' % (units, reference_date)
 

xarray/tests/test_coding_times.py

@@ -587,6 +587,9 @@ def test_infer_cftime_datetime_units():
                  'seconds since 1900-01-01 00:00:00.000000'),
                 ([date_type(1900, 1, 1),
                   date_type(1900, 1, 2, 0, 0, 0, 5)],
+                 'days since 1900-01-01 00:00:00.000000'),
+                ([date_type(1900, 1, 1), date_type(1900, 1, 8),
+                  date_type(1900, 1, 16)],
                  'days since 1900-01-01 00:00:00.000000')]:
             assert expected == coding.times.infer_datetime_units(dates)