[
https://issues.apache.org/jira/browse/ARROW-2653?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16689424#comment-16689424
]
Wes McKinney commented on ARROW-2653:
-------------------------------------
I agree with this, with the caveat that in my experience this type of code
(hashing) is very performance sensitive. I've been surprised at the changes
that can cause 30-50% change in execution time.
Luckily, we have plenty of other hash table implementations in other projects
to look at to see how efficient our implementations are relative to comparables
> [C++] Refactor hash table support
> ---------------------------------
>
> Key: ARROW-2653
> URL: https://issues.apache.org/jira/browse/ARROW-2653
> Project: Apache Arrow
> Issue Type: Task
> Components: C++
> Affects Versions: 0.11.1
> Reporter: Antoine Pitrou
> Assignee: Antoine Pitrou
> Priority: Major
>
> Currently our hash table support is scattered in several places:
> * {{compute/kernels/hash.cc}}
> * {{util/hash.h}} and {{util/hash.cc}}
> * {{builder.cc}} (in the DictionaryBuilder implementation)
> Perhaps we should have something like a type-parametered hash table class
> (perhaps backed by non-owned memory) with several primitives:
> * decide allocation size for a given number of items
> * lookup an item
> * insert an item
> * decide whether resizing is needed
> * resize to a new memory area
> * ...
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
