[
https://issues.apache.org/jira/browse/SOLR-7787?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14626799#comment-14626799
]
Dawid Weiss commented on SOLR-7787:
-----------------------------------
Well, I thought lack of initial feedback was lazy consensus :)
And seriously -- it can go either way. My original intention was to update HPPC
which is duplicated in Solr and the clustering contrib. These have to be
consistent. From there I observed that:
1) Solr uses HPPC in a small number of classes,
2) fastutil is present in solr's lib, but it's not used in any classes, it is
just a transitive dependency from hll,
3) fastutil is much larger than HPPC (roughly 15x).
I am really ok with any option. I decided to cut fastutil out because it's just
a much larger library (of which almost nothing is practically used in Solr).
Also, hll's implementation uses fastutil iterators very inefficiently (causing
intermediate autoboxing on every value) so I thought I'd take a stab at
improving that while converting to HPPC.
I can also leave everything as-is, really, but the patch I have locally seems
like a nice improvement on its own (and could be perhaps pruned further to get
rid of unnecessary stuff like serialization, etc.).
> Fork HyperLogLog and remove fastutil dependency
> -----------------------------------------------
>
> Key: SOLR-7787
> URL: https://issues.apache.org/jira/browse/SOLR-7787
> Project: Solr
> Issue Type: Task
> Reporter: Dawid Weiss
> Assignee: Dawid Weiss
> Priority: Trivial
> Fix For: 5.3, Trunk
>
>
> So fastutil is now part of Solr's distribution (because the stats component
> uses hyperloglog library, which in turn requires fastutil). I looked at the
> actual uses of fastutil and only java-hll uses it (and only a few classes).
> I've created a fork that uses HPPC instead (also randomized all tests, they
> pass). Since it's a relatively simple package I think it could be forked and
> imported into Solr's codebase entirely. I'd make a pull request but I see
> Hoss also created a few comments/ PRs and none of them received any
> attention; the project seems to be stale or dead?
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
