Let's do it as a separate patch? The clone patch is already scaring
me enough ;)
Maybe we can allow creation of a read-only IndexReader during open?
Ie, a new open method that takes a "boolean readOnly".
I had been previously thinking that we'd get to a read-only
IndexReader by slurping all write operations into IndexWriter, but, it
seems like that migration will take quite some time....
Mike
Jason Rutherglen wrote:
Should we do a read-only reader as part of the clone patch? Sounds
very similar. Also sounds like the delegate model will work best
instead of creating a SegmentReader subclass.
On Thu, Jul 3, 2008 at 7:55 AM, Michael McCandless <[EMAIL PROTECTED]
> wrote:
But I think you still need to synchronize, because the first thread
that does a deletion needs to create the deletedDocs BitVector and
others need to wait while that's happening?
I think [eventually] getting to a read-only reader is the best
approach (https://issues.apache.org/jira/browse/LUCENE-1030).
Mike
Jason Rutherglen wrote:
A possible solution to the IndexReader.isDeleted synchronization is
to have a delegate class that is different depending on the Java
version. For Java 1.4 the class defaults to the synchronization
used now. A Java 1.5 version would use a volatile deletedDocs
attribute. This would provide a backwards compatible solution and a
solution for Java versions with volatile working that can avoid the
use of synchronized.
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
