[
https://issues.apache.org/jira/browse/SPARK-61497?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hyukjin Kwon updated SPARK-61497:
---------------------------------
Fix Version/s: connect-rust-4.3.0
> Raise for a str that UTF-8 cannot encode in the options of the stream reader,
> and set the wholetext default of text(), in the Python client as PySpark does
> -----------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: SPARK-61497
> URL: https://issues.apache.org/jira/browse/SPARK-61497
> Project: Spark
> Issue Type: Bug
> Components: Connect
> Affects Versions: connect-rust-4.3.0
> Reporter: Hyukjin Kwon
> Priority: Major
> Fix For: connect-rust-4.3.0
>
>
> Follow-ups of a post-hoc review of SPARK-61475, checked against PySpark 4.3.0:
> - The stream reader's option() and its format methods (for example
> text(lineSep=...), csv(sep=...), and the path option of load()) sent a str
> that UTF-8 cannot encode, such as a lone surrogate from a
> surrogateescape-decoded path, as U+FFFD three times: another text, such as
> another path. SPARK-61475 made them convert through a lossy str(). The batch
> reader, readStream.options() and load(**options) raise protobuf's ValueError,
> as PySpark does.
> - text() of the batch and stream readers took wholetext=None, so an earlier
> wholetext option was kept. PySpark's text(paths, wholetext=False, ...) sets
> its default on every call through _set_opts. So
> spark.read.option("wholetext", True).text(path) reads one row per line in
> PySpark, and one row per file in the Python client. readStream.text(path)
> also leaves wholetext=False in the reader's options in PySpark.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]