[
https://issues.apache.org/jira/browse/LUCENE-7080?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15186829#comment-15186829
]
Simon Willnauer edited comment on LUCENE-7080 at 3/9/16 9:26 AM:
-----------------------------------------------------------------
[~mariusneo], the lucene test framework uses randomized runner which takes care
of the reproducibility of our PRNG. The framework / test uses a random seed to
initialize the PRNG and that allows use to draw random number like this in a
reproducible fashion. The problem here is that different JDKs can have
different hashing schemes for the hash set and that causes different iteration
order which in-turn breaks the reproducibility.
was (Author: simonw):
[~mariusneo] the lucene test framework uses randomized runner which takes care
of the reproducibility of our PRNG. The framework / test uses a random seed to
initialize the PRNG and that allows use to draw random number like this in a
reproducible fashion. The problem here is that different JDKs can have
different hashing schemes for the hash set and that causes different iteration
order which in-turn breaks the reproducibility.
> MockDirectoryWrapper relies on HashSet iteration order
> ------------------------------------------------------
>
> Key: LUCENE-7080
> URL: https://issues.apache.org/jira/browse/LUCENE-7080
> Project: Lucene - Core
> Issue Type: Bug
> Components: general/test
> Affects Versions: 5.5, 6.0
> Reporter: Simon Willnauer
> Assignee: Simon Willnauer
>
> MDW relies on HashSet iteration order in
> {code}
> public synchronized void corruptFiles(Collection<String> files) throws
> IOException {
> // Must make a copy because we change the incoming unsyncedFiles
> // when we create temp files, delete, etc., below:
> for(String name : new ArrayList<>(files)) { // <<<<< this should be sorted
> int damage = randomState.nextInt(6);
> {code}
> this causes reproducibility issues when files get corrupted.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
