http://code.google.com/p/pyftpdlib/issues/detail?id=143
This comes from a user who sent me a report via e-mail. Unfortunately I don't have an OSX box to test against.

Code which should replicate the problem is this:

import socket, struct
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.setsockopt(socket.SOL_SOCKET, socket.SO_LINGER, struct.pack('ii', 1, 0))
s.connect(('localhost', 21))
s.close()

...while this is a fix I think it should work:

Index: Lib/asyncore.py
===================================================================
--- Lib/asyncore.py	(revisione 86084)
+++ Lib/asyncore.py	(copia locale)
@@ -242,7 +242,7 @@
             try:
                 self.addr = sock.getpeername()
             except socket.error, err:
-                if err.args[0] == ENOTCONN:
+                if err.args[0] in (ENOTCONN, EINVAL):
                     # To handle the case where we got an unconnected
                     # socket.
                     self.connected = False


Nosying ixokai as I know he has an OSX box to test against.
Setting "high" priority and type == "security" as asyncore-based servers are likely to crash because of this.
It might even make sense to backport the fix in Python 2.6 because of the security implications.

I can verify the problem exists in asyncore at release27-maint on the mac, and that the below patch fixes it.

After applying, I ran a full regrtest and nothing new broke.

While writing a test case for this I found out another problem in asyncore: handle_connect was erroneously called when the dispatcher delegates the connection to a handler resulting in ENOTCONN being raised.
Patch in attachment targeted for python 2.7 should fix both issues.

Forgot to attach the patch.

"Might even make sense" to backport doesn't sound like a definite, so I've removed 2.6 and 2.5 from versions.  You'll want to ask the release managers for a decision if you want to backport.

New changeset 8c19c9914c22 by Giampaolo Rodola' in branch '2.7':
fix #10340: properly handle EINVAL on OSX and also avoid to call handle_connect() in case of a disconnetected socket which is not meant to connect.
http://hg.python.org/cpython/rev/8c19c9914c22

New changeset e2cddb3f4526 by Giampaolo Rodola' in branch '3.2':
fix #10340: properly handle EINVAL on OSX and also avoid to call handle_connect() in case of a disconnetected socket which is not meant to connect.
http://hg.python.org/cpython/rev/e2cddb3f4526

New changeset 6ffdca50a5ef by Giampaolo Rodola' in branch 'default':
merge 79422b3684f1 in 3.3 branch (issue 10340)
http://hg.python.org/cpython/rev/6ffdca50a5ef

New changeset 8c1fd9276b25 by Giampaolo Rodola' in branch '3.2':
issue 10340 - forgot to update Misc/NEWS
http://hg.python.org/cpython/rev/8c1fd9276b25

New changeset 13cefcbcc7da by Giampaolo Rodola' in branch 'default':
fix #10340: properly handle EINVAL on OSX and also avoid to call handle_connect() in case of a disconnetected socket which is not meant to connect.
http://hg.python.org/cpython/rev/13cefcbcc7da

This appears to be causing buildbot failures:

http://www.python.org/dev/buildbot/all/builders/x86%20debian%20parallel%203.x/builds/4077

http://www.python.org/dev/buildbot/all/builders/AMD64%20Gentoo%20Wide%203.x/builds/3520

New changeset e35a5bbb0b91 by Giampaolo Rodola' in branch 'default':
fix failing asyncore test as per http://bugs.python.org/issue10340#msg156586
http://hg.python.org/cpython/rev/e35a5bbb0b91

Sorry about that. It should now be fixed.

Better, but not stable:

http://www.python.org/dev/buildbot/all/builders/AMD64%20OpenIndiana%203.2/builds/984/steps/test/logs/stdio

http://www.python.org/dev/buildbot/all/builders/x86%20OpenIndiana%203.2/builds/998/steps/test/logs/stdio

http://hg.python.org/cpython/rev/0b960e41e533
Let's see how it goes.

This looks resolved. Can it be closed?

I'm assuming that the buildbots stabilized so I'm going to go ahead and close this.