os module have varios O_ declerations but it doesn't define O_ASYNC.
Same for fcntl module which doesn't define FASYNC.

Simply defining 

O_ASYNC = 020000

works fine but its not as elegant as saying os.O_ASYNC.

These flags are non-posix, linux-specific constants. Python 'os' module
uses 'posix' module for all *nix systems, including those, that do not
support O_ASYNC and FASYNC flags. I think your feature request should be
rejected.

I think they at least should be supported on Linux then. It does work if
you use the value itself anyway.

>I think they at least should be supported on Linux then. 

And what happens if some Unix flavor (i.e. Solaris) adds new flag, say
O_NONLINUXSYNC and it has the same value as linux's O_ASYNC? And then
FreeBSD adds O_BSDSYNC flag with the same number and so on. Then Python
will have to have separate system-specific module for each *nix system
that uses Python. Why do you think Python maintainers would want that?

>It does work if you use the value itself anyway.

It's the way fcntl works -- it's just a thin layer on top of fcntl() and
ioctl() calls. But that's not a good reason for including non-posix
flags to 'posix' module

Platform-specific values are not a problem.
On win32 os.O_NOINHERIT == 128, and on linux os.O_EXCL has the same value.

Python just exposes the available #defines at the module level.
(btw, the posix module is also available on windows. It is named 'nt',
but it is really the same module)

And in posixmodule.c, There are already some "GNU extensions":
os.O_DIRECT, os.O_NOFOLLOW...

Add O_ASYNC and FASYNC as GNU extensions.

Python code already has GNU/Windows only exceptions. Attached patch adds
O_ASYNC and FASYNC as GNU extensions. If accepted I can send a
complimentary documentation patch.

OK, committed patch in r63368. Thanks!

Thanks, can you also apply to py3k branch? :)

That will happen automatically.

Documentation patch is added too.

Thanks, applied in r63373.