We hit upon an an issue in hcat when a user was trying to read a table
that had a map column and one of the keys in the map column was null.
The core issue of the bug is easy enough to fix, but it did bring up a
question on whether or not this was a valid usecase and whether or not
we should be supporting nulls as keys in maps.
I've written up my thoughts on the matter on the jira :
https://issues.apache.org/jira/browse/HIVE-5020 . If any more of you
would like to weigh in on the matter, I'd appreciate it.
As of right now, I'm leaning towards supporting nulls, as detailed on the jira.
A further follow up here, would be that rcfile and orc seem to have
different behaviours for this issue, and if we decide to support
nulls, and that stripping them silently is bad, then rcfile needs to
be fixed as well. I can do that as part of this bug or as a separate