[
https://issues.apache.org/jira/browse/HIVE-15624?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sergey Shelukhin updated HIVE-15624:
------------------------------------
Description:
Noticed while fixing HIVE-15558
Schema evolution non-vectorized tests, or maybe their previous version on 2.1,
in some instances have nulls in query results for values that are padded due to
being char, when converting them to integers (e.g. '50000' or '-9999' are
converted, but '17 ' cannot be converted and becomes null).
HIVE-15558, while restoring proper padding to char, also made vectorization
consistent with non-vectorized Hive in making them null, however that is not a
desirable behavior.
I've tried it out in simple queries with casts, and it doesn't happen ('17 '
of type char(5) is cast to int as expected).
I think we should fix this behavior for both vectorized and non-vectorized case.
was:
Noticed while fixing HIVE-15558
Schema evolution non-vectorized tests, or maybe their previous version on 2.1,
in some instances have nulls in query results for values that are padded due to
being char (e.g. '50000' or '-9999' are converted, but '17 ' cannot be
converted and becomes null).
HIVE-15558, while restoring proper padding to char, also made vectorization
consistent with non-vectorized Hive in making them null, however that is not a
desirable behavior.
I've tried it out in simple queries with casts, and it doesn't happen ('17 '
of type char(5) is cast to int as expected).
I think we should fix this behavior for both vectorized and non-vectorized case.
> sometimes (with schema evolution?) padded char values cannot be converted to
> integers
> -------------------------------------------------------------------------------------
>
> Key: HIVE-15624
> URL: https://issues.apache.org/jira/browse/HIVE-15624
> Project: Hive
> Issue Type: Bug
> Reporter: Sergey Shelukhin
>
> Noticed while fixing HIVE-15558
> Schema evolution non-vectorized tests, or maybe their previous version on
> 2.1, in some instances have nulls in query results for values that are padded
> due to being char, when converting them to integers (e.g. '50000' or '-9999'
> are converted, but '17 ' cannot be converted and becomes null).
> HIVE-15558, while restoring proper padding to char, also made vectorization
> consistent with non-vectorized Hive in making them null, however that is not
> a desirable behavior.
> I've tried it out in simple queries with casts, and it doesn't happen ('17
> ' of type char(5) is cast to int as expected).
> I think we should fix this behavior for both vectorized and non-vectorized
> case.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
