This test failure just happened to me on a py3k checkout:

test test_urllibnet failed -- Traceback (most recent call last):
  File "/home/antoine/py3k/debug/Lib/test/test_urllibnet.py", line 191, in test_data_header
    time.strptime(datevalue, dateformat)
  File "/home/antoine/py3k/debug/Lib/_strptime.py", line 461, in _strptime_time
    return _strptime(data_string, format)[0]
  File "/home/antoine/py3k/debug/Lib/_strptime.py", line 332, in _strptime
    (data_string, format))
ValueError: time data 'Thu, 10 Jun 2010 19:03:39 GMT' does not match format '%a, %d %b %Y %H:%M:%S GMT'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/home/antoine/py3k/debug/Lib/test/test_urllibnet.py", line 193, in test_data_header
    self.fail('Date value not in %r format', dateformat)
TypeError: fail() takes at most 2 arguments (3 given)

How do I enable `network' resource?

I am getting

$ ./python.exe -m test.regrtest test_urllibnet
test_urllibnet
test_urllibnet skipped -- Use of the `network' resource not enabled
1 test skipped:
    test_urllibnet
Those skips are all expected on darwin.


Same on Linux ...

> How do I enable `network' resource?

Use the "-unetwork" flag to regrtest.

$ ./python.exe -m test.regrtest -unetwork test_urllibnet
test_urllibnet
1 test OK.

Also the arguments to strptime from reported error message work fine:

>>> import time
>>> time.strptime('Thu, 10 Jun 2010 19:03:39 GMT', '%a, %d %b %Y %H:%M:%S GMT')
time.struct_time(tm_year=2010, tm_mon=6, tm_mday=10, tm_hour=19, tm_min=3, tm_sec=39, tm_wday=3, tm_yday=161, tm_isdst=-1)
>>> from datetime import *
>>> datetime.strptime('Thu, 10 Jun 2010 19:03:39 GMT', '%a, %d %b %Y %H:%M:%S GMT')
datetime.datetime(2010, 6, 10, 19, 3, 39)

What is your locale?

> What is your locale?

$ locale
LANG=fr_FR.utf8
LC_CTYPE="fr_FR.utf8"
LC_NUMERIC="fr_FR.utf8"
LC_TIME="fr_FR.utf8"
LC_COLLATE="fr_FR.utf8"
LC_MONETARY="fr_FR.utf8"
LC_MESSAGES="fr_FR.utf8"
LC_PAPER="fr_FR.utf8"
LC_NAME="fr_FR.utf8"
LC_ADDRESS="fr_FR.utf8"
LC_TELEPHONE="fr_FR.utf8"
LC_MEASUREMENT="fr_FR.utf8"
LC_IDENTIFICATION="fr_FR.utf8"
LC_ALL=

Strangely, it also works here from the prompt:

>>> import time
>>> time.strptime('Thu, 10 Jun 2010 19:03:39 GMT', '%a, %d %b %Y %H:%M:%S GMT')
time.struct_time(tm_year=2010, tm_mon=6, tm_mday=10, tm_hour=19, tm_min=3, tm_sec=39, tm_wday=3, tm_yday=161, tm_isdst=-1)
>>> import locale
>>> locale.getlocale(locale.LC_TIME)
(None, None)

Apparently, something sets the locale before running the test.

I still cannot reproduce the failure, but please, try the attached patch.

The culprit seems to be test_bad_address. If I disable this test, or replace the bad URL by a good one, everything works fine. It seems that failing to resolve the domain name changes the current locale...

The patch is not sufficient, since other failures can then occur if test_strptime gets run after test_urllibnet:

======================================================================
ERROR: test_twelve_noon_midnight (test.test_strptime.Strptime12AMPMTests)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/antoine/py3k/debug/Lib/test/test_strptime.py", line 389, in test_twelve_noon_midnight
    eq(time.strptime('12 PM', '%I %p')[3], 12)
  File "/home/antoine/py3k/debug/Lib/_strptime.py", line 461, in _strptime_time
    return _strptime(data_string, format)[0]
  File "/home/antoine/py3k/debug/Lib/_strptime.py", line 335, in _strptime
    data_string[found.end():])
ValueError: unconverted data remains: PM

======================================================================
FAIL: test_pattern (test.test_strptime.TimeRETests)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/antoine/py3k/debug/Lib/test/test_strptime.py", line 124, in test_pattern
    pattern_string)
AssertionError: did not find abbreviated weekday in pattern string '(?P<a>lun\.|mar\.|mer\.|jeu\.|ven\.|sam\.|dim\.)\s+(?P<A>mercredi|vendredi|dimanche|samedi|lundi|mardi|jeudi)\s+(?P<d>3[0-1]|[1-2]\d|0[1-9]|[1-9]| [1-9])'

======================================================================
FAIL: test_hour (test.test_strptime.StrptimeTests)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/antoine/py3k/debug/Lib/test/test_strptime.py", line 266, in test_hour
    (strf_output, strp_output[3], self.time_tuple[3]))
AssertionError: testing of '%I %p' directive failed; '08 ' -> 8 != 20

----------------------------------------------------------------------

BTW, looking at test_bad_address(), I wonder why it puts extra dots in the url?  The comment above it suggests that the intent is to use a name within RFC 2606 .invalid TLD.  Not likely to be a problem in your case, but using made up TLD is not a good idea.

> The patch is not sufficient, since other
> failures can then occur if test_strptime gets
> run after test_urllibnet

This looks like a bug in support.run_with_locale decorator.  

It is described as
#=======================================================================                                                                                                                                     
# Decorator for running a function in a different locale, correctly resetting                                                                                                                                
# it afterwards.

but apparently does not.

Ok, so what it boils down to is the following behaviour:

>>> import locale, socket
>>> sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
>>> locale.getlocale(locale.LC_TIME)
(None, None)
>>> sock.connect(("invalidhost", 80))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
socket.gaierror: [Errno -2] Nom ou service inconnu
>>> locale.getlocale(locale.LC_TIME)
('fr_FR', 'UTF8')

In setipaddr() in socketmodule.c, the following line appears to change the current locale when DNS lookup fails:

    error = getaddrinfo(name, NULL, &hints, &res);

This is checked by making a call to setlocale(LC_TIME, NULL) before and after the aforementioned line.

This is with the following libc:
$ rpm -qv glibc
glibc-2.11.1-8mnb2

This looks like a glibc bug to me. I suspect an unauthorized redhat change; I hope Ulrich Drepper would have never accepted a glibc that causes getaddrinfo to implicitly call setlocale - see

http://sources.redhat.com/ml/libc-alpha/2004-03/msg00161.html

You might experiment with disabling IDN support in getaddrinfo.

I've reported the bug upstream at Mandriva:
https://qa.mandriva.com/show_bug.cgi?id=59736

It would be nice to know whether other distributions with a Redhat lineage are affected. Can someone with such a distribution the code in msg107484?

In the end, I'm not sure it's Python's task to workaround such bug.
However, the bug in error reporting should be fixed.

It looks like your libc calls setlocale(LC_ALL, "") on error.  This may or may not be right and not python's problem in any case.  What is worth to investigate, however is why @run_with_locale decorator fails to restore the locale after it was modified by libc.

Antoine,

Would s/LC_TIME/LC_ALL/ in my patch fix your problem?  I could not find an affected system, but I simulated the problem by adding locale.setlocale(locale.LC_ALL, "") call in the test.

I think the patch is worth applying. It fixes the typo in the except clause and makes the test independent of the locale settings.

> Would s/LC_TIME/LC_ALL/ in my patch fix your problem?

As I explained, this wouldn't fix the later failures in test_strptime.

I tested the code from msg107484 on Fedora 16 with no change in locale.

Probably OK to close?

Closing this based on - msg155387. The upstream bug is "unfortunately" closed as OLD. We can reopen if this stumble upon this again.