From 6d7686a6a2e57da31d08beebb2594119aeeb36cf Mon Sep 17 00:00:00 2001
From: immerrr <immerrr@gmail.com>
Date: Mon, 28 Apr 2014 13:11:28 +0400
Subject: [PATCH] CLN: drop internals._invert_reordering in favour of
 lib.get_reverse_indexer

---
 doc/source/release.rst   |  2 ++
 doc/source/v0.14.0.txt   |  6 ++++
 pandas/core/internals.py | 63 +---------------------------------------
 pandas/lib.pyx           | 11 +++++++
 4 files changed, 20 insertions(+), 62 deletions(-)

diff --git a/doc/source/release.rst b/doc/source/release.rst
index 38e95eaba0b0f..91e523cfc882c 100644
--- a/doc/source/release.rst
+++ b/doc/source/release.rst
@@ -295,6 +295,8 @@ Improvements to existing features
 - ``read_excel`` can now read milliseconds in Excel dates and times with xlrd >= 0.9.3. (:issue:`5945`)
 - ``pivot_table`` can now accept ``Grouper`` by ``index`` and ``columns`` keywords (:issue:`6913`)
 - Improved performance of compatible pickles (:issue:`6899`)
+- Refactor Block classes removing `Block.items` attributes to avoid duplication
+  in item handling (:issue:`6745`, :issue:`6988`).
 
 .. _release.bug_fixes-0.14.0:
 
diff --git a/doc/source/v0.14.0.txt b/doc/source/v0.14.0.txt
index 43096b133f26e..2c29804cbafa0 100644
--- a/doc/source/v0.14.0.txt
+++ b/doc/source/v0.14.0.txt
@@ -27,6 +27,12 @@ users upgrade to this version.
 
 - :ref:`Bug Fixes <release.bug_fixes-0.14.0>`
 
+.. warning::
+
+   In 0.14.0 all ``NDFrame`` based containers have underwent significant internal refactoring.  Before that each block of
+   homogeneous data had its own labels and extra care was necessary to keep those in sync with parent container's labels.
+   As stated, the refactoring is internal and no publicly visible changes should happen.
+
 .. _whatsnew_0140.api:
 
 API changes
diff --git a/pandas/core/internals.py b/pandas/core/internals.py
index 887f7562421d7..48d047baaa6c0 100644
--- a/pandas/core/internals.py
+++ b/pandas/core/internals.py
@@ -2316,7 +2316,7 @@ def combine(self, blocks, copy=True):
 
         # FIXME: optimization potential
         indexer = np.sort(np.concatenate([b.mgr_locs.as_array for b in blocks]))
-        inv_indexer = _invert_reordering(indexer)
+        inv_indexer = lib.get_reverse_indexer(indexer, self.shape[0])
         new_items = self.items.take(indexer)
 
         new_blocks = []
@@ -3506,71 +3506,10 @@ def _possibly_compare(a, b, op):
     return res
 
 
-
-
 def _concat_indexes(indexes):
     return indexes[0].append(indexes[1:])
 
 
-def _invert_reordering(reordering, minlength=None):
-    """
-    Invert reordering operation.
-
-    Given array `reordering`, make `reordering_inv` of it, such that::
-
-        reordering_inv[reordering[x]] = x
-
-    There are two types of indexers:
-
-    source
-        is when element *s* at position *i* means that values to fill *i-th*
-        item of reindex operation should be taken from *s-th* item of the
-        original (this is what is returned by `pandas.Index.reindex`).
-    destination
-        is when element *d* at position *i* means that values from *i-th* item
-        of source should be used to fill *d-th* item of reindexing operation.
-
-    This function will convert from *source* to *destination* and vice-versa.
-
-    .. note:: trailing ``-1`` may be lost upon conversion (this is what
-              `minlength` is there for).
-
-    .. note:: if *source* indexer is not unique, corresponding *destination*
-              indexer will have ``dtype=object`` and will contain lists.
-
-    Examples:
-
-    >>> _invert_reordering([3, -1, 2, 4, -1])
-    array([-1, -1,  2,  0,  3])
-    >>> _invert_reordering([-1, -1, 0, 2, 3])
-    array([3, -1,  2,  4])
-    >>> _invert_reordering([1,3,5])
-    array([-1,  0, -1,  1, -1,  2])
-
-    """
-    reordering = np.asanyarray(reordering, dtype=np.int64)
-    if not com.is_integer_dtype(reordering):
-        raise ValueError("Only integer indexers are supported")
-
-    nonneg_indices = reordering[reordering >= 0].astype(np.int_)
-    counts = np.bincount(nonneg_indices, minlength=minlength)
-    has_non_unique = (counts > 1).any()
-
-    dtype = np.dtype(np.object_) if has_non_unique else np.dtype(np.int64)
-    inverted = np.empty_like(counts, dtype=dtype)
-    inverted.fill(-1)
-
-    nonneg_positions = np.arange(len(reordering), dtype=np.int64)[reordering >= 0]
-    np.put(inverted, nonneg_indices, nonneg_positions)
-
-    if has_non_unique:
-        nonunique_elements = np.arange(len(counts))[counts > 1]
-        for elt in nonunique_elements:
-            inverted[elt] = nonneg_positions[nonneg_indices == elt].tolist()
-
-    return inverted
-
-
 def _get_blkno_placements(blknos, blk_count, group=True):
     """
 
diff --git a/pandas/lib.pyx b/pandas/lib.pyx
index c7494c2f4344f..53c4e0a44e8e9 100644
--- a/pandas/lib.pyx
+++ b/pandas/lib.pyx
@@ -495,6 +495,17 @@ def fast_zip(list ndarrays):
     return result
 
 def get_reverse_indexer(ndarray[int64_t] indexer, Py_ssize_t length):
+    """
+    Reverse indexing operation.
+
+    Given `indexer`, make `indexer_inv` of it, such that::
+
+        indexer_inv[indexer[x]] = x
+
+    .. note:: If indexer is not unique, only first occurrence is accounted.
+
+    """
+
     cdef:
         Py_ssize_t i, n = len(indexer)
         ndarray[int64_t] rev_indexer