[
https://issues.apache.org/jira/browse/LUCENE-6427?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14497795#comment-14497795
]
Luc Vanlerberghe commented on LUCENE-6427:
------------------------------------------
* I moved the "Depends on the ghost bits being clear!" line from the doc
comments to a regular comment, except for the constructor (where the fact is
verified if assertions are enabled)
* I renamed checkIfEmpty to scanIsEmpty and updated comments.
I don't see the problem with having a scanIsEmpty method available for those
willing to look for it. It probably will make a performance difference for
small bitsets because it avoids the overhead of two method calls and the
numberOfTrailingZeros call in nextSetBit on the first non-zero long.
LUCENE-5856 shows that even removing a useless & 0x3f from *BitSet.get and
company can have a noticeable effect (albeit probably in tight inner loops)
By the way, I did remove a useless & 0x3f from FixedBitSet.flip in this patch
as well
> BitSet fixes - assert on presence of 'ghost bits' and others
> ------------------------------------------------------------
>
> Key: LUCENE-6427
> URL: https://issues.apache.org/jira/browse/LUCENE-6427
> Project: Lucene - Core
> Issue Type: Bug
> Components: core/other
> Reporter: Luc Vanlerberghe
>
> Fixes after reviewing org.apache.lucene.util.FixedBitSet, LongBitSet and
> corresponding tests:
> * Some methods rely on the fact that no bits are set after numBits (what I
> call 'ghost' bits here).
> ** cardinality, nextSetBit, intersects and others may yield wrong results
> ** If ghost bits are present, they may become visible after ensureCapacity is
> called.
> ** The tests deliberately create bitsets with ghost bits, but then do not
> detect these failures
> * FixedBitSet.cardinality scans the complete backing array, even if only
> numWords are in use
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
