I noticed a significant performance deterioration when using polars dataframe join function after upgrading polars from 1.30.0 to 1.31.0. The code snippet is below:

import polars as pl
import time
import numpy as np

print(pl.__version__)
np.random.seed(0)

indices = np.arange(2_000)
columns = [f"col_{i}" for i in range(20_000)]

df_1 = pl.DataFrame({
    "index": indices,
    **{col: np.random.rand(len(indices)) for col in columns}
})

df_2 = pl.DataFrame({
    "index": indices,
    **{col: np.random.rand(len(indices)) for col in columns}
})

print("DataFrames created.")

t0 = time.time()
df_merged = df_1.join(df_2, on="index", how="left", suffix="_right")
t1 = time.time()
print(f"Time taken to merge: {t1 - t0:.2f} seconds") 

When using polars 1.30.0, the merge step takes 0.06 seconds,

1.30.0
DataFrames created.
Time taken to merge: 0.06 seconds

but when using polars 1.31.0, the merge step takes almost 30 seconds

1.31.0
DataFrames created.
Time taken to merge: 27.68 seconds

Anyone knows why that happened?