[
https://issues.apache.org/jira/browse/PARQUET-467?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15125213#comment-15125213
]
Wes McKinney commented on PARQUET-467:
--------------------------------------
Yes:
* BIT_PACKED uses {{BitReader}}
* RLE uses {{RleDecoder}}
The Impala code has a LevelDecoder class extending RleDecoder which calls into
BitReader where appropriate. I would prefer to create a clean abstraction (i.e.
do not use inheritance) that doesn't conflate the two encoding styles (which
are different: BIT_PACKED is not RLE encoding, even though the RLE encoding
uses bitpacking to save space). I would be happy to take on these JIRAs if it
is not clear, please let me know.
> Implement and test BIT_PACKED level encoding / decoding
> --------------------------------------------------------
>
> Key: PARQUET-467
> URL: https://issues.apache.org/jira/browse/PARQUET-467
> Project: Parquet
> Issue Type: Bug
> Components: parquet-cpp
> Reporter: Wes McKinney
>
> While RLE is the preferred encoding format (and BIT_PACKED is deprecated in
> Parquet 2.0), we will need to support this encoding format for legacy Parquet
> files that use it. As part of this JIRA we will verify round-tripping levels
> to this encoding format.
> See also PARQUET-462
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
