[
https://issues.apache.org/jira/browse/ARROW-17484?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17583015#comment-17583015
]
Weston Pace commented on ARROW-17484:
-------------------------------------
This is an invalid plan. The "sum" function is defined in Substrait as follows:
{noformat}
- name: "sum"
description: Sum a set of values. The sum of zero elements yields null.
impls:
- args:
- options: [ SILENT, SATURATE, ERROR ]
required: false
- value: i8
nullability: DECLARED_OUTPUT
decomposable: MANY
intermediate: i64?
return: i64?
{noformat}
The output type is always {{i64?}}. It is not legal in Substrait to have two
implementations that differ only by return type. This is very similar to
function overloading which has a similar constraint. So it would be illegal to
have an implementation that accepted {{i32}} and returns {{i32}}. So this plan
that you've created can never be legal.
As a consequence, the output type can always be calculated from the input types
and the function name. So the {{outputType}} in the plan is purely
informational and the consumer is free to ignore it.
> [C++] Substrait to Arrow Aggregate doesn't take the provided Output Type for
> aggregates
> ---------------------------------------------------------------------------------------
>
> Key: ARROW-17484
> URL: https://issues.apache.org/jira/browse/ARROW-17484
> Project: Apache Arrow
> Issue Type: Bug
> Components: C++
> Reporter: Vibhatha Lakmal Abeykoon
> Assignee: Vibhatha Lakmal Abeykoon
> Priority: Major
>
> The current Substrait to Aggregate deserializer doesn't take the plan
> provided output type as the output type of the execution plan.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
