Recently, while trying to resolve a bug in Bcfg2, I ran into a situation which can be summed up by the following:
Python 2.7.1 (r271:86832, Mar 26 2011, 11:26:21)
[GCC 4.4.5] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import os, stat
>>> dev = os.makedev(1, 3)
>>> mode = stat.S_IFCHR | 0777
>>> print(mode)
8703
>>> os.mknod('test', mode, dev)
>>> os.stat('test')
posix.stat_result(st_mode=8685, st_ino=1148358, st_dev=12L, st_nlink=1, st_uid=0, st_gid=0, st_size=0, st_atime=1314372451, st_mtime=1314372451, st_ctime=1314372451)
Above, you can see that the mode specified ends up being different than the mode which is set by os.mknod. Instead of a character device with permissions of 0777, I was ending up with permissions of 0755. If you follow the link, you will find no documentation mentioning the umask of the running process in the mknod section. However, you can search around the page and realize that the umask of the running process is masked out for other methods.
The inconsistency arises due to the implementation of mknod used by Python. For instance, if you run the above code on Windows under Cygwin, it does the Right Thing ™. This was my clue that there was something about the implementation that was off. Sure enough, after committing a simple fix, the problem disappeared.
I think this is simply a documentation issue, but I was unable to find any information on the problem while searching around. Hopefully this post will save someone from wasting a ton of time on the same issue.