FIX sql handling of timedelta64 columns (GH6921)

Author: jorisvandenbossche

doc/source/io.rst

@@ -809,8 +809,6 @@ two extra parameters:
     String value 'infer' can be used to instruct the parser to try detecting
     the column specifications from the first 100 rows of the data. Default
     behaviour, if not specified, is to infer.
-    As with regular python slices, you can slice to the end of the line
-    with ``None``, e.g. ``colspecs = [(0, 1), (1, None)]``.
   - ``widths``: A list of field widths which can be used instead of 'colspecs'
     if the intervals are contiguous.
 
@@ -3235,6 +3233,13 @@ the database using :func:`~pandas.DataFrame.to_sql`.
 
     data.to_sql('data', engine)
 
+.. note::
+
+    Due to the limited support for timedelta's in the different database
+    flavors, columns with type ``timedelta64`` will be written as integer
+    values as nanoseconds to the database and a warning will be raised.
+
+
 Reading Tables
 ~~~~~~~~~~~~~~
 

pandas/io/sql.py

@@ -730,7 +730,10 @@ def _sqlalchemy_type(self, arr_or_dtype):
             except:
                 return DateTime
         if com.is_timedelta64_dtype(arr_or_dtype):
-            return Interval
+            warnings.warn("the 'timedelta' type is not supported, and will be "
+                          "written as integer values (ns frequency) to the "
+                          "database.", UserWarning)
+            return Integer
         elif com.is_float_dtype(arr_or_dtype):
             return Float
         elif com.is_integer_dtype(arr_or_dtype):
@@ -973,6 +976,11 @@ def _sql_type_name(self, dtype):
         pytype_name = "text"
         if issubclass(pytype, np.floating):
             pytype_name = "float"
+        elif com.is_timedelta64_dtype(pytype):
+            warnings.warn("the 'timedelta' type is not supported, and will be "
+                          "written as integer values (ns frequency) to the "
+                          "database.", UserWarning)
+            pytype_name = "int"
         elif issubclass(pytype, np.integer):
             pytype_name = "int"
         elif issubclass(pytype, np.datetime64) or pytype is datetime:

pandas/io/tests/test_sql.py

@@ -29,12 +29,14 @@
 from datetime import datetime
 
 from pandas import DataFrame, Series, Index, MultiIndex, isnull
+from pandas import to_timedelta
 import pandas.compat as compat
 from pandas.compat import StringIO, range, lrange
 from pandas.core.datetools import format as date_format
 
 import pandas.io.sql as sql
 import pandas.util.testing as tm
+from pandas import _np_version_under1p7
 
 
 try:
@@ -480,6 +482,17 @@ def test_date_and_index(self):
         self.assertTrue(issubclass(df.IntDateCol.dtype.type, np.datetime64),
                         "IntDateCol loaded with incorrect type")
 
+    def test_timedelta(self):
+        # see #6921
+        if _np_version_under1p7:
+            raise nose.SkipTest("test only valid in numpy >= 1.7")
+
+        df = to_timedelta(Series(['00:00:01', '00:00:03'], name='foo')).to_frame()
+        with tm.assert_produces_warning(UserWarning):
+            df.to_sql('test_timedelta', self.conn)
+        result = sql.read_sql_query('SELECT * FROM test_timedelta', self.conn)
+        tm.assert_series_equal(result['foo'], df['foo'].astype('int64'))              
+        
     def test_to_sql_index_label(self):
         temp_frame = DataFrame({'col1': range(4)})