[ 
https://issues.apache.org/jira/browse/LUCENE-8574?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16698188#comment-16698188
 ] 

Robert Muir commented on LUCENE-8574:
-------------------------------------

Hopefully this is enough? If we have to handle the case where advanceExact() is 
called over and over again we have to add 'currentDoc' back too so it works 
more like jack's original cache. But we should avoid adding abstractions here 
that will make things harder on the compiler.

> ExpressionFunctionValues should cache per-hit value
> ---------------------------------------------------
>
>                 Key: LUCENE-8574
>                 URL: https://issues.apache.org/jira/browse/LUCENE-8574
>             Project: Lucene - Core
>          Issue Type: Bug
>    Affects Versions: 7.5, master (8.0)
>            Reporter: Michael McCandless
>            Assignee: Robert Muir
>            Priority: Major
>         Attachments: LUCENE-8574.patch
>
>
> The original version of {{ExpressionFunctionValues}} had a simple per-hit 
> cache, so that nested expressions that reference the same common variable 
> would compute the value for that variable the first time it was referenced 
> and then use that cached value for all subsequent invocations, within one 
> hit.  I think it was accidentally removed in LUCENE-7609?
> This is quite important if you have non-trivial expressions that reference 
> the same variable multiple times.
> E.g. if I have these expressions:
> {noformat}
> x = c + d
> c = b + 2 
> d = b * 2{noformat}
> Then evaluating x should only cause b's value to be computed once (for a 
> given hit), but today it's computed twice.  The problem is combinatoric if b 
> then references another variable multiple times, etc.
> I think to fix this we just need to restore the per-hit cache?
>  



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to