BUG: Bug in DataFrame column insertion when the column creation fails (GH3010)

RELEASE.rst

@@ -141,6 +141,8 @@ pandas 0.11.0
   - Fixed printing of ``NaT` in an index
   - Bug in idxmin/idxmax of ``datetime64[ns]`` Series with ``NaT`` (GH2982__)
   - Bug in ``icol`` with negative indicies was incorrect producing incorrect return values (see GH2922_)
+  - Bug in DataFrame column insertion when the column creation fails, existing frame is left in
+    an irrecoverable state (GH3010_)
 
 .. _GH622: https://github.com/pydata/pandas/issues/622
 .. _GH797: https://github.com/pydata/pandas/issues/797
@@ -166,6 +168,7 @@ pandas 0.11.0
 .. _GH2982: https://github.com/pydata/pandas/issues/2982
 .. _GH2989: https://github.com/pydata/pandas/issues/2989
 .. _GH3002: https://github.com/pydata/pandas/issues/3002
+.. _GH3010: https://github.com/pydata/pandas/issues/3010
 .. _GH3012: https://github.com/pydata/pandas/issues/3012
 
 

doc/source/io.rst

@@ -1138,11 +1138,14 @@ defaults to `nan`.
 
 .. ipython:: python
 
-    df_mixed             = df.copy()
-    df_mixed['string']   = 'string'
-    df_mixed['int']      = 1
-    df_mixed['bool']     = True
-    df_mixed['datetime64'] = Timestamp('20010102')
+    df_mixed = DataFrame({ 'A' : randn(8),
+                           'B' : randn(8),
+                           'C' : np.array(randn(8),dtype='float32'),
+                           'string' :'string',
+                           'int' : 1,
+                           'bool' : True,
+                           'datetime64' : Timestamp('20010102')},
+                         index=range(8))
     df_mixed.ix[3:5,['A', 'B', 'string', 'datetime64']] = np.nan
 
     store.append('df_mixed', df_mixed, min_itemsize = {'values': 50})
@@ -1445,8 +1448,7 @@ may not be installed (by Python) by default.
 
 Compression for all objects within the file
 
-   - ``store_compressed = HDFStore('store_compressed.h5', complevel=9,
-                                    complib='blosc')``
+   - ``store_compressed = HDFStore('store_compressed.h5', complevel=9, complib='blosc')``
 
 Or on-the-fly compression (this only applies to tables). You can turn
 off file compression for a specific table by passing ``complevel=0``

pandas/core/internals.py

@@ -1334,11 +1334,22 @@ def insert(self, loc, item, value):
         if item in self.items:
             raise Exception('cannot insert %s, already exists' % item)
 
-        new_items = self.items.insert(loc, item)
-        self.set_items_norename(new_items)
+        try:
+            new_items = self.items.insert(loc, item)
+            self.set_items_norename(new_items)
+
+            # new block
+            self._add_new_block(item, value, loc=loc)
 
-        # new block
-        self._add_new_block(item, value, loc=loc)
+        except:
+
+            # so our insertion operation failed, so back out of the new items
+            # GH 3010
+            new_items = self.items.delete(loc)
+            self.set_items_norename(new_items)
+
+            # re-raise
+            raise
 
         if len(self.blocks) > 100:
             self._consolidate_inplace()

pandas/tests/test_frame.py

@@ -1968,6 +1968,19 @@ def test_constructor_cast_failure(self):
         foo = DataFrame({'a': ['a', 'b', 'c']}, dtype=np.float64)
         self.assert_(foo['a'].dtype == object)
 
+        # GH 3010, constructing with odd arrays
+        df = DataFrame(np.ones((4,2)))
+
+        # this is ok
+        df['foo'] = np.ones((4,2)).tolist()
+
+        # this is not ok
+        self.assertRaises(AssertionError, df.__setitem__, tuple(['test']), np.ones((4,2)))
+
+        # this is ok
+        df['foo2'] = np.ones((4,2)).tolist()
+
+
     def test_constructor_dtype_nocast_view(self):
         df = DataFrame([[1, 2]])
         should_be_view = DataFrame(df, dtype=df[0].dtype)