[
https://issues.apache.org/jira/browse/SPARK-46502?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dongjoon Hyun reassigned SPARK-46502:
-------------------------------------
Assignee: L. C. Hsieh
> Support timestamp types in UnwrapCastInBinaryComparison
> -------------------------------------------------------
>
> Key: SPARK-46502
> URL: https://issues.apache.org/jira/browse/SPARK-46502
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 4.0.0
> Reporter: L. C. Hsieh
> Assignee: L. C. Hsieh
> Priority: Major
> Labels: pull-request-available
>
> We have an optimization rule `UnwrapCastInBinaryComparison` that handles
> similar cases but it doesn't cover timestamp types.
> For a query plan like
> ```
> == Analyzed Logical Plan ==
> batch: timestamp
> Project [batch#26466]
> +- Filter (batch#26466 >= cast(2023-12-21 10:00:00 as timestamp))
> +- SubqueryAlias spark_catalog.default.timestamp_view
> +- View (`spark_catalog`.`default`.`timestamp_view`, [batch#26466])
> +- Project [cast(batch#26467 as timestamp) AS batch#26466]
> +- Project [cast(batch#26463 as timestamp) AS batch#26467]
> +- SubqueryAlias spark_catalog.default.table_timestamp
> +- Relation
> spark_catalog.default.table_timestamp[batch#26463] parquet
> == Optimized Logical Plan ==
> Project [cast(batch#26463 as timestamp) AS batch#26466]
> +- Filter (isnotnull(batch#26463) AND (cast(batch#26463 as timestamp) >=
> 2023-12-21 10:00:00))
> +- Relation spark_catalog.default.table_timestamp[batch#26463] parquet
> ```
> The predicate compares a timestamp_ntz column with a literal value. As the
> column is wrapped in a cast expression to timestamp type, the literal
> (string) is wrapped with a cast to timestamp type. The literal with cast is
> foldable so it is evaluated to literal of timestamp earlier. So the predicate
> becomes `cast(batch#26463 as timestamp) >= 2023-12-21 10:00:00`. As the cast
> is in column side, it cannot be pushed down to data source/table.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
