On FreeBSD and MacOS 64-bit systems the ident field of a kevent is big
enough to hold a 64-bit integer (uintptr_t). Looks like Python is
casting it to an unsigned 32-bit integer.

This is inconvenient for implementing kqueue timers, where id(timer_obj)
is a natural choice for an ident.

Would you like to propose a patch?

I'm not sure how to patch this so that it will work on both 32 and 64
bit systems. Issues:

1. What would be an appropriate member type for ident in
kqueue_event_members? It seems like T_PYSSIZET might work. Otherwise, I
am guessing that this will involve some #if's.

2. I think the format spec in kqueue_event_repr needs to change. It
seems like this will also require some #if's.

3. kqueue_event_init uses PyObject_AsFileDescriptor to set the ident
field. This should be doing a PyLong_Check first to see if
PyLong_AsSomething would be more appropriate.

4. I think the type of the result variable in kqueue_event_richcompare
needs to be changed to long long int.

> 1. What would be an appropriate member type for ident in
> kqueue_event_members? It seems like T_PYSSIZET might work. Otherwise, I
> am guessing that this will involve some #if's.

IIUC, it needs to match *exactly* the field size inside struct kevent.
So you'll have to use #ifs, possible along with autoconf tests (to
find out the sizes of the fields that typically vary across
implementations).

> 2. I think the format spec in kqueue_event_repr needs to change. It
> seems like this will also require some #if's.

Here, I would widen the fields to size_t, and use the size_t formatter
(assuming that all systems supporting kqueue also know how to print
size_t, or know to print long long).

> 3. kqueue_event_init uses PyObject_AsFileDescriptor to set the ident
> field. This should be doing a PyLong_Check first to see if
> PyLong_AsSomething would be more appropriate.

Hmm. I think I need to understand the use case better. Can you post
some sample code where this all matters?

If this *is* a regular file-like object, then surely int is enough, no?

> 4. I think the type of the result variable in kqueue_event_richcompare
> needs to be changed to long long int.

Fine with me. The question then is whether long long is available on
all systems that support kqueue.

Martin, thanks for your responses. In regards to point three:

Kqueue's are not just used for file descriptors. I believe this is the
reason why the ident field is a uintptr_t and not an int.

The example I gave was for kqueue timers. Since the operating system
does not allocate 'timer descriptors' for you, I decided to use the
return value from the id function.

Here is some code that demonstrates:

import select
a = 1
b = id(a)
c = select.kevent(a)
d = select.kevent(b)
assert a == c.ident
assert b == d.ident
assert b & (1<<32) - 1 == d.ident

The second assert will fail on 64 bit systems if 'b' is too big.

Anyway, I will try to come up with a patch for this.

This is against release31-maint, if it matters.

According to man pages on the Web, the kevent structure is:

     struct kevent {
	     uintptr_t ident;	     /* identifier for this event */
	     short     filter;	     /* filter for event */
	     u_short   flags;	     /* action flags for kqueue */
	     u_int     fflags;	     /* filter flag value */
	     intptr_t  data;	     /* filter data value */
	     void      *udata;	     /* opaque user data identifier */
     };

So the `ident` field is indeed an uintptr_t.
However the patch is slightly wrong IMO:

- you should not blindly use `long long`. Python already defines a
"Py_uintptr_t" type. If you need more information you should conditional
compilation such as in (for the off_t type)
http://svn.python.org/view/python/trunk/Modules/_io/_iomodule.h?view=markup

- in tests, you should use sys.maxsize (which gives you the max value of
a ssize_t integer) rather than choosing based on platform.architecture():

>>> sys.maxsize
9223372036854775807
>>> 2**64*1
18446744073709551616L
>>> sys.maxsize*2 + 1
18446744073709551615L


PS : a patch against trunk or py3k is preferred, but in this case it
would probably apply cleanly anyway

Apparently Roundup borked the URL. Let's try another one (or look into
Modules/_io/_iomodule.h): 
http://code.python.org/hg/trunk/file/b9bc35171668/Modules/_io/_iomodule.h#l88

Antoine, thanks for the tips and the example. I have updated the patch.

I checked and this does apply cleanly to py3k.

The patch (http://bugs.python.org/file15228/kevent.patch) works for me
under OS X 10.5.8 Intel.

All tests pass, except test_telnetlib which fails intermittently with
and without the patch.

Python 3.2a0 (py3k:75951, Oct 29 2009, 11:38:58) 
[GCC 4.0.1 (Apple Inc. build 5465)] on darwin

For me, the patch is worth trying out on the buildbots - to see if there
are any configuration problems we might have overlooked.

Patch was committed in r76108 (trunk) and r76111 (py3k) and didn't
introduce any regression on the FreeBSD and OS X buildbots. Thanks!

This patch causes build errors on 32-bit/64-bit multi-architecture builds 
on OS X.  See Issue7416 for more info and a patch.