Github user dbtsai commented on a diff in the pull request:
https://github.com/apache/spark/pull/17606#discussion_r110979361
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/TypeCoercion.scala
---
@@ -571,6 +571,7 @@ object TypeCoercion {
NaNvl(l, Cast(r, DoubleType))
case NaNvl(l, r) if l.dataType == FloatType && r.dataType ==
DoubleType =>
NaNvl(Cast(l, DoubleType), r)
+ case NaNvl(l, r) if r.dataType == NullType => NaNvl(l, Cast(r,
l.dataType))
--- End diff --
Yeah, this PR prevents casting from `NaNvl(FloatType, NullType)` to
`NaNvl(DoubleType, DoubleType)` since we want to minimize the casting as much
as possible. Also, if we want to replace `NaN` by `null`, we want to keep the
output type the same as input type.
Whether `NaNvl(FloatType, DoubleType)` should be cast into
`NaNvl(DoubleType, DoubleType)` is another story. I agree with you, we should
downcast the replacement `DoubleType` into `FloatType`. And in my opinion,
doing this implicit casting is error-prone, and we should do explicit casting
by users instead.
@gatorsmile maybe you can chime in, and give the feedback whether we should
cast `NaNvl(FloatType, DoubleType)` to `NaNvl(DoubleType, DoubleType)`.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
