Performance regression in DataFrame[bool_indexer]

[TomAugspurger]

import numpy as np
import pandas as pd

idx_dupe = np.array(range(30)) * 99
df = pd.DataFrame(np.random.randn(10000, 5))
bool_indexer = [True] * 5000 + [False] * 5000


%timeit df[np.array(bool_indexer)]

# 1.0.2
2.58 ms ± 108 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

# master
5.47 ms ± 192 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

Note that this only affects the case where the indexer is a list. Performance looks fine when an ndarray is passed.

https://pandas.pydata.org/speed/pandas/index.html#indexing.DataFrameNumericIndexing.time_bool_indexer?commits=80d37adcc3d9bfbbe17e8aa626d6b5873465ca98-4f89c261f624305fc7bae6c43ae862663994be34 points to somewhere in 80d37ad..4f89c26, which has a bunch of commits.

[rohitkg98]

The performance issue seems to be in

pandas/pandas/core/indexers.py

Line 348 in e835b76

def check_array_indexer(array: AnyArrayLike, indexer: Any) -> Any:

in which

pandas/pandas/core/construction.py

Line 56 in 911e19b

def array(

is called which calls infer_dtype twice.
It seems to have started from this PR #31150.
The performance also seems to have worsened just a tiny bit after this commit.
I have used cProfile to perform the analysis. I apologise for any mistakes, I'm still a novice dev.