[
https://issues.apache.org/jira/browse/LUCENE-6299?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Michael McCandless updated LUCENE-6299:
---------------------------------------
Attachment: LUCENE-6299.patch
Thanks Rob, here's a new patch merged with yours.
I added tests + fixes for the other issues, and discovered a
ridiculous (index corruption) bug: on init IndexWriter does not set
its pendingNumDocs to the number of docs already in the index.
So this check is completely broken and it's trivial to make a corrupt
index today! Grrr... I added a test and fixed that.
I also added a test for addIndexes(CodecReader[]) too, fixed
BaseCompositeReader to throw CorruptIndexException when it's a
DirectoryReader, changed the exception to IllegalArgumentException,
changed reserveDocs to just take a long and simplified the checking in
addIndexes (just calls reserveDocs).
Given that this is an index corruption issue I think we should
back-port for 4.10.4, but on back-port I would only do the minimal bug
fixes here.
> IndexWriter's enforcement of 2.1B doc limits is buggy
> -----------------------------------------------------
>
> Key: LUCENE-6299
> URL: https://issues.apache.org/jira/browse/LUCENE-6299
> Project: Lucene - Core
> Issue Type: Bug
> Reporter: Michael McCandless
> Assignee: Michael McCandless
> Fix For: 4.10.4, 5.0, Trunk, 5.1
>
> Attachments: LUCENE-6299.patch, LUCENE-6299_addIndexes.patch
>
>
> E.g. if you pass an already > 2.1B docs to either addIndexes, it can fail to
> enforce properly.
> IW's private reserveDocs should refuse to accept negative values.
> IW.deleteAll fails to set the pendingNumDocs to 0.
--
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
