PERF: fix clean_index_list perf (#16295)

closes #16285

Author: jreback

asv_bench/benchmarks/indexing.py

@@ -19,6 +19,9 @@ def time_getitem_list_like(self):
     def time_getitem_array(self):
         self.s[np.arange(10000)]
 
+    def time_getitem_lists(self):
+        self.s[np.arange(10000).tolist()]
+
     def time_iloc_array(self):
         self.s.iloc[np.arange(10000)]
 

doc/source/whatsnew/v0.20.2.txt

@@ -26,6 +26,7 @@ Enhancements
 Performance Improvements
 ~~~~~~~~~~~~~~~~~~~~~~~~
 
+- Performance regression fix when indexing with a list-like (:issue:`16285`)
 
 
 .. _whatsnew_0202.bug_fixes:

pandas/_libs/lib.pyx

@@ -950,7 +950,6 @@ def clean_index_list(list obj):
     Utility used in pandas.core.index._ensure_index
     """
     cdef:
-        ndarray[object] converted
         Py_ssize_t i, n = len(obj)
         object v
         bint all_arrays = 1
@@ -964,15 +963,20 @@ def clean_index_list(list obj):
     if all_arrays:
         return obj, all_arrays
 
-    converted = np.empty(n, dtype=object)
-    for i in range(n):
-        v = obj[i]
-        if PyList_Check(v) or np.PyArray_Check(v) or hasattr(v, '_data'):
-            converted[i] = tuple(v)
-        else:
-            converted[i] = v
+    # don't force numpy coerce with nan's
+    inferred = infer_dtype(obj)
+    if inferred in ['string', 'bytes', 'unicode',
+                    'mixed', 'mixed-integer']:
+        return np.asarray(obj, dtype=object), 0
+    elif inferred in ['integer']:
+
+        # TODO: we infer an integer but it *could* be a unint64
+        try:
+            return np.asarray(obj, dtype='int64'), 0
+        except OverflowError:
+            return np.asarray(obj, dtype='object'), 0
 
-    return maybe_convert_objects(converted), 0
+    return np.asarray(obj), 0
 
 
 ctypedef fused pandas_string:

pandas/core/indexes/base.py

@@ -3960,7 +3960,7 @@ def _ensure_index(index_like, copy=False):
     if isinstance(index_like, list):
         if type(index_like) != list:
             index_like = list(index_like)
-        # 2200 ?
+
         converted, all_arrays = lib.clean_index_list(index_like)
 
         if len(converted) > 0 and all_arrays: