You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@calcite.apache.org by "Rui Wang (Jira)" <ji...@apache.org> on 2019/11/21 18:16:00 UTC
[jira] [Commented] (CALCITE-3529) TIMESTAMPDIFF function may
overflow when handle unit of MICROSECOND/NANOSECOND
[ https://issues.apache.org/jira/browse/CALCITE-3529?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16979476#comment-16979476 ]
Rui Wang commented on CALCITE-3529:
-----------------------------------
I had similar problem in BeamSQL, and in Beam community we had several discussions there. In Java world, due to wide-used Joda library, usually "timestamp" only supports up to millisecond precision. In order to support beyond millis, essentially we need extra storage to save sub-millis, which might be just another int.
Back to Calcite sql function implementation. I believe currently relevant implementation converts "timestamp" to long and assume it's millisecond.
> TIMESTAMPDIFF function may overflow when handle unit of MICROSECOND/NANOSECOND
> ------------------------------------------------------------------------------
>
> Key: CALCITE-3529
> URL: https://issues.apache.org/jira/browse/CALCITE-3529
> Project: Calcite
> Issue Type: Bug
> Components: core
> Affects Versions: 1.21.0
> Reporter: Zhenghua Gao
> Priority: Major
>
> For unit of NANOSECOND, the TIMESTAMPDIFF function returns a BIGINT, which may overflow since BIGINT can't cover the huge range of nanosecond.
> For unit of MICROSECOND, the TIMESTAMPDIFF function returns a INTEGER, which may overflow too.
> TIMESTAMPDIFF(MICROSECOND, TIMESTAMP '9999-12-31 23:59:59.999', TIMESTAMP '0001-01-01 00:00:00.000') should returns '315537897599999000' [1], calcite returns '879764032'.
>
> TIMESTAMPDIFF(NANOSECOND, TIMESTAMP '9999-12-31 23:59:59.999', TIMESTAMP '0001-01-01 00:00:00.000') should returns '315537897599999000*000*' which is bigger than LONG.MAX_VALUE, calcite returns '-1943248345937622528'.
>
> My suggestion is:
> # Change the return type to BIGINT for unit of MICROSECOND
> # Disable support for unit of NANOSECOND or give a RISK message in the document of TIMESTAMPDIFF
> [1] the value is calculated by MySQL 5.6 from [http://sqlfiddle.com/]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)