Hyukjin Kwon created SPARK-61040:
------------------------------------
Summary: Parse the DDL strings of DataType.fromDDL and of the
return types of UDFs on the server in the Python client as PySpark does
Key: SPARK-61040
URL: https://issues.apache.org/jira/browse/SPARK-61040
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
In PySpark Connect, {{DataType.fromDDL(ddl)}}, {{StructType.fromDDL}} and
{{pyspark.sql.types._parse_datatype_string(s)}} parse a DDL string on the
server of the active session ({{SparkSession._parse_ddl}}, the {{DDLParse}} of
an {{AnalyzePlanRequest}}), as does a {{UserDefinedFunction}} of its DDL return
type, as {{udf(f, "a INT NOT NULL")}}. So they take any DDL that Spark does,
such as {{"a INT NOT NULL COMMENT 'c'"}}, {{"st struct<a:int,b:double>"}} and
{{"s STRING COLLATE UTF8_LCASE"}}, and raise the errors of Spark's parser, such
as {{PARSE_SYNTAX_ERROR}} for {{"a INT,"}}.
The Rust-backed Python client parses them on the client. It raises
{{ValueError: [INVALID_DATATYPE_FORMAT]}} for {{NOT NULL}} and nested types,
takes {{"s STRING COLLATE UTF8_LCASE"}} as a string of the collation {{collate
utf8_lcase}}, and {{"a INT,"}} as {{a INT}}. It has no
{{pyspark.sql.types._parse_datatype_string}}. Parse them as PySpark does.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
