As Martin commented on my patch at bpo-26392 the socketserver.server_close is like the file close. That made me think that we should add a context manager to the socketserver.

I would support this change, as long as we aren’t supporting using server_close() to stop the server loop at the same time.

It might be worth updating other example code that uses server_close(), in particular in the xmlrpc.server documentation.

Only closing the server :).

1. Did the changes requested in the CR.
2. Changed the example's in xmlrpc.server, http.server to use context manager.
3. Changed the xmlrpc.server, http.server server implementation when running python -m {xmlrpc.server, http.server} to use context manager.

Thanks, the patch looks pretty good to me. I left some comments about minor things. Also, someone will need to write a What’s New entry.

Updated the patch:

1. Did the changes requested in the CR.
2. Changed the example's in wsgiref.simple_server to use context manager.
3. Added What's New entry.

updated the patch according to the CR comments.

This seems like an appropriate enhancement. I notice that the serve_forever examples did not previously have server_close(). Was this an oversight or will the server_close in __exit__ just be a no-op?

The examples did not close the server but I am not sure if that was a mistake. The server is not being closed only on examples where it is expected to run until there is a keyboard interrupt. However you can see that when you send keyboard interrupt to a server you start using python -m http.server you will do close the server.

Server_close() was only documentated last year; see bpo-23254. For the examples that run until you interrupt them, the servers currently emit a resource warning (in addition to the KeyboardInterrupt traceback and the Python 2 bytes warnings):

$ python -bWall TCPServer.py
127.0.0.1 wrote:
TCPServer.py:16: BytesWarning: str() on a bytes instance
  print(self.data)
b'hello world with TCP'
127.0.0.1 wrote:
TCPServer.py:16: BytesWarning: str() on a bytes instance
  print(self.data)
b'python is nice'
^CTraceback (most recent call last):
  File "TCPServer.py", line 28, in <module>
    server.serve_forever()
  File "/usr/lib/python3.5/socketserver.py", line 237, in serve_forever
    ready = selector.select(poll_interval)
  File "/usr/lib/python3.5/selectors.py", line 367, in select
    fd_event_list = self._poll.poll(timeout)
KeyboardInterrupt
sys:1: ResourceWarning: unclosed <socket.socket fd=3, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 9999)>
[Exit 1]

If you ignore the warning, there isn’t much effective difference, because the socket gets closed when the process exits, or if you are lucky, when Python garbage collects the global “server” object. But IMO it is bad practice not to clean up resources properly, especially in an API example.

I agree and consider this is an argument for adding the context manager methods and using them with 'with' in the examples.

New changeset 5c4303c46a18 by Martin Panter in branch 'default':
Issue bpo-26404: Add context manager to socketserver, by Aviv Palivoda
https://hg.python.org/cpython/rev/5c4303c46a18

Thanks for the patch Aviv. I made a few minor English grammar etc tweaks in the version I committed, as pointed out in the review comments.