ENH: avoid creating reference cycle on indexing (#15746)

Author: Antoine Pitrou

pandas/_libs/lib.pyx

@@ -1839,5 +1839,27 @@ cdef class BlockPlacement:
         return self._as_slice
 
 
+cdef class _NDFrameIndexerBase:
+    '''
+    A base class for _NDFrameIndexer for fast instantiation and attribute
+    access.
+    '''
+    cdef public object obj, name, _ndim
+
+    def __init__(self, name, obj):
+        self.obj = obj
+        self.name = name
+        self._ndim = None
+
+    @property
+    def ndim(self):
+        # Delay `ndim` instantiation until required as reading it
+        # from `obj` isn't entirely cheap.
+        ndim = self._ndim
+        if ndim is None:
+            ndim = self._ndim = self.obj.ndim
+        return ndim
+
+
 include "reduce.pyx"
 include "inference.pyx"

pandas/core/generic.py

@@ -1841,23 +1841,12 @@ def to_latex(self, buf=None, columns=None, col_space=None, header=True,
     @classmethod
     def _create_indexer(cls, name, indexer):
         """Create an indexer like _name in the class."""
+        from functools import partial
 
         if getattr(cls, name, None) is None:
-            iname = '_%s' % name
-            setattr(cls, iname, None)
-
-            def _indexer(self):
-                i = getattr(self, iname)
-                if i is None:
-                    i = indexer(self, name)
-                    setattr(self, iname, i)
-                return i
-
+            _indexer = partial(indexer, name)
             setattr(cls, name, property(_indexer, doc=indexer.__doc__))
 
-            # add to our internal names set
-            cls._internal_names_set.add(iname)
-
     def get(self, key, default=None):
         """
         Get item from object for given key (DataFrame column, Panel slice,

pandas/core/indexing.py

@@ -23,6 +23,7 @@
 from pandas.core.common import (is_bool_indexer, _asarray_tuplesafe,
                                 is_null_slice, is_full_slice,
                                 _values_from_object)
+from pandas._libs.lib import _NDFrameIndexerBase
 
 
 # the supported indexers
@@ -85,19 +86,14 @@ class IndexingError(Exception):
     pass
 
 
-class _NDFrameIndexer(object):
+class _NDFrameIndexer(_NDFrameIndexerBase):
     _valid_types = None
     _exception = KeyError
     axis = None
 
-    def __init__(self, obj, name):
-        self.obj = obj
-        self.ndim = obj.ndim
-        self.name = name
-
     def __call__(self, axis=None):
         # we need to return a copy of ourselves
-        new_self = self.__class__(self.obj, self.name)
+        new_self = self.__class__(self.name, self.obj)
 
         if axis is not None:
             axis = self.obj._get_axis_number(axis)
@@ -1321,7 +1317,7 @@ class _IXIndexer(_NDFrameIndexer):
 
     """
 
-    def __init__(self, obj, name):
+    def __init__(self, name, obj):
 
         _ix_deprecation_warning = textwrap.dedent("""
             .ix is deprecated. Please use
@@ -1333,7 +1329,7 @@ def __init__(self, obj, name):
 
         warnings.warn(_ix_deprecation_warning,
                       DeprecationWarning, stacklevel=3)
-        super(_IXIndexer, self).__init__(obj, name)
+        super(_IXIndexer, self).__init__(name, obj)
 
     def _has_valid_type(self, key, axis):
         if isinstance(key, slice):

pandas/tests/indexing/test_indexing.py

@@ -5,6 +5,7 @@
 
 import pytest
 
+import weakref
 from warnings import catch_warnings
 from datetime import datetime
 
@@ -881,6 +882,14 @@ def test_partial_boolean_frame_indexing(self):
                                 columns=list('ABC'))
         tm.assert_frame_equal(result, expected)
 
+    def test_no_reference_cycle(self):
+        df = pd.DataFrame({'a': [0, 1], 'b': [2, 3]})
+        for name in ('loc', 'iloc', 'ix', 'at', 'iat'):
+            getattr(df, name)
+        wr = weakref.ref(df)
+        del df
+        assert wr() is None
+
 
 class TestSeriesNoneCoercion(object):
     EXPECTED_RESULTS = [