Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/22494
Sorry my mistake. I'm talking about the specific query reported at
https://issues.apache.org/jira/browse/SPARK-22036?focusedCommentId=16618104&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-16618104
, which needs to turn off `DECIMAL_OPERATIONS_ALLOW_PREC_LOSS`.
SPARK-25454 is a long-standing bug and currently we can't help users to
work around it.
My point is, to work around [this
regression](https://issues.apache.org/jira/browse/SPARK-22036?focusedCommentId=16618104&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-16618104),
user must turn off both `DECIMAL_OPERATIONS_ALLOW_PREC_LOSS` and the new
config, which makes me think we should not create a new config.
> After this patch, this query would return null instead, as an overflow
would happen. So this patch is "correcting" a regression from 2.2 but it is
introducing another one from 2.3.0-2.3.1.
I don't agree with it. Users can turn on
`DECIMAL_OPERATIONS_ALLOW_PREC_LOSS` to make the query work. We should not fix
values of some configs and then define regression, that's not a regression.
The reason why
[this](https://issues.apache.org/jira/browse/SPARK-22036?focusedCommentId=16618104&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-16618104)
is a regression is: users have no way to get the same result of 2.3 in 2.4.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
