[
https://issues.apache.org/jira/browse/ARROW-14104?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17430530#comment-17430530
]
Sarah Gilmore commented on ARROW-14104:
---------------------------------------
So sorry about the delay. [~jorisvandenbossche] and [~westonpace].
I've attached two files ([^exampleArrow4.parq] and [^exampleArrow5.parq]) that
were both created with the following code:
{code:java}
// code placeholder
import pyarrow as pa
import pyarrow.parquet as pq
import datetime
>>> column = pa.array([[datetime.datetime(2023, 9, 23, 11)]],
>>> pa.list_(pa.timestamp('us', 'America/New_York')));
>>> t = pa.Table.from_arrays([column], names=['TimestampColumn']);
>>> pq.write_table(t, "example.parq");
{code}
I ran the code snippet above in pyarrow 4.0.0 and pyarrow 5.0.0 to create
exampleArrow4.parq and exampleArrow5.parq, respectively.
Here's the output of reading both files in pyarrow 4.0.0:
{code:java}
// code placeholder
import pyarrow as pa
import pyarrow.parquet as pq
>>> t1 = pq.read_table("exampleArrow4.parq")
>>> t1
pyarrow.Table
TimestampColumn: list<item: timestamp[us, tz=America/New_York]>
child 0, item: timestamp[us, tz=America/New_York]
>>> t2 = pq.read_table("exampleArrow5.parq")
>>> t2
pyarrow.Table
TimestampColumn: list<item: timestamp[us, tz=America/New_York]>
child 0, item: timestamp[us, tz=America/New_York]{code}
The TimeZone is read in properly for both files.
Here's the output of reading both files in pyarrow 5.0.0:
{code:java}
// code placeholder
import pyarrow as pa import pyarrow.parquet as pq
>>> t1 = pq.read_table("exampleArrow4.parq")
>>> t1
pyarrow.Table
TimestampColumn: list<item: timestamp[us, tz=UTC]>
child 0, item: timestamp[us, tz=UTC]
>>> t2 = pq.read_table("exampleArrow5.parq")
>>> t2
pyarrow.Table
TimestampColumn: list<item: timestamp[us, tz=UTC]>
child 0, item: timestamp[us, tz=UTC]
{code}
It looks like pyarrow 5.0.0 writes out the TimeZone information, but doesn't
read it in properly.
> Reading Lists of Timestamps from parquet files in Arrow 5.0.0 fails to
> preserve the TimeZone - unlike in Arrow 4.0.0
> --------------------------------------------------------------------------------------------------------------------
>
> Key: ARROW-14104
> URL: https://issues.apache.org/jira/browse/ARROW-14104
> Project: Apache Arrow
> Issue Type: Bug
> Components: C++, Parquet, Python
> Affects Versions: 5.0.0
> Reporter: Sarah Gilmore
> Priority: Minor
> Attachments: exampleArrow4.parq, exampleArrow5.parq
>
>
> In Arrow 4.0.0 it is possible to round-trip the TimeZone property of
> List<Timestamp> columns to and from parquet files:
> {code:java}
> >>> import pyarrow as pa
> >>> import pyarrow.parquet as pq
> >>> import datetime
> >>> column = pa.array([[datetime.datetime(2023, 9, 23, 11)]],
> >>> pa.list_(pa.timestamp('us', 'America/New_York')));
> >>> t = pa.Table.from_arrays([column], names=['TimestampColumn']);
> >>> pq.write_table(t, "example.parq");
> >>> t2 = pq.read_table("example.parq");
> >>> t2
> pyarrow.Table
> Dates: list<item: timestamp[us, tz=America/New_York]>
> child 0, item: timestamp[us, tz=America/New_York]
> {code}
> However, if you read the same parquet file in pyarrow 5.0.0, the TimeZone is
> set to UTC:
> {code:java}
> >>> t3 = pq.read_table("example.parq");
> >>> t3
> pyarrow.Table
> Dates: list<item: timestamp[us, tz=UTC]>
> child 0, item: timestamp[us, tz=UTC]
> {code}
>
> I noticed that the TimeZone is preserved in Arrow 5.0 when reading non-nested
> timestamp columns.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
