That example is wrong, but that fact weighs heavily in favor of the idea that the "natural" way to write it is the way it is in the example.

Since this part of the email api is still provisional, I propose we change the signature of Address to be:

  Address(display_name='', addr_spec=None, *, username='', domain='')

and make it a ValueError to specify both addr_spec and one or both of username and domain.  For 3.5 I'd just make that change, but for 3.4 I'd make a backward compatibility check and deprecation warning looking for an '@' in the second argument and treat it as username if there's no @ and an addr_spec if there is.