[
https://issues.apache.org/jira/browse/SPARK-56854?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hyukjin Kwon reassigned SPARK-56854:
------------------------------------
Assignee: Lavan Vivekanandasarma
> Filter None values in PySpark DataFrame[Stream]Reader/Writer .option(s) for
> parity with Spark Connect
> -----------------------------------------------------------------------------------------------------
>
> Key: SPARK-56854
> URL: https://issues.apache.org/jira/browse/SPARK-56854
> Project: Spark
> Issue Type: Bug
> Components: PySpark
> Affects Versions: 5.0.0
> Reporter: Lavan Vivekanandasarma
> Assignee: Lavan Vivekanandasarma
> Priority: Major
> Labels: pull-request-available
>
> Classic PySpark's DataFrame[Stream]Reader/Writer.option(key, None) and
> .options(**\{k: None}) forward None to the JVM as Java null. This diverges
> from the Spark Connect Python client (which has filtered None since
> SPARK-49263) and from OptionUtils._set_opts at
> python/pyspark/sql/readwriter.py:41-53, which already filters None.
> Example: spark.read.options(nullValue=None).schema("a STRING, b
> STRING").csv(path)
> For a row '"",val', Classic returns [Row(a='', b='val')] while Connect
> returns [Row(a=None, b='val')].
> Proposal: filter None from the public option and options methods on
> DataFrameReader, DataFrameWriter, DataFrameWriterV2, DataStreamReader, and
> DataStreamWriter, so Classic matches Connect and _set_opts. After the change,
> option(k, None) is a no-op and options(**\{k: None}) drops None entries
> before forwarding.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
