BUG: Revert sort optimization in np.unique.

The optimization was to sort integer subarrays by treating them as
strings of unsigned bytes. That worked fine for finding the unique
subarrays, but the sort order of the results could be unexpected.

Closes #10495.

1 files changed, 20 insertions, 17 deletions

diff --git a/numpy/lib/arraysetops.py b/numpy/lib/arraysetops.py
index 7b103ef3e..e8eda297f 100644
--- a/numpy/lib/arraysetops.py
+++ b/numpy/lib/arraysetops.py
@@ -148,16 +148,15 @@ def unique(ar, return_index=False, return_inverse=False,
         .. versionadded:: 1.9.0
 
     axis : int or None, optional
-        The axis to operate on. If None, `ar` will be flattened beforehand.
-        Otherwise, duplicate items will be removed along the provided axis,
-        with all the other axes belonging to the each of the unique elements.
-        Object arrays or structured arrays that contain objects are not
-        supported if the `axis` kwarg is used.
+        The axis to operate on. If None, `ar` will be flattened. If an integer,
+        the subarrays indexed by the given axis will be flattened and treated
+        as the elements of a 1-D array with the dimension of the given axis,
+        see the notes for more details.  Object arrays or structured arrays
+        that contain objects are not supported if the `axis` kwarg is used. The
+        default is None.
 
         .. versionadded:: 1.13.0
 
-
-
     Returns
     -------
     unique : ndarray
@@ -179,6 +178,17 @@ def unique(ar, return_index=False, return_inverse=False,
     numpy.lib.arraysetops : Module with a number of other functions for
                             performing set operations on arrays.
 
+    Notes
+    -----
+    When an axis is specified the subarrays indexed by the axis are sorted.
+    This is done by making the specified axis the first dimension of the array
+    and then flattening the subarrays in C order. The flattened subarrays are
+    then viewed as a structured type with each element given a label, with the
+    effect that we end up with a 1-D array of structured types that can be
+    treated in the same way as any other 1-D array. The result is that the
+    flattened subarrays are sorted in lexicographic order starting with the
+    first element.
+
     Examples
     --------
     >>> np.unique([1, 1, 2, 2, 3, 3])
@@ -223,25 +233,18 @@ def unique(ar, return_index=False, return_inverse=False,
         ret = _unique1d(ar, return_index, return_inverse, return_counts)
         return _unpack_tuple(ret)
 
+    # axis was specified and not None
     try:
         ar = np.swapaxes(ar, axis, 0)
     except np.AxisError:
         # this removes the "axis1" or "axis2" prefix from the error message
         raise np.AxisError(axis, ar.ndim)
 
-    orig_shape, orig_dtype = ar.shape, ar.dtype
     # Must reshape to a contiguous 2D array for this to work...
+    orig_shape, orig_dtype = ar.shape, ar.dtype
     ar = ar.reshape(orig_shape[0], -1)
     ar = np.ascontiguousarray(ar)
-
-    if ar.dtype.char in (np.typecodes['AllInteger'] +
-                         np.typecodes['Datetime'] + 'S'):
-        # Optimization: Creating a view of your data with a np.void data type of
-        # size the number of bytes in a full row. Handles any type where items
-        # have a unique binary representation, i.e. 0 is only 0, not +0 and -0.
-        dtype = np.dtype((np.void, ar.dtype.itemsize * ar.shape[1]))
-    else:
-        dtype = [('f{i}'.format(i=i), ar.dtype) for i in range(ar.shape[1])]
+    dtype = [('f{i}'.format(i=i), ar.dtype) for i in range(ar.shape[1])]
 
     try:
         consolidated = ar.view(dtype)