[ 
https://issues.apache.org/jira/browse/LUCENE-1261?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12586880#action_12586880
 ] 

Karl Wettin commented on LUCENE-1261:
-------------------------------------

Hi John,

see LUCENE-1260

    karl

> Impossible to use custom norm encoding/decoding
> -----------------------------------------------
>
>                 Key: LUCENE-1261
>                 URL: https://issues.apache.org/jira/browse/LUCENE-1261
>             Project: Lucene - Java
>          Issue Type: Bug
>          Components: Query/Scoring
>    Affects Versions: 2.3.1
>         Environment: All
>            Reporter: John Adams
>
> Although it is possible to override methods encodeNorm and decodeNorm in a 
> custom Similarity class, these methods are not actually used by the query 
> processing and scoring functions, not by the indexing functions. The relevant 
> Lucene classes all call "Similarity.decodeNorm" rather than 
> "similarity.decodeNorm", i.e. the norm encoding/decoding is fixed to use that 
> of the base Similarity class. Also index writing classes such as 
> DocumentWriter use "Similarity.decodeNorm" rather than 
> "similarity.decodeNorm", so we are stuck with the 3 bit mantissa encoding 
> implemented by SmallFloat.floatToByte315 and SmallFloat.byte315ToFloat.
> This is very restrictive and annoying, since in practice many users would 
> prefer an encoding that allows finer distinctions for boost and normalisation 
> factors close to 1.0. For example. SmallFloat.floatToByte52 uses 5 bits of 
> mantissa, and this would be of great help in distinguishing much better 
> between subtly different lengthNorms and FieldBoost/DocumentBoost values.
> It hsould be easy to fix this by changing all instances of 
> "Similarity.decodeNorm" and "Similarity.encodeNorm" to 
> "similarity.decodeNorm" and "similarity.encodeNorm" in the Lucene code (there 
> are only a few of each).

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]

Reply via email to