Message1672

it's not a problem with accept because it creates a new socket object, but there's definitely a race with dup and even makefile. I've added locking for this in the latest patch.

_fileobject.close() has a similar race, but the worst thing that can happen with it is the underlying socket and its streams having close() called more than once. Subsequent close() calls are noops (and synchronized) anyway. The same thing was possible with the original FileWrapper closer

I also fixed a problem with makefile and _socketobject, they weren't handling _closedsocket situations correctly (don't assume they should always reference_count++). Added tests for that, and some basic tests for dup() and close() with dup()
File Added: socket_fileobject4-r3591.diff