[ 
https://issues.apache.org/jira/browse/SPARK-61849?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Hyukjin Kwon reassigned SPARK-61849:
------------------------------------

    Assignee: Hyukjin Kwon

> Load the data types that PySpark pickled, and make the parameters of a type 
> its attributes as PySpark's
> -------------------------------------------------------------------------------------------------------
>
>                 Key: SPARK-61849
>                 URL: https://issues.apache.org/jira/browse/SPARK-61849
>             Project: Spark
>          Issue Type: Bug
>          Components: Connect
>    Affects Versions: connect-rust-4.4.0
>            Reporter: Hyukjin Kwon
>            Assignee: Hyukjin Kwon
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: connect-rust-4.4.0
>
>
> PySpark's data types are plain classes, which pickle as their class and their 
> __dict__ (copyreg.__newobj__ at protocols 2 and up). The drop-in's native 
> type classes require their constructor arguments and have no __setstate__, so 
> the pickle of a type that official PySpark made does not load right in the 
> drop-in:
> - DecimalType(10,2) loads as DecimalType(10,0), StringType('UTF8_LCASE') as 
> StringType(), TimeType(3) as TimeType(6), DayTimeIntervalType(1,2) as 
> DayTimeIntervalType(0,3);
> - ArrayType, MapType, StructField, StructType, CharType, VarcharType, 
> GeometryType and GeographyType raise TypeError ("missing 1 required 
> positional argument"), so a pickled df.schema does not load at all.
> Also, the parameters of a type (DecimalType.scale, StringType.collation, 
> CharType.length, TimeType.precision, the interval fields, ...) are plain 
> attributes in PySpark, whose assignment changes the type; in the drop-in they 
> are read-only.
> Make PySpark's pickles of the types load as the same types at protocols 2 to 
> 5, and the parameters assignable as PySpark's.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to