Josiah, that solution isn't working for me;  it looks as though there's 
a deeper weirdness:  what I'm seeing is that on OS X, in e.g. 
test_emptyline, we end up calling the readwrite function in asyncore.py 
with flags = POLLIN | POLLPRI | POLLHUP.  The first thing that gets 
called is obj.handle.read_event, and bizarrely it seems to be the 
read_event call that has the effect of closing the socket.  So the 
socket's already closed by the time we get to obj.handle_expt_event and 
the same error as before occurs.

I'm still trying to figure out why.