Bug: Incorrect date 1970-01-01 for the second row when querying from Polars to_arrow() table

[baiyus]

chdb version: v3.6.0 polars version: 1.34.0 python version: 3.12.3

Bug Description:
When querying a PyArrow table that was generated by polars.DataFrame.to_arrow(), chdb incorrectly parses the date32 column starting from the second row. The first row's date is read correctly, but subsequent identical dates (2000-01-01) are returned as 1970-01-01 (the Unix epoch date).

This issue seems specific to the Arrow table created by Polars, as tables created from Pandas do not exhibit this behavior.

Steps to Reproduce:
The following Python script consistently reproduces the bug using simulated data.

import chdb
import polars as pl

# 1. Create a Polars DataFrame with a date column from strings
data_dict = {
    'id': [1, 1],
    'td_date': ['2000-01-01', '2000-01-01'],
    'value': [1234.1, 2345.7]
}
df_polars = pl.DataFrame(data_dict)
df_polars = df_polars.with_columns(pl.col("td_date").str.to_date())

# 2. Convert the Polars DataFrame to a PyArrow Table
arrow_table_from_polars = df_polars.to_arrow()

print("--- Initial Polars DataFrame ---")
print(df_polars)
print("\n--- Arrow Table from Polars (looks correct) ---")
print(arrow_table_from_polars)

# 3. Query with chdb, which triggers the bug
result = chdb.query('''
    SELECT *
    FROM Python(arrow_table_from_polars)
    LIMIT 2
''', output_format='Markdown')

print("\n--- chdb Query Result (Buggy) ---")
print(result)

Actual Behavior: The query returns the incorrect date 1970-01-01 for the second row.

--- Initial Polars DataFrame ---
shape: (2, 3)
┌─────┬────────────┬───────────┐
│ id  ┆ td_date    ┆ value     │
│ --- ┆ ---        ┆ ---       │
│ i64 ┆ date       ┆ f64       │
╞═════╪════════════╪═══════════╡
│ 1   ┆ 2000-01-01 ┆ 1234.1000 │
│ 1   ┆ 2000-01-01 ┆ 2345.7000 │
└─────┴────────────┴───────────┘

--- Arrow Table from Polars (looks correct) ---
pyarrow.Table
id: int64
td_date: date32[day]
value: double
----
id: [[1,1]]
td_date: [[2000-01-01,2000-01-01]]
value: [[1234.1,2345.7]]

--- chdb Query Result (Buggy) ---
| id | td_date | value |
|-:|-:|-:|
| 1 | 2000-01-01 | 1234.1 |
| 1 | 1970-01-01 | 2345.7 |

Workaround:
A functioning workaround is to convert the date column back to a string within Polars before converting the DataFrame to an Arrow table. This suggests the issue is specific to how chdb parses the date32 array type when it originates from Polars.

# Workaround implementation
df_polars_str_date = df_polars.with_columns(
    pl.col("td_date").dt.strftime("%Y-%m-%d")
)
arrow_table_workaround = df_polars_str_date.to_arrow()

# A chdb query on 'arrow_table_workaround' works correctly.

[kafka1991]

Hi @baiyus
Thans for your report!
I can repro this and will fix it in next chdb release.

[wudidapaopao]

Hi @baiyus, @kafka1991, the issue will be resolved by #385