[ARROW-13876] [C++] Uniform null handling in compute functions

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 6.0.0
• Component/s: C++
• Labels: kernel pull-request-available types
• External issue URL: https://github.com/apache/arrow/issues/29494
• Assignee: David Li
• Reporter: Weston Pace
• Created: 03/Sep/21 01:16
• Updated: 11/Jan/23 08:36
• Resolved: 28/Sep/21 18:18

Description

The compute functions today have mixed support for null types.

Unary arithmetic functions (e.g. abs) don't support null arrays

Binary arithmetic functions (e.g. add) support one null array (e.g. int32 + null) but not both null arrays (i.e. null + null) but they do support both values being null (e.g. [null] + [null] = [null] if dtype=int32 but not supported if dtype=null)

sort_indices should support null arrays.

Some functions do forward null arrays:

• unique

Some functions output a non-null type given null inputs

• is_null (=> boolean)
• is_valid (=> boolean)
• value_counts (=> struct)
• dictionary_encode (=> dictionary<null>)
• count (=> int64)

Some functions throw an error other than "not implemented"

• list_parent_indices

Issue Links

• links to: GitHub Pull Request #11255