the url 
  http://docs.python.org/lib/socket-example.html
gives an example that uses socket.getaddrinfo that has this code

   HOST=''
   ...
   for res in socket.getaddrinfo(HOST, PORT, socket.AF_UNSPEC,
     socket.SOCK_STREAM, 0, socket.AI_PASSIVE)

but this fails on freebsd 6.1/python 2.5 as follows

> /usr/home/mrm/s2.py(30)<module>()
-> for res in socket.getaddrinfo(HOST, PORT, socket.AF_UNSPEC, 
socket.SOCK_STREAM, 0, socket.AI_PASSIVE):
(Pdb) p HOST
''
(Pdb) n
gaierror: (8, 'hostname nor servname provided, or not known')
> /usr/home/mrm/s2.py(30)<module>()

but setting 
  HOST=None
works fine.

either this is a doc bug or a code bug, your pick.

What operating system are you using? It could be a bug in the operating
system as well.

Martin v. Löwis wrote:
> Martin v. Löwis <martin@v.loewis.de> added the comment:
> 
> What operating system are you using? It could be a bug in the
> operating system as well. 

hi martin,

what i've tested so far...

  freebsd 6.2 - release 12 jan 2007
    socket.gaierror: (8, 'hostname nor servname provided, or not known')

  Red Hat Enterprise Linux Client release 5.1 (Tikanga), gnu/linux
2.6.18-53.el5
     socket.gaierror: (-2, 'Name or service not known')

mike

This also affects Python 2.4 and 2.6 on Linux systems. Bug
http://bugs.python.org/issue2763 is a duplicate of this one.

The issue is that socketmodule.c doesn't convert empty strings to NULLs
before passing hptr through to the underlying system getaddrinfo(...).
The question is whether to fix the documentation and examples or the
socketmodule.c code.

Attached a patch to correct the getaddrinfo(...) documentation and the
code example in socket.rst.

The documentation says: 

For *host* and *port*, by passing either an empty string or ``None``,
you can pass ``NULL`` to the C API, so the documentation says it should
work.

This doesn't work because PyString_AsString retruns an empty string, not
NULL. The attached patch adds a check in socketmodule.c to fix this for
host and port.

It's highly debatable whether this is better than fixing the
documentation, though.

I see. I just reread the getaddrinfo man page, and it says you must pass
NULL to leave host unspecified, so passing None is correct, and the
documentation is wrong.

Applied the documentation patch in r63039. Thanks for your efforts!