HTTPResponse.msg is documented as a http.client.HTTPMessage object containing the headers of the response [1].

But in fact this is a string containing the status code:

>>> import urllib.request
>>> req=urllib.request.urlopen('http://heise.de')
>>> content = req.read()
>>> type(req.msg)
<class 'str'>
>>> req.msg
'OK'

This value is apparently overriden in urllib/request.py:

./urllib/request.py:1246: # This line replaces the .msg attribute of the HTTPResponse
./urllib/request.py-1247- # with .headers, because urllib clients expect the response to
./urllib/request.py:1248: # have the reason in .msg. It would be good to mark this
./urllib/request.py-1249- # attribute is deprecated and get then to use info() or
./urllib/request.py-1250- # .headers.
./urllib/request.py:1251: r.msg = r.reason

Anyhow, it should be documented, that is not safe to retrieve the headers with HTTPResponse.msg and maybe add HTTPResponse.headers to the documentation.

[1] https://docs.python.org/3/library/http.client.html

See bpo-21228 for a patch which documents the “.msg” hack, and that the info() method is available for HTTP responses.

I think documenting the “.headers” attribute would be a bad idea, because it is introducing yet another way to do what the almost-documented info() method already does.

Documenting the “headers” attribute is also discussed in bpo-12707

New changeset fa3c9faabfb0 by Martin Panter in branch '3.4':
Issues bpo-22989, bpo-21228: Document HTTP response object for urlopen()
https://hg.python.org/cpython/rev/fa3c9faabfb0

New changeset b55c006b79bc by Martin Panter in branch '3.5':
Issue bpo-22989, bpo-21228: Merge urlopen() doc from 3.4 into 3.5
https://hg.python.org/cpython/rev/b55c006b79bc

New changeset c6930661599b by Martin Panter in branch 'default':
Issue bpo-22989, bpo-21228: Merge urlopen() doc from 3.5
https://hg.python.org/cpython/rev/c6930661599b

The documentation now mentions the “msg” quirk and the info() method.