Github user HyukjinKwon commented on the issue:
https://github.com/apache/spark/pull/22144
If we were going for 3.0, then I would definitely leave +1 and I agree that
we should rather focus on Spark itself as a higher priority - we should do that
when we go 3.0 and rather drop such cases IMHO. However, looks we are going for
2.4 for now.
It looks indeed a mistake and bringing a feature back that we had in
previous releases, and sounds like the problem is not isolated (let's say, it's
not specific to a few set UDAFs only), which, IMHO, looks a proper blocker.
How about we bring this back if the changes proposed here are all what we
need? If the current change is not enough, we could at least mention this
feature was dropped intentionally in migration guide.
About testing, this PR described "How was this patch tested?" although it
has no test. @pgandhi999, are you able to write a test?
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
