BUG: fix mutation of DTI backing Series/DataFrame

pandas/core/internals/blocks.py

@@ -2923,7 +2923,9 @@ def _try_coerce_result(self, result):
             # allow passing of > 1dim if its trivial
             if result.ndim > 1:
                 result = result.reshape(np.prod(result.shape))
-            result = self.values._shallow_copy(result)
+
+            # GH#24096 new values invalidates a frequency
+            result = self.values._shallow_copy(result, freq=None)
 
         return result
 

pandas/core/internals/construction.py

@@ -196,6 +196,14 @@ def init_dict(data, index, columns, dtype=None):
             arrays.loc[missing] = [v] * missing.sum()
 
     else:
+
+        for key in data:
+            if (isinstance(data[key], ABCDatetimeIndex) and
+                    data[key].tz is not None):
+                # GH#24096 need copy to be deep for datetime64tz case
+                # TODO: See if we can avoid these copies
+                data[key] = data[key].copy(deep=True)
+
         keys = com.dict_keys_to_ordered_list(data)
         columns = data_names = Index(keys)
         arrays = [data[k] for k in keys]

pandas/core/series.py

@@ -21,7 +21,7 @@
     is_extension_array_dtype, is_extension_type, is_hashable, is_integer,
     is_iterator, is_list_like, is_scalar, is_string_like, is_timedelta64_dtype)
 from pandas.core.dtypes.generic import (
-    ABCDataFrame, ABCSeries, ABCSparseArray, ABCSparseSeries)
+    ABCDataFrame, ABCDatetimeIndex, ABCSeries, ABCSparseArray, ABCSparseSeries)
 from pandas.core.dtypes.missing import (
     isna, na_value_for_dtype, notna, remove_na_arraylike)
 
@@ -182,6 +182,11 @@ def __init__(self, data=None, index=None, dtype=None, name=None,
                 else:
                     # need to copy to avoid aliasing issues
                     data = data._values.copy()
+                    if (isinstance(data, ABCDatetimeIndex) and
+                            data.tz is not None):
+                        # GH#24096 need copy to be deep for datetime64tz case
+                        # TODO: See if we can avoid these copies
+                        data = data._values.copy(deep=True)
                 copy = False
 
             elif isinstance(data, np.ndarray):

pandas/tests/frame/test_block_internals.py

@@ -28,6 +28,22 @@
 
 
 class TestDataFrameBlockInternals():
+    def test_setitem_invalidates_datetime_index_freq(self):
+        # GH#24096 altering a datetime64tz column inplace invalidates the
+        #  `freq` attribute on the underlying DatetimeIndex
+
+        dti = date_range('20130101', periods=3, tz='US/Eastern')
+        ts = dti[1]
+
+        df = DataFrame({'B': dti})
+        assert df['B']._values.freq == 'D'
+
+        df.iloc[1, 0] = pd.NaT
+        assert df['B']._values.freq is None
+
+        # check that the DatetimeIndex was not altered in place
+        assert dti.freq == 'D'
+        assert dti[1] == ts
 
     def test_cast_internals(self, float_frame):
         casted = DataFrame(float_frame._data, dtype=int)

pandas/tests/series/test_block_internals.py

@@ -0,0 +1,42 @@
+# -*- coding: utf-8 -*-
+
+import pandas as pd
+
+# Segregated collection of methods that require the BlockManager internal data
+# structure
+
+
+class TestSeriesBlockInternals(object):
+
+    def test_setitem_invalidates_datetime_index_freq(self):
+        # GH#24096 altering a datetime64tz Series inplace invalidates the
+        #  `freq` attribute on the underlying DatetimeIndex
+
+        dti = pd.date_range('20130101', periods=3, tz='US/Eastern')
+        ts = dti[1]
+        ser = pd.Series(dti)
+        assert ser._values is not dti
+        assert ser._values._data.base is not dti._data.base
+        assert dti.freq == 'D'
+        ser.iloc[1] = pd.NaT
+        assert ser._values.freq is None
+
+        # check that the DatetimeIndex was not altered in place
+        assert ser._values is not dti
+        assert ser._values._data.base is not dti._data.base
+        assert dti[1] == ts
+        assert dti.freq == 'D'
+
+    def test_dt64tz_setitem_does_not_mutate_dti(self):
+        # GH#21907, GH#24096
+        dti = pd.date_range('2016-01-01', periods=10, tz='US/Pacific')
+        ts = dti[0]
+        ser = pd.Series(dti)
+        assert ser._values is not dti
+        assert ser._values._data.base is not dti._data.base
+        assert ser._data.blocks[0].values is not dti
+        assert ser._data.blocks[0].values._data.base is not dti._data.base
+
+        ser[::3] = pd.NaT
+        assert ser[0] is pd.NaT
+        assert dti[0] == ts