[
https://issues.apache.org/jira/browse/LUCENE-5953?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Uwe Schindler updated LUCENE-5953:
----------------------------------
Attachment: LUCENE-5953.patch
Sorry that it took a while, I was very busy.
Attached is a patch that refactors LockFactory handling. In addition to the
discussion before, this one makes handling of locks much simplier:
- LockFactory is no longer used or exposed by the Directory implementation it
is just some factory class, that can be used by a directory to implement
makeLock(name). BaseDirectory is the only place where the lockFactory is
actually used, its not part of Directory interface.
- IndexWriter and all other components solely use Directory.makeLock(name),
which returns a Lock instance
- FSDirectory subclasses are no longer allowed to take "null" as LockFactory
instance. You have to pass one explicit (or use a constructor that is
documented to take default).
- FSLockFactory.getDefault() can be used to get a default instance for the
platform used. Currently this always returns NativeFSLockFactory.INSTANCE.
- Most lock factories are now singletons: NativeFSLockFactory.INSTANCE,
SimpleFSLockFactory.INSTANCE, NoLockFactory.INSTANCE.
I also rewrote the broken SnapShooter (the way how it locks) in Solr (but not
fix it). As it is documented as "deprecated" and no longer used, we should
remove it in Solr 5.0. I will open another issue about that.
I would like to commit this asap, because the patch may get out of sync very
fast. I will now run all test in a loop with explicit -Dtests.directory=... to
check all combinations, because disallowing "null" as LockFactory may cause
hidden bugs. So we should run this patch for a while on Jenkins :-)
> Make LockFactory final on Directory
> -----------------------------------
>
> Key: LUCENE-5953
> URL: https://issues.apache.org/jira/browse/LUCENE-5953
> Project: Lucene - Core
> Issue Type: Improvement
> Components: core/store
> Reporter: Uwe Schindler
> Assignee: Uwe Schindler
> Fix For: Trunk
>
> Attachments: LUCENE-5953.patch
>
>
> We should remove the setters for the LockFactory from Directory and make the
> field final. It is a bug to change the LockFactory after creating a
> directory, because you may break locking (if locks are currently held).
> The LockFactory should be passed on ctor only.
> The other suggestion: Should LockFactory have a directory at all? We moved
> away from having the lock separately from the index directory. This is no
> longer a supported configuration (since approx Lucene 2.9 or 3.0). I would
> like to remove the directory from LockFactory and make it part of the
> Directory only.
--
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
