[ARROW-16819] [C++] arrow::compute::CallFunction needs a batch length for nullary functions

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 9.0.0
• Component/s: C++
• External issue URL: https://github.com/apache/arrow/issues/32149
• Assignee: Wes McKinney
• Reporter: Wes McKinney
• Created: 12/Jun/22 13:08
• Updated: 11/Jan/23 11:46
• Resolved: 13/Jun/22 16:12

Description

This is a design deficiency in CallFunction. If a function is nullary (zero input arguments), the execution machinery has no way to determine the output length from an empty vector of datums. We should add CallFunction variants based on ExecBatch and ExecSpan (from ARROW-16755) and figure out how to deal with the split between scalar (where inputs must all be the same length or scalar) and vector functions (whose input arguments can have different length)

Issue Links

• relates to: ARROW-16755 [C++] Improve array expression and kernel evaluation performance on small inputs
• relates to: ARROW-16756 [C++] Introduce initial ArraySpan, ExecSpan non-owning / shared_ptr-free data structures for kernel execution, refactor scalar kernels