Hyukjin Kwon created SPARK-60921:
------------------------------------
Summary: Make data types in Spark Connect Rust compare and hash as
PySpark's do
Key: SPARK-60921
URL: https://issues.apache.org/jira/browse/SPARK-60921
Project: Spark
Issue Type: Bug
Components: Connect
Affects Versions: connect-rust-4.3.0
Reporter: Hyukjin Kwon
In PySpark, two data types are equal when they are of the same class and have
the same attributes ({{DataType.__eq__}}), and a data type hashes as
{{hash(str(self))}}. The Rust-backed Python client's data types, except
{{StructField}}, define no equality, so they compare by identity, and so these
PySpark 4.3.0 results differ:
* {{IntegerType() == IntegerType()}}, {{DecimalType(10, 2) == DecimalType(10,
2)}}, {{ArrayType(IntegerType()) == ArrayType(IntegerType())}},
{{MapType(StringType(), LongType()) == MapType(StringType(), LongType())}} and
the same for {{StructType}} are true. The client returns false for all of them,
and true for {{!=}}.
* So does every comparison of a schema or a field type, as in {{df.schema ==
StructType(...)}}, {{field.dataType == StringType()}}, or a data type and its
pickled copy.
* Equal data types hash equally, and {{StructField}} is hashable. The client's
hashes differ, and its {{StructField}} is unhashable.
Make data types compare and hash as PySpark's do.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
