None of my arguments have really changed since 2.4.  I still believe 
that this is a poor choice of default behaviour (and if it is meant to 
be overridden to be useable, then 'pass' or 'raise 
NotYetImplementedError' would be a better choice).

However, my impression is that nothing I say will convince you of that, 
so I'll give up on that.  It would have been interesting to hear from 
Andrew the reason for the code.

However, this is definitely a documentation bug (as outlined 
previously), and so should be fixed, not closed.  I can provide a patch 
if it stands a chance of being accepted.