Github user srowen commented on the issue:
https://github.com/apache/spark/pull/22144
@markhamstra I do think there's an unspoken but legitimate consideration
here, and that's that there is also a cost to not shipping the N thousand other
things users are waiting on in this release. This one sounds like it may cause
weeks of delay, which is a lot. (And while the fact that this is already well
delayed shouldn't matter to that logic, it kind of does in practice.)
That cost is why I'd say this change shouldn't block, rather than an
erosion of some standard. In fact, there seemed to be plenty of appetite to
back and fix several things in 2.4 that weren't, probably, correctness or
regression problems.
As to whether this is a regression -- does sound like the current behavior
is not what's intended and even wrong in some cases and that has to be fixed.
It seems it was a regression in 2.2 that wasn't caught, and therefore should
have blocked that release, but that ship has sailed. Now, it's become simply an
established bug like any other. That doesn't make it less bad but it should be
treated like something we found today vis-a-vis the release process.
I am left a little unclear on whether there's a correctness issue here.
And: does 'fixing' this cause some user UDAFs that worked around this to break
too?
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
