In Modules/selectmodule.c it is assumed that the kevent structure is defined on FreeBSD and NetBSD as:

     struct kevent {
             uintptr_t ident;
             short     filter;
             u_short   flags;
             u_int     fflags;
             intptr_t  data;
             uintptr_t udata;
     };

and on OpenBSD as:

     struct kevent {
             u_int     ident;
             short     filter;
             u_short   flags;
             u_int     fflags;
             intptr_t  data;
             int       udata;
     };

Actually it is defined on FreeBSD as:

     struct kevent {
             uintptr_t ident;
             short     filter;
             u_short   flags;
             u_int     fflags;
             intptr_t  data;
             void      *udata;
     };

On OpenBSD as:

     struct kevent {
             uintptr_t ident;
             short     filter;
             u_short   flags;
             u_int     fflags;
             int64_t   data;
             void      *udata;
     };

And on NetBSD as:

     struct kevent {
             uintptr_t ident;
             uint32_t  filter;
             uint32_t  flags;
             uint32_t  fflags;
             int64_t   data;
             intptr_t  udata;
     };

Other issues are related to rich comparison. Due to integer overflows the ordering is not transitive. The rich comparison protocol is not properly supported, comparing a kevent_event object with a non-kevent_event object don't falls back to the rich comparison methods of the right arguments.

New changeset b9052a0 by Serhiy Storchaka in branch 'master':
bpo-31893: Fixed select.kqueue(). (bpo-4166)
b9052a0

New changeset f9a639b by Serhiy Storchaka (Miss Islington (bot)) in branch '3.6':
bpo-31893: Fixed select.kqueue(). (GH-4166) (bpo-4190)
f9a639b

The commit b9052a0 broke compilation on FreeBSD:

building 'select' extension
cc -pthread -fPIC -Wno-unused-result -Wsign-compare -g -O0 -Wall -Wstrict-prototypes -std=c99 -Wextra -Wno-unused-result -Wno-unused-parameter -Wno-missing-field-initializers -Werror=implicit-function-declaration -I./Include -I. -I/usr/local/include -I/usr/home/buildbot/python/3.x.koobs-freebsd10/build/Include -I/usr/home/buildbot/python/3.x.koobs-freebsd10/build -c /usr/home/buildbot/python/3.x.koobs-freebsd10/build/Modules/selectmodule.c -o build/temp.freebsd-10.3-STABLE-amd64-3.7-pydebug/usr/home/buildbot/python/3.x.koobs-freebsd10/build/Modules/selectmodule.o
/usr/home/buildbot/python/3.x.koobs-freebsd10/build/Modules/selectmodule.c:1889:63: error: expected ';' at end of declaration
FILTER_FMT_UNIT FLAGS_FMT_UNIT FFLAGS_FMT_UNIT DATA_FMT_UNIT
^
;
1 error generated.
*** WARNING: renaming "_asyncio" since importing it failed: No module named 'select'

http://buildbot.python.org/all/#/builders/87/builds/95/steps/3/logs/stdio

New changeset 8cbf4e1 by Serhiy Storchaka in branch '2.7':
[2.7] bpo-31893: Fixed select.kqueue(). (GH-4166) (bpo-4193)
8cbf4e1

Oh, the commit also broke the "x86 Tiger 3.x buildbot:

http://buildbot.python.org/all/#/builders/30/builds/93

======================================================================
FAIL: test_create_event (test.test_kqueue.TestKQueue)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/Users/db3l/buildarea/3.x.bolen-tiger/build/Lib/test/test_kqueue.py", line 71, in test_create_event
    self.assertEqual(ev.data, 5)
AssertionError: 25769803781 != 5

Oh, I forgot that Mac OS X also in the BSD family.

Thank you Victor for signaling errors.

New changeset 2298fad by Serhiy Storchaka in branch 'master':
bpo-31893: Fix errors in b9052a0. (bpo-4196)
2298fad

New changeset 84e252b by Serhiy Storchaka (Miss Islington (bot)) in branch '3.6':
bpo-31893: Fix errors in b9052a0. (GH-4196) (bpo-4201)
84e252b

New changeset e7531e5 by Serhiy Storchaka (Miss Islington (bot)) in branch '2.7':
bpo-31893: Fix errors in b9052a0. (GH-4196) (bpo-4202)
e7531e5

Tests on all buildbots are passed.

New changeset ce51890 by Serhiy Storchaka in branch '2.7':
bpo-31893: Fix a backporting error in 8cbf4e1. (bpo-4219)
ce51890

Test fails on x86 Tiger 2.7:

http://buildbot.python.org/all/#/builders/59/builds/17

======================================================================
FAIL: test_create_event (test.test_kqueue.TestKQueue)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/Users/db3l/buildarea/2.7.bolen-tiger/build/Lib/test/test_kqueue.py", line 39, in test_create_event
    self.assertRaises(TypeError, cmp, ev, None)
AssertionError: TypeError not raised

Same failure on x86-64 El Capitan 2.7/
http://buildbot.python.org/all/#builders/98/builds/17

======================================================================
FAIL: test_create_event (test.test_kqueue.TestKQueue)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/Users/buildbot/buildarea/2.7.billenstein-elcapitan/build/Lib/test/test_kqueue.py", line 39, in test_create_event
    self.assertRaises(TypeError, cmp, ev, None)
AssertionError: TypeError not raised

On also AMD64 FreeBSD 10.x Shared 2.7:

http://buildbot.python.org/all/#/builders/97/builds/17

======================================================================
FAIL: test_create_event (test.test_kqueue.TestKQueue)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/usr/home/buildbot/python/2.7.koobs-freebsd10/build/Lib/test/test_kqueue.py", line 39, in test_create_event
    self.assertRaises(TypeError, cmp, ev, None)
AssertionError: TypeError not raised

Oh, I didn't test on 2.7.

The old code contradicted the common comparison behavior. It allowed comparing select.kevent with other objects only for equality as for identity. kevent == other always returned False even if other == kevent returned True. kevent < other raised a TypeError.

The new code uses the default behavior. kevent == other returns the same as other == kevent (False by default if types are not comparable, but may return True if other.__eq__(kevent) returns True). kevent < other returns the same as other > kevent. E.g. raises a TypeError by default in Python 3. But in Python 2 all objects are comparable by default, and this breaks a test which expects that select.kevent is not comparable.

There are two ways to fix this:

1. Make select.kevent non-comparable again. This contradicts the default behavior and I don't know reasons why it should be non-comparable, but this was an existing behavior.

2. Remove the tests or change them to test that comparing select.kevent with other object doesn't raise an error.

kqueue_event_richcompare() returns NotImplemented if other is not an event:

    if (!kqueue_event_Check(o)) {
        Py_INCREF(Py_NotImplemented);
        return Py_NotImplemented;
    }

So I don't understand why tests started to fail with the commit ce51890.

Because the default behavior is different in Python 2 and Python 3. In Python 3 this is a TypeError. In Python 2 all objects are comparable by default.

I suggest to fix the tests, the code looks good to me.

Fixed by PR 4349 (changeset 15e1453).