Github user mgaido91 commented on the issue:
https://github.com/apache/spark/pull/22494
> So I don't see any harm of this PR.
If the user doesn't turn off the flag, of course nothing changes. If the
user does, then let's imagine this case. A user has this: `select 1234567891 /
(cast(2 as bigint) * (a decimal(1, 1) number) * cast(2 as bigint))`. Before
this PR, this operation works fine. After the PR the right operand would have a
precision which is at least 38 (while it could be much less, ie. 4 before this
change). The result of the division would be a decimal(38, 38): so the
operation would overflow.
It is true that after this PR, Spark will support (with the same exact
precision) all the operations that worked fine before #20023. But this PR also
can break some operation that started working fine after #20023 (even with the
flag turned off).
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
