[
https://issues.apache.org/jira/browse/SPARK-41226?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Apache Spark reassigned SPARK-41226:
------------------------------------
Assignee: Apache Spark
> Refactor Spark types by introducing physical types
> --------------------------------------------------
>
> Key: SPARK-41226
> URL: https://issues.apache.org/jira/browse/SPARK-41226
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 3.4.0
> Reporter: Gengliang Wang
> Assignee: Apache Spark
> Priority: Major
>
> I am creating this one for Desmond Cheong since he can't signup for an
> account because of
> [https://infra.apache.org/blog/jira-public-signup-disabled.html.|https://infra.apache.org/blog/jira-public-signup-disabled.html]
>
> His description for this improvement:
> The Spark type system currently supports multiple data types with the same
> physical representation in memory. For example {{DateType}} and
> {{YearMonthIntervalType}} are both implemented using {{{}IntegerType{}}}.
> Because of this, operations on data types often involve case matching where
> multiple data types match to the same effects.To simplify this case matching
> logic, we can introduce the notion of logical and physical data types where
> multiple logical data types can be implemented with the same physical data
> type, then perform case matching on physical data types.Some areas that can
> utilize this logical/physical type separation are: *
> {{SpecializedGettersReader}} in {{SpecializedGettersReader.java}}
> * {{copy}} in {{ColumnarBatchRow.java}} and {{ColumnarRow.java}}
> * {{getAccessor}} in {{InternalRow.scala}}
> * {{externalDataTypeFor}} in {{RowEncoder.scala}}
> * {{unsafeWriter}} in {{InterpretedUnsafeProjection.scala}}
> * {{getValue}} and {{javaType}} in {{CodeGenerator.scala}}
> * {{doValidate}} in {{literals.scala}}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
