Skip to content

Add PyarrowScalarUDF and convert PyScalarUDF to API recommended upstream  #907

@Michael-J-Ward

Description

@Michael-J-Ward

Is your feature request related to a problem or challenge? Please describe what you are trying to do.

Up until #906, PyScalarUDF has used datafusion_physical_expr::functions::make_scalar_function which has been deprecated since v36 and is being removed from the public API in v43.

The deprecation warning advised:

Implement your function directly in terms of ColumnarValue or use ScalarUDF instead

datafusion-python users can not currently implement ScalarUDF in that recommended way.

Describe the solution you'd like

  • Change PyScalarUDF to take functions that operate on ColumnarValue, as recommended upstream.
  • Add PyarrowScalarUDF that let's python users continue writing UDF's as they do today.

Describe alternatives you've considered
We could remove the ability to write pyarrow UDF's, forcing users to migrate as directed in the deprecation warning.

Additional context
It may be useful to review how upstream previously converted functions that use ArrayRef to ScalarUDF.

https://github.com/apache/datafusion/blob/8e8bb89a285c6fe6e8186378b7798d3d213f08f7/datafusion/physical-expr/src/functions.rs#L79-L122

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions