[
https://issues.apache.org/jira/browse/LUCENE-8296?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16465190#comment-16465190
]
Simon Willnauer commented on LUCENE-8296:
-----------------------------------------
I think this is mostly a relict from before I started refactoring
ReadersAndUpdates. I would love to even go further and down the road make the
returned Bits instance immutable. I think we should have a very very simple
base class that FixedBitSet can extend that knows how to read from the array.
This way we know nobody ever mutates it. Today you can just cast the liveDocs
from a NRT reader and change it's private instance. I am going to look into
this unless anybody beats me.
One thing that I am feel is missing is an explicit test that the returned bits
don't change in subsequent modifications.
+1 to the change!
> PendingDeletes shouldn't write to live docs that it shared
> ----------------------------------------------------------
>
> Key: LUCENE-8296
> URL: https://issues.apache.org/jira/browse/LUCENE-8296
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Adrien Grand
> Priority: Minor
> Attachments: LUCENE-8296.patch
>
>
> PendingDeletes has a markAsShared mechanism that allow to make sure that the
> latest livedocs are not going to receive more updates. But it is not always
> used, and I was able to verify that in some cases we end up with readers
> whose live docs disagree with the number of deletes. Even though this might
> not be causing bugs, it feels dangerous to me so I think we should consider
> always marking live docs as shared in #getLiveDocs.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
