lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Uwe Schindler (JIRA)" <>
Subject [jira] Updated: (LUCENE-2719) Re-add SorterTemplate and use it to provide fast ArraySorting and replace BytesRefHash sorting
Date Fri, 22 Oct 2010 21:50:19 GMT


Uwe Schindler updated LUCENE-2719:

    Attachment: LUCENE-2719.patch

Attached you find the patch. Robert offered to do benchmarks with Automaton.

The patch can be applied to a clean checkout, you no longer need to svn copy old SorterTemplate,
as this is a almost complete rewrite.

This patch removes the CHANGES.txt entry for 3.x, as it readds the class. If we don't merge
this to 3.x, the CHANGES should be reverted. As Lucene uses Arrays.sort(Object[]) which is
slow at other places, I recommend to add it also to 3.x.

Please test the stuff with large -Dtests.multiplier! Maybe also verify my modified quickSort!

> Re-add SorterTemplate and use it to provide fast ArraySorting and replace BytesRefHash
> ----------------------------------------------------------------------------------------------
>                 Key: LUCENE-2719
>                 URL:
>             Project: Lucene - Java
>          Issue Type: Improvement
>    Affects Versions: 3.1
>            Reporter: Uwe Schindler
>            Assignee: Uwe Schindler
>             Fix For: 3.1, 4.0
>         Attachments: LUCENE-2719.patch
> This patch adds back an optimized and rewritten SorterTemplate back to Lucene (removed
after release of 3.0). It is of use for several components:
> - Automaton: Automaton needs to sort States and other things. Using Arrays.sort() is
slow, because it clones internally to ensure stable search. This component is much faster.
This patch adds Arrays.sort() replacements in ArrayUtil that work with natural order or using
a Comparator<?>. You can choose between quickSort and mergeSort.
> - BytesRefHash uses another QuickSort algorithm without insertionSort for very short
ord arrays. This class uses SorterTemplate to provide the same with insertionSort fallback
in a very elegant way. Ideally this class can be used everywhere, where the sort algorithm
needs to be separated from the underlying data and you can implement a swap() and compare()
function (that get slot numbers instead of real values). This also applies to Solr (Yonik?).
> SorterTemplate provides quickSort and mergeSort algorithms. Internally for short arrays,
it automatically chooses insertionSort (like JDK's Arrays). The quickSort algorith was copied
modified from old BytesRefHash. This new class only shares MergeSort with the original CGLIB
SorterTemplate, which is no longer maintained.

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:
For additional commands, e-mail:

View raw message