I helped develop a small C++/Python library for interacting with databases like Hive and Impala via the HiveServer2 Thrift protocol and making them accessible to Python / pandas:
Internally this interfaces with HS2's own columnar representation. Arrow is a natural partner for this project, much of which could be discarded. I think Arrow would make as much sense as any place to develop this codebase further. It could be later split off into a new project if a large enough community develops
If we did this, do we need to do a software grant (essentially what I'm proposing is to fork)? Can we just attribute the original Cloudera authors in LICENSE.txt?