[
https://issues.apache.org/jira/browse/HIVE-7262?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14048142#comment-14048142
]
Matt McCline commented on HIVE-7262:
------------------------------------
In talking to Harish, the issue is we should not try to vectorize "pure" or
"true' table functions like NOOP.
We should only vectorize PTF when it is strictly for windowing only. Then, the
automatically added virtual columns like FILENAME and BLOCKOFFSET will get
pruned away very early and not be an issue.
Separately, there is another issue HIVE-5570 Handle virtual columns and schema
evolution in vector code path when someone actually want one of those virtual
columns.
> Partitioned Table Function (PTF) query fails on ORC table when attempting to
> vectorize
> --------------------------------------------------------------------------------------
>
> Key: HIVE-7262
> URL: https://issues.apache.org/jira/browse/HIVE-7262
> Project: Hive
> Issue Type: Bug
> Reporter: Matt McCline
> Assignee: Matt McCline
>
> In ptf.q, create the part table with STORED AS ORC and SET
> hive.vectorized.execution.enabled=true;
> Queries fail to find BLOCKOFFSET virtual column during vectorization and
> suffers an exception.
> ERROR vector.VectorizationContext
> (VectorizationContext.java:getInputColumnIndex(186)) - The column
> BLOCK__OFFSET__INSIDE__FILE is not in the vectorization context column map.
> Jitendra pointed to the routine that returns the VectorizationContext in
> Vectorize.java needing to add virtual columns to the map, too.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
