I would suggest to make fastavro a default option on Python 3, for the lack of alternative at the moment, but keep current default on Python 2.
I would also keep both avro and avro-python3 dependencies and associated codepaths. This way, we will gradually increase the usage of fastavro, but keep an alternative available in case users encounter issues. Hopefully with next Avro release, the non-default alternative on Python 3 will be viable, we can keep the tests around to check that. As usage of Python 3 ramps up, we will have more confidence in making the switch to fastavro a default implementation on both Python 2 and Python 3 if we so choose. Thanks, Valentyn
