The specification for seek() says:

 seek(  	offset[, whence])
    Note that if the file is opened for appending (mode
'a' or 'a+'), any seek() operations will be undone at
the next write.

Consider operating on an fdopen()ed file. The Python
source simply calls into the OS-provided fdopen():

http://pxr.openlook.org/pxr/source/Modules/posixmodule.c#3530

However, the POSIX standard

http://www.opengroup.org/onlinepubs/009695399/functions/fdopen.html

says:

"Although not explicitly required by this volume of
IEEE Std 1003.1-2001, a good implementation of append
(a) mode would cause the O_APPEND flag to be set."

Thus, to ensure Python semantics, Python's fdopen()
must perform an fcntl() to ensure O_APPEND is set.

For example, on Solaris, this optional O_APPEND
behaviour is not applied:

http://cvs.opensolaris.org/source/xref/on/usr/src/lib/libc/port/stdio/fdopen.c?r=1.22#97

This has recently caused serious problems with the
Mercurial SCM.

Logged In: YES 
user_id=17929

freebsd 4.11 shows the same behaviour.

import fcntl
import os

def test_fdopen_append():
    
    def is_append(fd):
        return bool(fcntl.fcntl(fd, fcntl.F_GETFL) &
os.O_APPEND)


    fd = os.open("foo.txt", os.O_RDWR | os.O_CREAT)
    assert fd != -1
    
    print "is_append:", is_append(fd)
    
    f=os.fdopen(fd, 'a')
    print "after fdopen is_append:", is_append(fd)

    assert is_append(fd)

test_fdopen_append()

Logged In: YES 
user_id=17929

patch here:
http://sourceforge.net/tracker/index.php?func=detail&aid=1462227&group_id=5470&atid=305470

Logged In: YES 
user_id=849994

Applied patch in rev. 43501, 43502.

Logged In: YES 
user_id=53034

Shouldn't the documentation now state this change in
behaviour? i.e. if fdopen() fails, O_APPEND may still have
become set.

This behaviour seems a little odd to me in fact. Can't we set
fcntl after a successful fdopen()[1]? Also wouldn't it be
useful to point to the standards sentence I quote above?

[1] not sure what the ALLOW_THREAD thing does?