[SPARK-30983] Support more than 5 typed column in typed Dataset.select API

• Type: Improvement
• Status: In Progress
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.1.0
• Component/s: SQL
• Assignee: Unassigned
• Reporter: L. C. Hsieh
• Created: 28/Feb/20 07:31
• Updated: 19/May/22 17:19

Description

Because Dataset only provides overloading typed select API to at most 5 typed columns, once more than 5 typed columns given, the select API call will go for untyped one.

Currently users cannot call typed select with more than 5 typed columns. There are few options:

1. Expose Dataset.selectUntyped (could rename it) to accept any number (due to the limit of ExpressionEncoder.tuple, at most 22 actually) of typed columns. Pros: not need to add too much code in Dataset. Cons: The returned type is generally Dataset[_], not a specified one like Dataset[(U1, U2)] for the overloading method.

2. Add more overloading typed select APIs up to 22 typed column inputs. Pros: Clear returned type. Cons: A lot of code to be added to Dataset for just corner cases. It can be a breaking change to existing user code that calls untyped select API.

Issue Links

• links to: [Github] Pull Request #36613 (nkronenfeld)