[
https://issues.apache.org/jira/browse/LUCENE-6500?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14558874#comment-14558874
]
Uwe Schindler commented on LUCENE-6500:
---------------------------------------
The anonymous subclassing was added to explicitely don't call closelisteners on
the "synthetic" readers. The same code is there 2 times... (for composites and
for atomics).
Closing all subreaders is done in ParallelCompositeReader's doClose() where it
iterates over all "real childs" which are passed in ctor.
I think the new problem is that (as you say) the close listeners may be
registered on the synthetic leaves... I have to think about how to better
handle this, with your patch it may now happen that a reader is closed multiple
times. Which is not a problem in your test, but could be in reality.
In addition there is now (after your patch) an assymetry between atomic leaves
and composite sub readers... Does't the same problem apply to atomic readers?
Unfrotuentaly my Eclipse strikes like Deutsche Bahn and says "GC overhead
limit". I am unable to dig...
> ParallelCompositeReader does not always call closed listeners
> -------------------------------------------------------------
>
> Key: LUCENE-6500
> URL: https://issues.apache.org/jira/browse/LUCENE-6500
> Project: Lucene - Core
> Issue Type: Bug
> Reporter: Adrien Grand
> Assignee: Adrien Grand
> Priority: Minor
> Attachments: LUCENE-6500.patch
>
>
> CompositeParallelReader misses to call closed listeners when the reader which
> is provided at construction time does not wrap leaf readers directly, such as
> a multi reader over directory readers.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
