[jira] [Commented] (ARROW-11538) [Python] Seg Fault with pandas.Timestamp filter

["Joris Van den Bossche (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ARROW-11538?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17280887#comment-17280887 ] 

Joris Van den Bossche commented on ARROW-11538:
-----------------------------------------------

[~josham] thanks a lot for the clear report! 
We got a similar report about int96 timestamps (ARROW-11480), but apparently it also happens with nanosecond timestamps (because you are passing {{version="2.0"}}, you save the pandas nanosecond timestamps like that in the parquet file. With the default of {{version="1.0"}}, we convert them to microseconds to write in the parquet file).

> [Python] Seg Fault with pandas.Timestamp filter
> -----------------------------------------------
>
>                 Key: ARROW-11538
>                 URL: https://issues.apache.org/jira/browse/ARROW-11538
>             Project: Apache Arrow
>          Issue Type: Bug
>          Components: Python
>    Affects Versions: 3.0.0
>         Environment: platform: Linux 64bit
> conda env:
> conda create -n pya python=3.8 pyarrow=3.0.0 pandas=1.2.1 pytest -c conda-forge
>            Reporter: Josh
>            Priority: Minor
>             Fix For: 4.0.0, 3.0.1
>
>
> The first two tests pass but the third gives: Fatal Python error: Segmentation fault
> All three pass in with pyarrow=2.0.0
> {code:java}
> import pandas
> import pyarrow as pa
> import pyarrow.dataset as ds
> import pyarrow.parquet as pq
> import pytest
> @pytest.fixture
> def data_path(tmp_path):
>     path = tmp_path / "data.parquet"
>     df = pandas.DataFrame(
>         [
>             ["A", pandas.Timestamp("2020-11-04")],
>         ],
>         columns=["name", "date"],
>     )
>     table = pa.Table.from_pandas(df)
>     pq.write_table(table, path, version="2.0")
>     return df, path
> @pytest.mark.parametrize(
>     "filter",
>     [
>         None,
>         ds.field("date") == "2020-11-04",
>         ds.field("date") == pandas.Timestamp("2020-11-04"),
>     ],
> )
> def test_dataset_filter(filter, data_path):
>     data, path = data_path
>     dataset = ds.dataset(path, format="parquet")
>     assert data.equals(dataset.to_table(filter=filter).to_pandas())
> {code}



--
This message was sent by Atlassian Jira
(v8.3.4#803005)