Specify stable sort in indexing

astropy/table/index.py

@@ -94,7 +94,7 @@
             raise ValueError("Cannot create index without at least one column")
         elif len(columns) == 1:
             col = columns[0]
-            row_index = Column(col.argsort())
+            row_index = Column(col.argsort(kind="stable"))
             data = Table([col[row_index]])
         else:
             num_rows = len(columns[0])
@@ -117,7 +117,7 @@
             try:
                 lines = table[np.lexsort(sort_columns)]
             except TypeError:  # arbitrary mixins might not work with lexsort
-                lines = table[table.argsort()]
+                lines = table[table.argsort(kind="stable")]
             data = lines[lines.colnames[:-1]]
             row_index = lines[lines.colnames[-1]]
 

astropy/table/tests/test_groups.py

@@ -690,3 +690,23 @@ def test_group_mixins_unsupported(col):
     tg = t.group_by("a")
     with pytest.warns(AstropyUserWarning, match="Cannot aggregate column 'mix'"):
         tg.groups.aggregate(np.sum)
+
+
+@pytest.mark.parametrize("add_index", [False, True])
+def test_group_stable_sort(add_index):
+    """Test that group_by preserves the order of the table.
+
+    This table has 5 groups with an average of 200 rows per group, so it is not
+    statistically possible that the groups will be in order by chance.
+
+    This tests explicitly the case where grouping is done via the index sort.
+    See: https://github.com/astropy/astropy/issues/14882
+    """
+    a = np.random.randint(0, 5, 1000)
+    b = np.arange(len(a))
+    t = Table([a, b], names=["a", "b"])
+    if add_index:
+        t.add_index("a")
+    tg = t.group_by("a")
+    for grp in tg.groups:
+        assert np.all(grp["b"] == np.sort(grp["b"]))

astropy/time/core.py

@@ -1441,13 +1441,28 @@ def argmax(self, axis=None, out=None):
 
         return dt.argmax(axis, out)
 
-    def argsort(self, axis=-1):
+    def argsort(self, axis=-1, kind="stable"):
         """Returns the indices that would sort the time array.
 
-        This is similar to :meth:`~numpy.ndarray.argsort`, but adapted to ensure
-        that the full precision given by the two doubles ``jd1`` and ``jd2``
-        is used, and that corresponding attributes are copied.  Internally,
-        it uses :func:`~numpy.lexsort`, and hence no sort method can be chosen.
+        This is similar to :meth:`~numpy.ndarray.argsort`, but adapted to ensure that
+        the full precision given by the two doubles ``jd1`` and ``jd2`` is used, and
+        that corresponding attributes are copied.  Internally, it uses
+        :func:`~numpy.lexsort`, and hence no sort method can be chosen.
+
+        Parameters
+        ----------
+        axis : int, optional
+            Axis along which to sort. Default is -1, which means sort along the last
+            axis.
+        kind : 'stable', optional
+            Sorting is done with :func:`~numpy.lexsort` so this argument is ignored, but
+            kept for compatibility with :func:`~numpy.argsort`. The sorting is stable,
+            meaning that the order of equal elements is preserved.
+
+        Returns
+        -------
+        indices : ndarray
+            An array of indices that sort the time array.
         """
         # For procedure, see comment on argmin.
         jd1, jd2 = self.jd1, self.jd2

docs/changes/table/14907.bugfix.rst

@@ -0,0 +1,3 @@
+Fix a bug where table indexes were not using a stable sort order. This was causing the
+order of rows within groups to not match the original table order when an indexed table
+was grouped.

docs/table/operations.rst

@@ -111,6 +111,13 @@ values and the indices of the group boundaries for those key values. The groups
 here correspond to the row slices ``0:4``, ``4:7``, and ``7:10`` in the
 ``obs_by_name`` table.
 
+The output grouped table has two important properties:
+
+- The groups in the order of the lexically sorted key values (``M101``, ``M31``,
+  ``M82`` in our example).
+- The rows within each group are in the same order as they appear in the
+  original table.
+
 The initial argument (``keys``) for the :func:`~astropy.table.Table.group_by`
 function can take a number of input data types: