[
https://issues.apache.org/jira/browse/LUCENE-9207?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17030680#comment-17030680
]
Michael Gibney commented on LUCENE-9207:
----------------------------------------
I think the special logic building SpanQueries for the slop=0 case was left in
place by LUCENE-8531 because the resulting behavior is functionally identical
to the MultiPhraseQuery approach, and SpanQueries for slop=0 are more efficient
(potentially _vastly_ more efficient) than the exponential expansion that can
result from MultiPhraseQuery over graph TokenStreams (e.g., for bigrams,
synonyms, wdgf, etc.).
[~romseygeek], do you think the code simplification is worth the potential
performance hit for the {{slop=0}} case? [~jim.ferenczi], [~sarowe],
[~uschindler], I'm curious for your perspectives (having been involved in the
discussion around LUCENE-8531). For heavily branching token streams (e.g.,
bigrams, certain tYpEs 0f 1nPuT to common WGDF configurations), the performance
impact is substantial. I know of (and in fact personally know) many people who
have been bitten by this in the form of SOLR-13336; but the underlying
performance issue is not Solr-specific and is not directly addressed by the fix
for SOLR-13336, which simply restores Lucene's maxBooleanClauses threshold for
shortcircuiting individual queries.
FWIW, I think LUCENE-7398 is a bit of a red herring here; I'm shooting from the
hip a bit, but I'm 90% confident that the LUCENE-7398 issues don't affect the
slop=0 case for _query_-time graph TokenStreams; and to the extent that they
affect _index_-time graph TokenStreams, they affect SpanQueries and
MultiPhraseQuery equally (that's a whole separate question!).
> Don't build SpanQuery in QueryBuilder
> -------------------------------------
>
> Key: LUCENE-9207
> URL: https://issues.apache.org/jira/browse/LUCENE-9207
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Alan Woodward
> Assignee: Alan Woodward
> Priority: Major
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Subtask of LUCENE-9204. QueryBuilder currently has special logic for graph
> phrase queries with no slop, constructing a spanquery that attempts to follow
> all paths using a combination of OR and NEAR queries. Given the known bugs
> in this type of query (LUCENE-7398) and that we would like to move span
> queries out of core in any case, we should remove this logic and just build a
> disjunction of phrase queries, one phrase per path.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
