If you're dealing with emoji -- and yes, they are incredibly common -- you'd better not use any Python before 3.3.
Non-BMP characters in 2.x-3.2 give you implementation-specific behavior. It can vary from one system to the next. Some string methods fail on "narrow builds" of Python (the ones that store their characters in UCS-2), while others return different results than they would on a "wide build".