[
https://issues.apache.org/jira/browse/ARROW-17965?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17614229#comment-17614229
]
Yaron Gvili commented on ARROW-17965:
-------------------------------------
Not sure how much context you are asking for. I bumped into this when working
on ordered aggregation. The test cases I worked out for this involved chunked
arrays (but not an `ExecPlan`) that triggered the described failure that led to
this issue.
I suppose it is possible to split to multiple `ExecBatch` instances, but I
think this is not convenient for the user and potentially less efficient, e.g.,
in the context of streaming it is more efficient to consume a large `ExecBatch`
with a chunked array than to consume multiple smaller `ExecBatch` instances
with the same data.
I believe the proposed code is simple enough to warrant a review. Let me know
your thoughts.
> [C++] ExecBatch support for ChunkedArray values
> -----------------------------------------------
>
> Key: ARROW-17965
> URL: https://issues.apache.org/jira/browse/ARROW-17965
> Project: Apache Arrow
> Issue Type: New Feature
> Components: C++
> Reporter: Yaron Gvili
> Assignee: Yaron Gvili
> Priority: Major
>
> Currently, `ExecBatch` does not handle chunked arrays when printing or
> slicing. The code assumes that if a value is not a scalar then it is an
> array, and so will fail on chunked array values.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
