[
https://issues.apache.org/jira/browse/LUCENE-6250?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14324071#comment-14324071
]
Adrien Grand commented on LUCENE-6250:
--------------------------------------
bq. fails because SpanScorer.score() is called while it is still at docID -1,
and I am stuck trying to figure this out.
I just tried out the seed and I think I understand why: approximations are
supposed to be a view and advancing the approximation should advance the scorer
and vice-versa. Yet SpanScorer caches the current doc ID (the 'doc' variable)
so when you advance the approximation, SpanScorer.doc is still -1 so when you
call score() it tries to get the norm for doc ID -1.
bq. When I started investigating this test case I was confused by BooleanWeight
creating an unused SpansScorer while trying to create a BooleanScorer. Shall I
open another issue to separate the condition to actually use a BooleanScorer
from the creation of the BooleanScorer (in BooleanWeight)?
I agree this would be nicer!
bq. Can the TestFilterSpans class be removed? It only checks that all abstract
Span methods are implemented.
I never used spans, but looking at
https://issues.apache.org/jira/browse/LUCENE-5933 this seems important in order
to make sure that FilterSpans delegate everything from the wrapped instance?
> Approximations on Spans
> -----------------------
>
> Key: LUCENE-6250
> URL: https://issues.apache.org/jira/browse/LUCENE-6250
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Paul Elschot
> Attachments: ApproxSpans-20150216a.patch
>
>
> Approximate spans using existing conjunction/disjunction approximations
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
