You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "David Smiley (JIRA)" <ji...@apache.org> on 2015/04/03 19:48:53 UTC
[jira] [Commented] (SOLR-6692) hl.maxAnalyzedChars should apply
cumulatively on a multi-valued field
[ https://issues.apache.org/jira/browse/SOLR-6692?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14394764#comment-14394764 ]
David Smiley commented on SOLR-6692:
------------------------------------
bq. Furthermore, DefaultSolrHighligher.doHighlightingByHighlighter should exit early from it's field value loop if it reaches hl.snippets.
Actually it can never do that because it takes the top hl.snippets from every value and then takes the top hl.snippets of that. Anyway; there's multiple mechanisms to exit early now -- hl.maxAnalyzedChars, and both hl.multiValued* options.
> hl.maxAnalyzedChars should apply cumulatively on a multi-valued field
> ---------------------------------------------------------------------
>
> Key: SOLR-6692
> URL: https://issues.apache.org/jira/browse/SOLR-6692
> Project: Solr
> Issue Type: Improvement
> Components: highlighter
> Reporter: David Smiley
> Assignee: David Smiley
> Fix For: 5.2
>
> Attachments: SOLR-6692_hl_maxAnalyzedChars_cumulative_multiValued,_and_more.patch
>
>
> in DefaultSolrHighlighter, the hl.maxAnalyzedChars figure is used to constrain how much text is analyzed before the highlighter stops, in the interests of performance. For a multi-valued field, it effectively treats each value anew, no matter how much text it was previously analyzed for other values for the same field for the current document. The PostingsHighlighter doesn't work this way -- hl.maxAnalyzedChars is effectively the total budget for a field for a document, no matter how many values there might be. It's not reset for each value. I think this makes more sense. When we loop over the values, we should subtract from hl.maxAnalyzedChars the length of the value just checked. The motivation here is consistency with PostingsHighlighter, and to allow for hl.maxAnalyzedChars to be pushed down to term vector uninversion, which wouldn't be possible for multi-valued fields based on the current way this parameter is used.
> Interestingly, I noticed Solr's use of FastVectorHighlighter doesn't honor hl.maxAnalyzedChars as the FVH doesn't have a knob for that. It does have hl.phraseLimit which is a limit that could be used for a similar purpose, albeit applied differently.
> Furthermore, DefaultSolrHighligher.doHighlightingByHighlighter should exit early from it's field value loop if it reaches hl.snippets, and if hl.preserveMulti=true
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org