PERF: accessing sliced indexes with populated indexing engines

Author: topper-123

doc/source/whatsnew/v2.1.0.rst

@@ -105,6 +105,7 @@ Performance improvements
 - Performance improvement in :meth:`DataFrame.first_valid_index` and :meth:`DataFrame.last_valid_index` for extension array dtypes (:issue:`51549`)
 - Performance improvement in :meth:`DataFrame.clip` and :meth:`Series.clip` (:issue:`51472`)
 - Performance improvement in :func:`read_parquet` on string columns when using ``use_nullable_dtypes=True`` (:issue:`47345`)
+- Performance improvement when searching an :class:`Index` sliced from other indexes (:issue:`xxxxx`)
 - Performance improvement in :meth:`read_orc` when reading a remote URI file path. (:issue:`51609`)
 
 .. ---------------------------------------------------------------------------

pandas/_libs/index.pyx

@@ -230,6 +230,18 @@ cdef class IndexEngine:
     def __sizeof__(self) -> int:
         return self.sizeof()
 
+    cpdef _update_from_other(self, IndexEngine other, reverse: bool):
+        self.unique = other.unique
+        self.need_unique_check = other.need_unique_check
+        self.need_monotonic_check = other.need_monotonic_check
+        if reverse and not self.need_monotonic_check:
+            # the slice reverses the index
+            self.monotonic_inc = other.monotonic_dec
+            self.monotonic_dec = other.monotonic_inc
+        else:
+            self.monotonic_inc = other.monotonic_inc
+            self.monotonic_dec = other.monotonic_dec
+
     @property
     def is_unique(self) -> bool:
         if self.need_unique_check:

pandas/core/indexes/base.py

@@ -5153,9 +5153,7 @@ def __getitem__(self, key):
         if isinstance(key, slice):
             # This case is separated from the conditional above to avoid
             # pessimization com.is_bool_indexer and ndim checks.
-            result = getitem(key)
-            # Going through simple_new for performance.
-            return type(self)._simple_new(result, name=self._name)
+            return self._getitem_slice(key)
 
         if com.is_bool_indexer(key):
             # if we have list[bools, length=1e5] then doing this check+convert
@@ -5181,7 +5179,12 @@ def _getitem_slice(self: _IndexT, slobj: slice) -> _IndexT:
         Fastpath for __getitem__ when we know we have a slice.
         """
         res = self._data[slobj]
-        return type(self)._simple_new(res, name=self._name)
+        result = type(self)._simple_new(res, name=self._name)
+        if isinstance(result._engine, libindex.IndexEngine): # may also be IntervalTree
+            reverse = slobj.step is not None and slobj.step < 0
+            result._engine._update_from_other(self._engine, reverse=reverse)
+
+        return result
 
     @final
     def _can_hold_identifiers_and_holds_name(self, name) -> bool: