You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@arrow.apache.org by "Joris Van den Bossche (Jira)" <ji...@apache.org> on 2021/05/14 13:30:00 UTC
[jira] [Comment Edited] (ARROW-12736) [C++] Eliminate unnecessary
copy in FieldPath::Get()
[ https://issues.apache.org/jira/browse/ARROW-12736?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17344592#comment-17344592 ]
Joris Van den Bossche edited comment on ARROW-12736 at 5/14/21, 1:29 PM:
-------------------------------------------------------------------------
Worth (and safe enough) to include in 4.0.1? (since this addresses (part of) a performance regression)
was (Author: jorisvandenbossche):
Worth (and safe enough) to include in 4.0.1?
> [C++] Eliminate unnecessary copy in FieldPath::Get()
> ----------------------------------------------------
>
> Key: ARROW-12736
> URL: https://issues.apache.org/jira/browse/ARROW-12736
> Project: Apache Arrow
> Issue Type: Improvement
> Components: C++
> Reporter: David Li
> Assignee: David Li
> Priority: Major
> Labels: pull-request-available
> Fix For: 5.0.0
>
> Time Spent: 17h 20m
> Remaining Estimate: 0h
>
> FieldPath::Get() uses RecordBatch::column_data which returns a vector<shared_ptr<>> by _value_ instead of by _reference_ - so with wide schemas, we end up copying and then destroying a lot of shared_ptrs. When done in a tight loop - as with dataset projection - this is a major performance pessimization.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)