This is very odd, but it was reproduced by people in #python as well. 
Compare, in python 2.5:

>>> 
urllib.urlopen('http://bugs.debian.org/cgi-bin/bugreport.cgi?mbox=yes;bug=123456').readline()
'From mechanix@lucretia.debian.net Tue Dec 11 11:32:47 2001\n'

To the equivalent in python 3.0:

>>> 
urllib.request.urlopen('http://bugs.debian.org/cgi-bin/bugreport.cgi?mbox=yes;bug=123456').readline()
b'f65\r\n'

I don't reproduce the problem:

>>>
urllib.request.urlopen('http://bugs.debian.org/cgi-bin/bugreport.cgi?mbox=yes;bug=123456').readline()
b'From mechanix@lucretia.debian.net Tue Dec 11 11:32:47 2001\n'

I connect through a http proxy.

Confirmed:

Python 3.1a0 (py3k:67702, Dec 11 2008, 11:09:14)
[GCC 4.2.4 (Ubuntu 4.2.4-1ubuntu3)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import urllib.request
>>>
urllib.request.urlopen('http://bugs.debian.org/cgi-bin/bugreport.cgi?mbox=yes;bug=123456').readlines()
[b'f65\r\n', b'From mechanix@lucretia.debian.net Tue Dec 11 11:32:47
2001\n',  ...

Perhaps it's related to the \r at read boundaries bug?

The "f65" is the chunk length for the first chunk returned when
requesting that URL.  A proxy could easily hide this by switching to a
different transfer encoding.

Does the same thing happen with 2.6?

Jeremy

On Thu, Dec 11, 2008 at 8:53 AM, Jean-Paul Calderone
<report@bugs.python.org> wrote:
>
> Jean-Paul Calderone <exarkun@divmod.com> added the comment:
>
> The "f65" is the chunk length for the first chunk returned when
> requesting that URL.  A proxy could easily hide this by switching to a
> different transfer encoding.
>
> ----------
> nosy: +exarkun
>
> _______________________________________
> Python tracker <report@bugs.python.org>
> <http://bugs.python.org/issue4631>
> _______________________________________
> _______________________________________________
> Python-bugs-list mailing list
> Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/jeremy%40alum.mit.edu
>
>

> Does the same thing happen with 2.6?

No, I can't reproduce with 2.6.1.

Jeremy: no, it doesn't.

Python 2.6.1+ (release26-maint:67716M, Dec 13 2008, 10:30:52)
[GCC 4.2.4 (Ubuntu 4.2.4-1ubuntu3)] on linux2

~/release26-maint$ ./python -c "import urllib; print
urllib.urlopen('http://bugs.debian.org/cgi-bin/bugreport.cgi?mbox=yes;bug=123456').readlines()[0]"
From mechanix@lucretia.debian.net Tue Dec 11 11:32:47 2001

~/release26-maint$ ./python -c "from __future__ import unicode_literals;
import urllib; print
urllib.urlopen('http://bugs.debian.org/cgi-bin/bugreport.cgi?mbox=yes;bug=123456').readlines()[0]"
From mechanix@lucretia.debian.net Tue Dec 11 11:32:47 2001


FWIW, there are trailing spurious bytes too (note read() gives bytes,
while readlines() both bytes and strings in 3.0):
>>> import urllib.request; content =
urllib.request.urlopen('http://bugs.debian.org/cgi-bin/bugreport.cgi?mbox=yes;bug=123456').read()

Python 3.1a0 (py3k:67702, Dec 11 2008, 11:09:14)
[GCC 4.2.4 (Ubuntu 4.2.4-1ubuntu3)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import urllib.request

>>> content =
urllib.request.urlopen('http://bugs.debian.org/cgi-bin/bugreport.cgi?mbox=yes;bug=123456').read()

>>> content[-30:]
b'PGP SIGNATURE-----\n\n\n\n\n\r\n0\r\n\r\n'

>>> content[:10]
b'f65\r\nFrom '

While in 2.6:
>>> import urllib
>>> content =
urllib.urlopen('http://bugs.debian.org/cgi-bin/bugreport.cgi?mbox=yes;bug=123456').read()
>>> content[-30:]
'---END PGP SIGNATURE-----\n\n\n\n\n'

> FWIW, there are trailing spurious bytes too

And in the middle of the document as well. Each time there's a chunk, I
guess?

I have the same problem with that code:

(exchange USERNAME with your delicious username and PASSWORD with your 
delicious password):
 import urllib.request
 auth_handler = urllib.request.HTTPBasicAuthHandler()
 auth_handler.add_password('del.icio.us API', 'api.del.icio.us',
USERNAME, PASSWORD)
 opener = urllib.request.build_opener(auth_handler)
 print(str(opener.open('https://api.del.icio.us/v1/posts/all').read(20),
"utf-8"))

And I don't use a proxy or anything like that. This makes python 3
completely unusable for me. And python 2.6 gives me what I want (the
content of that virtual file) without any extra data in front or in the
middle of the content.

Took me a bit of Wiresharking to find this out, the problem is that we
are asking for the page using HTTP 1.1 in 3.0.

Here's a workaround patch for those who need it quick, I have yet to
look at urllib to see what can be fixed there.

---
Index: Lib/http/client.py
===================================================================
--- Lib/http/client.py  (revision 67716)
+++ Lib/http/client.py  (working copy)
@@ -600,7 +600,7 @@
class HTTPConnection:

_http_vsn = 11
-    _http_vsn_str = 'HTTP/1.1'
+    _http_vsn_str = 'HTTP/1.0'

response_class = HTTPResponse
default_port = HTTP_PORT
---


This is what we send in 2.5 and 3.0:

GET /cgi-bin/bugreport.cgi?mbox=yes;bug=123456 HTTP/1.0
               User-Agent: Python-urllib/1.17

GET /cgi-bin/bugreport.cgi?mbox=yes;bug=123456 HTTP/1.1
    Accept-Encoding: identity
    User-Agent: Python-urllib/3.1

IMO we should downgrade urlopen to HTTP 1.0 in 3.0.1. Implementing
chunked encoding will come later if people are interested in doing it.

Clarifying the diagnosis, the offending spurious bytes are only present
when we use 3.0's GET above.

That's because urllib.request.HTTPHandler asks for a vanilla
http.client.HTTPConnection, which uses HTTP 1.1.

IIUC, either we change the request version back to 1.0 (attached patch)
or correct the way the response is processed (is it at all?).

I think HTTPSHandler will also suffer from this, perhaps
[Fancy]URLopener too.

[Antoine: cool, an edit conflict that agrees with what I was about to
post :D]

Brief update:  The Python 2.x code works because readline() is provided
by socket._fileobject.  The Python 3.x code fails because it grabs the
HTTPResponse.fp instance variable at the end of
AbstractHTTPHandler.do_open.  That method needs to pass the response to
addinfourl(), but needs to have support for readline / readlines before
it can do that.

I have a patch here that seems to work for the specific url and that
passes all the tests.  Can anyone check whether it works for a larger
set of cases?

I'm a little concerned because I don't understand the new io library in
much detail.  There's an override for _checkClosed() in the HTTPResponse
that seems a little dodgy.  I'll try to get someone to review that
specifically.

I think your patch is good, but there may be another bug around:

I wrote a script to check results of 3.x against 2.x, but many pages
(http://groups.google.com/, http://en.wikipedia.org/) give 403:
Forbidden for 3.x... but work with 2.x!

If you think of this as a bug in 3.x, it could retry the request
identifying as 2.x on 403.

Other than that, your patch gives me identical results to 2.5/2.6 for
128 sites I tested (only a read(100) for each).

Interestingly, my patched version gives a file closer to the buggy
version in size, at 12700 bytes versus 12707. Your version agrees with
2.x and simple maths (128 x 100) in giving a 12799 bytes result. I have
no idea why.

HTH,
Daniel

The patch should have at least a test so that we don't have a regression
on this one.

Here's a test (in test_urllib2_localnet) that fails before the patch and
passes after, mostly lifted from test_httplib:

    def test_chunked(self):
        expected_response = b"hello world"
        chunked_start = (
                        b'a\r\n'
                        b'hello worl\r\n'
                        b'1\r\n'
                        b'd\r\n'
                        )
        response = [(200, [("Transfer-Encoding", "chunked")],
chunked_start)]
        handler = self.start_server(response)
        data = self.urlopen("http://localhost:%s/" % handler.port)
        self.assertEquals(data, expected_response)

Output:

test test_urllib2_localnet failed -- Traceback (most recent call last):
  File "~/py3k/Lib/test/test_urllib2_localnet.py", line 390, in test_chunked
    self.assertEquals(data, expected_response)
AssertionError: b'a\r\nhello worl\r\n1\r\nd\r\n' != b'hello world'

To allow this test to work, the attached patch also touches
FakeHTTPRequestHandler and TestUrlopen.urlopen.

On the principle, the test looks good.
If you want to avoid the 'if "%" in value' hack, you can use the
named-parameter form of string formatting:

>>> "localhost:%(port)s" % dict(port=8080)
'localhost:8080'
>>> "localhost" % dict(port=8080)
'localhost'

Antoine,
Thanks for reviewing, here's an updated version.

The test looks good to me.
I can't comment on the bugfix patch, but if it's ok to you, you can go
ahead :)

I took a look at the patch and it looks ok, apart from the
_checkClosed() hack (but I don't think there's any immediate solution).
It should be noted that HTTPResponse.readline() will be awfully slow
since, as HTTPResponse doesn't have peek(), readline() will call read()
one byte at a time...

(slow I/O is nothing new in py3k, however :-))

Here is a patch without the _checkClosed() hack. The solution is simply
to remove redundant _checkClosed() calls in IOBase (for example,
readline() doesn't need to do an explicit `closed` check as it calls
read()).

Committed in r69513, r69514. Thanks everyone!