*Or*, it now occurs to me, just stick with the ASCII-only policy that's already in place. I'd even say this is more correct since porting isn't supposed to change behavior. Leave support for some other codec for another ticket (perhaps #989). Apart from being simpler (I hope) and avoiding breaching the documented porting guidelines, this also means someone will actually have to think about unicode support on Python 2 as well. Saying we support unicode in the logging system is a lot better than saying we support unicode in the logging system on Python 3 only.