[Python-Dev] RE: test_asynchat.py broken on Windows

[Tim Peters]

[Jeremy Hylton]
> I think there are at least three things wrong, depending on what you
> count as wrong.

I would have guessed five, if you were counting Windows too <wink>.

> The test is hanging because it spawns a thread, which Python will
> prevent Python from exiting until the thread exits.  The thread
> doesn't exit because it's waiting for input from the client.  The
> client doesn't send it any input because of the exception.
>
> thing wrong #1: It would probably make sense to change
> test_asynchat.py to use a daemon thread so that it can't prevent the
> test suite from exiting.

I don't much care about this.

> thing wrong #2: I don't think the ENONET being returned by
> connect_ex() makes any sense.  My guess is that connect_ex() is (and
> always has been) broken on Windows.  Could you apply this patch and
> see what happens:
> ...

Fixed the problem, so I checked it in.  Thanks!

> thing wrong #3: From the winsock documentation for connect(), it looks
> like there are more errors we should be catching in the test that
> includes EWOULDBLOCK and EALREADY.  I don't think this would affect
> your test, though.

I'm not sure I parsed that as intended; asyncore.dispatcher.connect (in the
traceback when test_asynchat failed) is already checking for EINPROGRESS,
EALREADY, EWOULDBLOCK, 0 and EISCONN returns from connect_ex.  But, for
posterity, I will inscribe my socket programming knowledge in the period at
the end of this sentence.

> thing wrong #4: It doesn't look to me like the connect() call in
> test_asynchat.py is guaranteed to succeed.  It should succeed almost
> every time, but there's some small chance it will fail.  Perhaps the
> test should be more robust about that.

FWIW, I've never seen it fail before.

> That's four things, but I don't think the last thing is all that wrong.

Here's a fifth:

def main():
    s = echo_server()
    s.start()
    time.sleep(1) # Give server time to initialize
    c = echo_client()
    ...

time.sleep()-is-not-a-synchronization-gimmick-ly y'rs  - tim