BUG: catch almost-null-slice in _convert_slice_indexer (#31866)

Author: jbrockmendel

pandas/core/indexes/base.py

@@ -3146,8 +3146,7 @@ def _convert_slice_indexer(self, key: slice, kind: str_t):
         def is_int(v):
             return v is None or is_integer(v)
 
-        is_null_slicer = start is None and stop is None
-        is_index_slice = is_int(start) and is_int(stop)
+        is_index_slice = is_int(start) and is_int(stop) and is_int(step)
         is_positional = is_index_slice and not (
             self.is_integer() or self.is_categorical()
         )
@@ -3177,7 +3176,7 @@ def is_int(v):
             except KeyError:
                 pass
 
-        if is_null_slicer:
+        if com.is_null_slice(key):
             indexer = key
         elif is_positional:
             indexer = key

pandas/tests/indexes/test_base.py

@@ -2611,3 +2611,18 @@ def test_validate_1d_input():
     ser = pd.Series(0, range(4))
     with pytest.raises(ValueError, match=msg):
         ser.index = np.array([[2, 3]] * 4)
+
+
+def test_convert_almost_null_slice(indices):
+    # slice with None at both ends, but not step
+    idx = indices
+
+    key = slice(None, None, "foo")
+
+    if isinstance(idx, pd.IntervalIndex):
+        with pytest.raises(ValueError, match="cannot support not-default step"):
+            idx._convert_slice_indexer(key, "loc")
+    else:
+        msg = "'>=' not supported between instances of 'str' and 'int'"
+        with pytest.raises(TypeError, match=msg):
+            idx._convert_slice_indexer(key, "loc")