BUG: fix datetimes.should_cache() error for deque (GH 29403)

itertools.islice() should be used to get slice of a deque. itertools.islice() also can be used (and is efficient) for other collections. So
unique(arg[:check_count])
was replaced with
set(islice(arg, check_count))

Author: nathalier

pandas/core/tools/datetimes.py

@@ -1,6 +1,7 @@
 from collections import abc
 from datetime import datetime, time
 from functools import partial
+from itertools import islice
 from typing import Optional, TypeVar, Union
 
 import numpy as np
@@ -118,7 +119,7 @@ def should_cache(
 
     assert 0 < unique_share < 1, "unique_share must be in next bounds: (0; 1)"
 
-    unique_elements = unique(arg[:check_count])
+    unique_elements = set(islice(arg, check_count))
     if len(unique_elements) > check_count * unique_share:
         do_caching = False
     return do_caching

pandas/tests/indexes/datetimes/test_tools.py

@@ -3,6 +3,7 @@
 import calendar
 from datetime import datetime, time
 import locale
+from collections import deque
 
 from dateutil.parser import parse
 from dateutil.tz.tz import tzoffset
@@ -2274,6 +2275,16 @@ def test_should_cache(listlike, do_caching):
         == do_caching
     )
 
+@pytest.mark.parametrize(
+    "listlike",
+    [(deque([pd.Timestamp('2010-06-02 09:30:00')] * 51)),
+     ([pd.Timestamp('2010-06-02 09:30:00')] * 51),
+     (tuple([pd.Timestamp('2010-06-02 09:30:00')] * 51))],
+)
+def test_no_slicing_errors_in_should_cache(listlike):
+    # GH 29403
+    assert (tools.should_cache(listlike) is True)
+
 
 @pytest.mark.parametrize(
     "unique_share,check_count, err_message",