[
https://issues.apache.org/jira/browse/HIVE-19681?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16487858#comment-16487858
]
Matt McCline commented on HIVE-19681:
-------------------------------------
Oh, it is a two-step generation process:
1) Generate the random rows.
2) Then, based on more random numbers, decide how to arrange the random rows
into interesting batches. The "trick" is to do some modification of the
original random rows like copying repeated NULLs or repeated values down
columns for the interesting batch being configured.
The interesting batch configuration is kept and used to generate the batches
from the modified random rows. So, the data is the same for row-mode and
vector-mode.
> Fix TestVectorIfStatement
> -------------------------
>
> Key: HIVE-19681
> URL: https://issues.apache.org/jira/browse/HIVE-19681
> Project: Hive
> Issue Type: Test
> Components: Vectorization
> Affects Versions: 3.1.0, 4.0.0
> Reporter: Vihang Karajgaonkar
> Assignee: Vihang Karajgaonkar
> Priority: Major
>
> {{TestVectorIfStatement}} generates interesting batches (injection of random
> repeating null column values and repeating non-null values) when evaluating
> the vectorized expressions. But the modification of random rows is done after
> the row mode is evaluated. Hence it is likely that comparison results will
> fail. I am not sure how its working in the first place.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
