Logged In: YES 
user_id=31435

No disagreement here.  I'd rather fix the code than the docs, 
and a fix can take two forms:  raise an exception if the full 
range of ints can't be delivered, or deliver the full range of 
ints.

If anyone wants to take this on, I think the latter ("do a right 
thing") may be much more practical than it used to be, given 
the new-in-2.3 Twister impelementation -- we can generate 
long pseudo-random bitstrings quickly now (at least at the C 
level), but couldn't before 2.3.