[
https://issues.apache.org/jira/browse/HIVE-5563?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13797062#comment-13797062
]
Owen O'Malley commented on HIVE-5563:
-------------------------------------
RCFile doesn't have a vectorized reader so it wasn't broken by HIVE-5391.
HIVE-4113 changed both RCFile and ORC.
There certainly does need to be more testcases for the interface between Hive
and the file formats although testing the effects of column projection are
pretty tricky.
The current InputFormat API is a mess with a lot of special cases and hidden
settings in the configuration. It seriously needs to be redesigned at some
point.
> Skip reading columns in ORC for count(*)
> ----------------------------------------
>
> Key: HIVE-5563
> URL: https://issues.apache.org/jira/browse/HIVE-5563
> Project: Hive
> Issue Type: Bug
> Components: File Formats
> Reporter: Owen O'Malley
>
> With HIVE-4113, the semantics of ColumnProjectionUtils.getReadColumnIds was
> fixed so that an empty list means no columns instead of all columns. (Except
> the caveat of the override of ColumnProjectionUtils.isReadAllColumns.)
> However, ORC's reader wasn't updated so it still reads all columns.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
