[
https://issues.apache.org/jira/browse/LUCENE-4161?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Adrien Grand updated LUCENE-4161:
---------------------------------
Attachment: LUCENE-4161.patch
First version of the patch.
A few things that were internal now need to be exposed, so I tried to do some
clean up:
* {{CODEC_NAME}} and CODEC_VERSION{START,CURRENT} are public,
* the format is an enum (PackedInts.Format.{PACKED,PACKED_SINGLE_BLOCK}),
* improved docs overall.
There are new factory methods get{Reader,ReaderIterator,Writer}NoHeader that do
the same as their get{Reader,ReaderIterator,Writer} counterpart, but with no
header writing/checking.
Improved performance of Reader/Mutable bulk methods (using code generation, see
http://people.apache.org/~jpountz/packed_ints.html vs.
http://people.apache.org/~jpountz/packed_ints2.html).
{{ReaderIterator}} and {{Writer}} now use the same code as
{{Reader}}/{{Mutable}} bulk methods so they are likely to be much faster too.
In addition, ReaderIterator now allows consumers to retrieve several values at
the same time.
{{Direct*}} and {{Packed*ThreeBlocks}} had a lot of duplicate code that was not
factorizable so I created scripts to generate them.
Something that might still slow down ReaderIterator (probably the most useful
class for codecs) a bit is that ReaderIterator always reads one long at a time.
Adding a method to bulk-read longs to DataInput (similarly to readBytes) might
improve performance. This probably deserves an other issue in JIRA and can be
done later.
> Make PackedInts usable by codecs
> --------------------------------
>
> Key: LUCENE-4161
> URL: https://issues.apache.org/jira/browse/LUCENE-4161
> Project: Lucene - Java
> Issue Type: Improvement
> Components: core/store
> Reporter: Adrien Grand
> Assignee: Adrien Grand
> Priority: Minor
> Attachments: LUCENE-4161.patch
>
>
> Some codecs might be interested in using
> PackedInts.{Writer,Reader,ReaderIterator} to read and write fixed-size values
> efficiently.
> The problem is that the serialization format is self contained, and always
> writes the name of the codec, its version, its number of bits per value and
> its format. For example, if you want to use packed ints to store your
> postings list, this is a lot of overhead (at least ~60 bytes per term, in
> case you only use one Writer per term, more otherwise).
> Users should be able to externalize the storage of metadata to save space.
> For example, to use PackedInts to store a postings list, one should be able
> to store the codec name, its version and the number of bits per doc in the
> header of the terms+postings list instead of having to write it once (or
> more!) per term.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
