[
https://issues.apache.org/jira/browse/LUCENE-6019?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14180413#comment-14180413
]
Michael McCandless commented on LUCENE-6019:
--------------------------------------------
Grrrrrr!
DefaultIndexingChain has this nice (wrong) comment:
{noformat}
// This will throw an exc if the caller tried to
// change the DV type for the field:
fp.fieldInfo.setDocValuesType(dvType);
if (hasDocValues == false) {
fieldInfos.globalFieldNumbers.setDocValuesType(fp.fieldInfo.number,
fp.fieldInfo.name, dvType);
}
{noformat}
... it's wrong because in FieldInfos.setDocValuesType we have this:
{noformat}
synchronized void setDocValuesType(int number, String name, DocValuesType
dvType) {
assert containsConsistent(number, name, dvType);
docValuesType.put(name, dvType);
}
{noformat}
So indeed it only throws exc when assertions are enabled.
bq. NOTE: this only fails if you run without assertions which I think lucene
should do in CI once in a while too.
+1, I think we should do this first. We have existing tests that would have
caught this bug had we already done this...
> IndexWriter allows to add same field with different docvlaues type
> -------------------------------------------------------------------
>
> Key: LUCENE-6019
> URL: https://issues.apache.org/jira/browse/LUCENE-6019
> Project: Lucene - Core
> Issue Type: Bug
> Components: core/index
> Affects Versions: 4.10.1
> Reporter: Simon Willnauer
> Priority: Critical
> Fix For: 4.10.2, 5.0
>
>
> IndexWriter checks if the DV types are consitent in multiple places but if
> due to some problems in Elasticsearch users where able to add the same field
> with different DV types causing merges to fail. Yet I was able to reduce this
> to a lucene testcase but I was puzzled since it always failed. Yet, I had to
> run it without assertions and that cause the bug to happen. I can add field
> foo with BINARY and SORTED_SET causing a merge to fail. Here is a gist
> https://gist.github.com/s1monw/8707f924b76ba40ee5f3 /
> https://github.com/elasticsearch/elasticsearch/issues/8009
> While this is certainly a problem in Elasticsearch Lucene also allows to
> corrupt an index due to user error which I think should be prevented. NOTE:
> this only fails if you run without assertions which I think lucene should do
> in CI once in a while too.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
