DEPR: ignoring missing labels when indexing on MultiIndex level (#42351)

doc/source/whatsnew/v1.4.0.rst

@@ -154,7 +154,7 @@ Deprecations
 - Deprecated ``method`` argument in :meth:`Index.get_loc`, use ``index.get_indexer([label], method=...)`` instead (:issue:`42269`)
 - Deprecated treating integer keys in :meth:`Series.__setitem__` as positional when the index is a :class:`Float64Index` not containing the key, a :class:`IntervalIndex` with no entries containing the key, or a :class:`MultiIndex` with leading :class:`Float64Index` level not containing the key (:issue:`33469`)
 - Deprecated treating ``numpy.datetime64`` objects as UTC times when passed to the :class:`Timestamp` constructor along with a timezone. In a future version, these will be treated as wall-times. To retain the old behavior, use ``Timestamp(dt64).tz_localize("UTC").tz_convert(tz)`` (:issue:`24559`)
--
+- Deprecated ignoring missing labels when indexing with a sequence of labels on a level of a MultiIndex (:issue:`42351`)
 
 .. ---------------------------------------------------------------------------
 

pandas/core/indexes/multi.py

@@ -3268,6 +3268,18 @@ def _update_indexer(idxr: Index, indexer: Index) -> Index:
                         )
                     except KeyError:
                         # ignore not founds; see discussion in GH#39424
+                        warnings.warn(
+                            "The behavior of indexing on a MultiIndex with a nested "
+                            "sequence of labels is deprecated and will change in a "
+                            "future version. `series.loc[label, sequence]` will "
+                            "raise if any members of 'sequence' or not present in "
+                            "the index's second level. To retain the old behavior, "
+                            "use `series.index.isin(sequence, level=1)`",
+                            # TODO: how to opt in to the future behavior?
+                            # TODO: how to handle IntervalIndex level? (no test cases)
+                            FutureWarning,
+                            stacklevel=7,
+                        )
                         continue
                     else:
                         idxrs = _convert_to_indexer(item_lvl_indexer)

pandas/tests/indexing/multiindex/test_loc.py

@@ -398,14 +398,21 @@ def test_loc_getitem_duplicates_multiindex_missing_indexers(indexer, pos):
     idx = MultiIndex.from_product(
         [["A", "B", "C"], ["foo", "bar", "baz"]], names=["one", "two"]
     )
-    s = Series(np.arange(9, dtype="int64"), index=idx).sort_index()
-    expected = s.iloc[pos]
+    ser = Series(np.arange(9, dtype="int64"), index=idx).sort_index()
+    expected = ser.iloc[pos]
 
     if expected.size == 0 and indexer != []:
         with pytest.raises(KeyError, match=str(indexer)):
-            s.loc[indexer]
+            ser.loc[indexer]
     else:
-        result = s.loc[indexer]
+        warn = None
+        msg = "MultiIndex with a nested sequence"
+        if indexer == (slice(None), ["foo", "bah"]):
+            # "bah" is not in idx.levels[1], so is ignored, will raise KeyError
+            warn = FutureWarning
+
+        with tm.assert_produces_warning(warn, match=msg):
+            result = ser.loc[indexer]
         tm.assert_series_equal(result, expected)
 
 

pandas/tests/io/formats/style/test_style.py

@@ -627,10 +627,27 @@ def test_applymap_subset(self, slice_):
     def test_applymap_subset_multiindex(self, slice_):
         # GH 19861
         # edited for GH 33562
+        warn = None
+        msg = "indexing on a MultiIndex with a nested sequence of labels"
+        if (
+            isinstance(slice_[-1], tuple)
+            and isinstance(slice_[-1][-1], list)
+            and "C" in slice_[-1][-1]
+        ):
+            warn = FutureWarning
+        elif (
+            isinstance(slice_[0], tuple)
+            and isinstance(slice_[0][1], list)
+            and 3 in slice_[0][1]
+        ):
+            warn = FutureWarning
+
         idx = MultiIndex.from_product([["a", "b"], [1, 2]])
         col = MultiIndex.from_product([["x", "y"], ["A", "B"]])
         df = DataFrame(np.random.rand(4, 4), columns=col, index=idx)
-        df.style.applymap(lambda x: "color: red;", subset=slice_).render()
+
+        with tm.assert_produces_warning(warn, match=msg, check_stacklevel=False):
+            df.style.applymap(lambda x: "color: red;", subset=slice_).render()
 
     def test_applymap_subset_multiindex_code(self):
         # https://github.com/pandas-dev/pandas/issues/25858
@@ -1438,6 +1455,19 @@ def test_non_reducing_multi_slice_on_multiindex(self, slice_):
         idxs = MultiIndex.from_product([["U", "V"], ["W", "X"], ["Y", "Z"]])
         df = DataFrame(np.arange(64).reshape(8, 8), columns=cols, index=idxs)
 
-        expected = df.loc[slice_]
-        result = df.loc[non_reducing_slice(slice_)]
+        msg = "indexing on a MultiIndex with a nested sequence of labels"
+        warn = None
+        for lvl in [0, 1]:
+            key = slice_[lvl]
+            if isinstance(key, tuple):
+                for subkey in key:
+                    if isinstance(subkey, list) and "-" in subkey:
+                        # not present in the index level, ignored, will raise in future
+                        warn = FutureWarning
+
+        with tm.assert_produces_warning(warn, match=msg):
+            expected = df.loc[slice_]
+
+        with tm.assert_produces_warning(warn, match=msg):
+            result = df.loc[non_reducing_slice(slice_)]
         tm.assert_frame_equal(result, expected)