[
https://issues.apache.org/jira/browse/SPARK-53779?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yicong Huang updated SPARK-53779:
---------------------------------
Description:
Proposal to introduce a transform API for Column in Spark, inspired by Scala’s
pipe operator and SQL pipeline syntax. This would allow chaining
transformations in a pipeline style, improving readability compared to nested
function calls.
{*}Motivation{*}{*}{*}
* Scala’s pipe API and SQL pipeline syntax provide a cleaner,
pipeline-oriented style.
* Current nested function invocations (e.g., f2(f1(col))) are less readable
than a chained style (col.transform(f1).transform(f2)).
* AI code generators also tend to produce pipeline style code more cleanly.
* This aligns with the existing DataFrame API pipeline style (df.transform(f)
→ DataFrame).
> Implement transform in column API
> ---------------------------------
>
> Key: SPARK-53779
> URL: https://issues.apache.org/jira/browse/SPARK-53779
> Project: Spark
> Issue Type: New Feature
> Components: Spark Core
> Affects Versions: 4.1.0
> Reporter: Yicong Huang
> Priority: Major
> Labels: pull-request-available
>
> Proposal to introduce a transform API for Column in Spark, inspired by
> Scala’s pipe operator and SQL pipeline syntax. This would allow chaining
> transformations in a pipeline style, improving readability compared to nested
> function calls.
>
> {*}Motivation{*}{*}{*}
>
> * Scala’s pipe API and SQL pipeline syntax provide a cleaner,
> pipeline-oriented style.
> * Current nested function invocations (e.g., f2(f1(col))) are less readable
> than a chained style (col.transform(f1).transform(f2)).
> * AI code generators also tend to produce pipeline style code more cleanly.
> * This aligns with the existing DataFrame API pipeline style
> (df.transform(f) → DataFrame).
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
