[
https://issues.apache.org/jira/browse/ARROW-7754?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17030394#comment-17030394
]
Micah Kornfield commented on ARROW-7754:
----------------------------------------
I'm sorry it is that much slower it is probably due to as Antoine points out to
using a variant instead of managing some of the state manually. May I ask
which function it was?
I agree performance critical functions probably shouldn't use result (and
probably not even status as a return type).
> [C++] Result<T> is slow
> -----------------------
>
> Key: ARROW-7754
> URL: https://issues.apache.org/jira/browse/ARROW-7754
> Project: Apache Arrow
> Issue Type: Improvement
> Components: C++
> Reporter: Antoine Pitrou
> Assignee: Antoine Pitrou
> Priority: Major
>
> When converting a short performance-critical function to return a Result
> (instead of returning a Status and filling a out-parameter), I noticed a
> catastrophic performance regression (around 2x or 3x slower).
> It seems the current Result implementation is very slow, for several reasons:
> - it imposes "safety" features even in release mode, for example on the
> critical path of move operators
> - the underlying mpark variant implementation is not optimized for
> performance-critical data structures
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
