Moving to newer Python versions

Contents

Moving to newer Python versions

With Lucid fast approaching, efforts to actually migrate Launchpad to Python 2.6 are getting under way.

Relevant tagged bugs: https://bugs.launchpad.net/launchpad-foundations/+bugs?field.tag=python-upgrade

Branches

https://code.launchpad.net/~launchpad-committers/launchpad/python2.6 -- collaboration branch created by maxb based on salgado's personal branch
https://code.launchpad.net/~salgado/shipit/shipit-python2.6

Known/Open issues

Things partially dealt with by Salgado, who's going on leave and might not be able to finish them:
- The failure when setting up the GoogleServiceLayer seems to be a consequence of the server answering with a errno.ECONNABORTED when it's not yet ready, so I've just changed the wait_for_service function to treat that error the same way it does for errno.ECONNREFUSED.
- The test hang is because of the change done on bug 349316, so I've just commented out that addCleanup() line for now, to see if we can get a full test run.

I ran the test suite with devel r10700, the python2.6 branch, and the two Python 2.6-related branches that landed overnight.

Full log is here: tests-20100414.log

Total: 25792 tests, 25 failures, 9 errors in 204 minutes 11.495 seconds.

Unfortunately I forgot to stop my launchpad-buildd before running the suite, so some test failures are spurious:

test_manager
buildd-dispatching.txt
BuilddSlaveTestSetup
translationtemplatesbuildbehavior.txt

And three are known Lucid failures:

xx-distribution.txt (fixed in lazr.restful, but not yet used by LP)
soyuz-upload.txt (apt-ftparchive hash changes)
test_generateConfig (ditto)

Old results

With the two changes above I was able to get the test suite to run to completion: Total: 25856 tests, 71 failures, 119 errors in 300 minutes 55.792 seconds.

Here's the full log: tests.log

This appears to have been run without the Lucid pygpgme fix, so there are lots of "Unknown error code" failures in Soyuz and a couple of Registry tests. I'm rerunning the Soyuz-related tests to see what the real failures are. -- wgrant

Things needing revising

A multitude of modules of modules (sha, md5, sets, ...) are deprecated in Python 2.6, and many tests would fail merely because of unexpected DeprecationWarnings, so we've had to jump through multiple hoops all over the place to suppress these. Ultimately we need a plan for removing these suppressions, because we want to know about deprecation long-term.

Setup caveat

The extension modules in your sourcecode directory will have been built for Python 2.5 and these will break for your 2.6 branch. The simple thing to do is to move your old sourcecode aside and run utilities/rocketfuel-get to get it again with Python 2.6. A more complex but useful thing to do to manage sourcecode directories for multiple python versions is:
```
cd ~/launchpad/lp-sourcedeps
mkdir 2.6
cd 2.6
ln -s ../download-cache
ln -s ../eggs
mkdir sourcecode
cd sourcecode
for i in ../../sourcecode/*; do bzr checkout --lightweight "$i"; done
```
and now point utilities/link-external-sourcecode to ../../lp-sourcedeps/2.6 - if you do that you'll need to be aware that rocketfuel-get won't know about it, and your extra directories will require occasional manual updating.

Cleanup potential

Is there any reason we need ipython-0.9.1-py2.5.egg in egg, not tarball, form?
- Presumably we could bin the wadllib 0.1 eggs too.
Eventually we should get rid of all the scripts hardcoded /usr/bin/pythonX.Y and use buildout's shebang munging.
scripts/branch-rewrite.py uses /usr/bin/pythonX.Y -u. We should figure out if this is significant and why this script has to be different from other scripts.
scripts/ftpmaster-tools/_syncorigins.py uses /usr/bin/env pythonX.Y... why?!
cronscripts/mirror-prober.sh and cronscripts/nightly.sh are invoking python scripts with explicit interpreters. Why are they not using the shebangs? This leads to versions being defined in two places.

Known test failures and notes thereon

These results assume you:

Use the lp:~launchpad-committers/launchpad/python2.6 branch

Have a locally patched sourcecode/pygpgme:

--- src/pygpgme-context.c       2008-10-16 07:44:55 +0000
+++ src/pygpgme-context.c       2010-02-19 01:48:46 +0000
@@ -92,6 +92,8 @@
         return -1;
     }
 
+    gpgme_check_version(NULL);
+
     if (pygpgme_check_error(gpgme_new(&self->ctx)))
         return -1;

Do not have python-tdb installed, which otherwise causes bzr-hg and bzr-git to behave in ways the tests do not expect.
Have the following not-yet-landed branches merged:
- lp:~maxb/launchpad/trivial-bad-httpcaller-constructor-invocation
- lp:~maxb/launchpad/py2.6-warnings-monkeypatching

BaseLayer: OK (281 tests)

MemcachedLayer: OK (8 tests)

LibrarianLayer: OK (9 tests)

DatabaseLayer: OK (46 tests)

LaunchpadLayer: OK (24 tests)

ZopelessLayer: OK (20 tests)

MailmanLayer: ??

Note: These do not run automatically. Use bin/test --layer=MailmanLayer
You may get failures when you run all the tests in one go. Try running each individually for more accurate results.

LaunchpadScriptLayer: OK

LaunchpadZopelessLayer

test_generateConfig (lp.archivepublisher.tests.test_ftparchive.TestFTPArchive): Is benign, caused by apt-ftparchive generating SHA-1 and SHA-256 checksums in more places in Lucid.

test_bug_496988 (lp.bugs.scripts.tests.test_checkwatches.TestCheckwatchesWithSyncableGnomeProducts)

Traceback (most recent call last):
_StringException: Text attachment: traceback
------------
Traceback (most recent call last):
  File "/home/maxb/launchpad/lp-sourcedeps/eggs/testtools-0.9.2-py2.6.egg/testtools/runtest.py", line 128, in _run_user
    return fn(*args)
  File "/home/maxb/launchpad/lp-sourcedeps/eggs/testtools-0.9.2-py2.6.egg/testtools/testcase.py", line 368, in _run_test_method
    testMethod()
  File "/home/maxb/launchpad/lp-branches/real-python-2.6/lib/lp/bugs/scripts/tests/test_checkwatches.py", line 109, in test_bug_496988
    gnome_bugzilla, [bug_watch_1, bug_watch_2])
  File "/home/maxb/launchpad/lp-branches/real-python-2.6/lib/lp/bugs/scripts/checkwatches/updater.py", line 522, in _getExternalBugTrackersAndWatches
    remote_bug_ids))
  File "/home/maxb/launchpad/lp-branches/real-python-2.6/lib/lp/bugs/externalbugtracker/bugzilla.py", line 617, in getProductsForRemoteBugs
    self.initializeRemoteBugDB(bug_ids)
  File "/home/maxb/launchpad/lp-branches/real-python-2.6/lib/lp/bugs/externalbugtracker/isolation.py", line 64, in wrapper
    return func(*args, **kwargs)
  File "/home/maxb/launchpad/lp-branches/real-python-2.6/lib/lp/bugs/externalbugtracker/bugzilla.py", line 568, in initializeRemoteBugDB
    'permissive': True,
  File "/usr/lib/python2.6/xmlrpclib.py", line 1199, in __call__
    return self.__send(self.__name, args)
  File "/usr/lib/python2.6/xmlrpclib.py", line 1489, in __request
    verbose=self.__verbose
  File "/home/maxb/launchpad/lp-branches/real-python-2.6/lib/lp/bugs/externalbugtracker/xmlrpc.py", line 98, in request
    response = self._parse_response(self.opener.open(request), None)
  File "/usr/lib/python2.6/xmlrpclib.py", line 1392, in _parse_response
    return u.close()
  File "/usr/lib/python2.6/xmlrpclib.py", line 838, in close
    raise Fault(**self._stack[0])
Fault: <Fault Client: "Content-Type must be 'text/xml,' 'multipart/*,' or 'application/dime' instead of 'application/x-www-form-urlencoded'">
------------

lib/canonical/launchpad/ftests/../../../lp/bugs/tests/bugs-emailinterface.txt -- tests for http://bugs.python.org/issue7082, which has been fixed.

lib/canonical/launchpad/ftests/../doc/emailauthentication.txt

Failure in test lib/canonical/launchpad/ftests/../doc/emailauthentication.txt
Failed doctest test for emailauthentication.txt
  File "lib/canonical/launchpad/ftests/../doc/emailauthentication.txt", line 0

----------------------------------------------------------------------
File "lib/canonical/launchpad/ftests/../doc/emailauthentication.txt", line 150, in emailauthentication.txt
Failed example:
    print msg.as_string() #doctest: -NORMALIZE_WHITESPACE
Differences (ndiff with -expected +actual):
    - Date:...
    - ...
    + Date: Wed, 13 Apr 2005 19:18:42 -0400
    + From: Sample Person <test@canonical.com>
    + To: someone@somhost
    + Subject: Signed Email
    + Message-ID: <20050413231842.GB16989@localhost.localdomain>
    + Mime-Version: 1.0
    + Content-Type: multipart/signed; micalg=pgp-sha1;
    +   protocol="application/pgp-signature"; boundary="8NvZYKFJsRX2Djef"
    + Content-Disposition: inline
    + Status: RO
    + <BLANKLINE>
    + <BLANKLINE>
    + --8NvZYKFJsRX2Djef
    + Content-Type: multipart/mixed;
    - Content-Type: multipart/mixed; boundary="--------------------EuxKj2iCbKjpUGkD"
    ? ------------------------------
    +  boundary="--------------------EuxKj2iCbKjpUGkD"
    - ...
    + Content-Disposition: inline
    + <BLANKLINE>
    + <BLANKLINE>
    + ----------------------EuxKj2iCbKjpUGkD
    + Content-Type: text/plain; charset=us-ascii
    + Content-Disposition: inline
    + <BLANKLINE>
    + Some signed content.
    + <BLANKLINE>
    + ----------------------EuxKj2iCbKjpUGkD
    + Content-Type: text/plain; charset=us-ascii
    + Content-Disposition: attachment; filename="test.txt"
    + <BLANKLINE>
    + A signed attachment.
    + <BLANKLINE>
    + ----------------------EuxKj2iCbKjpUGkD--
    + <BLANKLINE>
    + --8NvZYKFJsRX2Djef
    + Content-Type: application/pgp-signature; name="signature.asc"
    + Content-Description: Digital signature
    + Content-Disposition: inline
    + -----BEGIN PGP SIGNATURE-----
    + Version: GnuPG v1.4.1 (GNU/Linux)
    + <BLANKLINE>
    + iD8DBQBDoAFKuiuTid/SBUMRAvzNAKCPHJgldGMFXh5SvKEufqB8N6IkoACdEu+8
    + Go2JWmGWaY8Nqquwk/Tr0pU=
    + =OhN7
    + -----END PGP SIGNATURE-----
    + <BLANKLINE>
    + --8NvZYKFJsRX2Djef--
    + <BLANKLINE>
    + <BLANKLINE>
    + <BLANKLINE>
    + <BLANKLINE>

FunctionalLayer: OK (174 tests)

LaunchpadFunctionalLayer

lib/canonical/launchpad/ftests/../doc/profiling.txt -- HTTPCaller takes no constructor arguments but the doctest tries to give it one
valid_absolute_url (canonical.launchpad.validators.url) -- Python 2.6's urlparse disregards the uses_netloc list when parsing urls and treats anything beginning with // as netloc based. This makes maxb and wgrant say WTF!?! The untested valid_absolute_url DB constraint is also affected!
lib/lp/soyuz/tests/../doc/soyuz-upload.txt - Is benign, caused by apt-ftparchive generating SHA-1 and SHA-256 checksums in more places in Lucid.

AppServerLayer: OK (143 tests)

DatabaseFunctionalLayer: OK (4461 tests)

GoogleLaunchpadFunctionalLayer: OK (72 tests)

PageTestLayer

TwistedLayer: OK (235 tests)

TwistedLaunchpadZopelessLayer: OK (16 tests)

TwistedAppServerLayer

ZopelessAppServerLayer

test_hosted_branch_stacked_on_mirrored_branch (lp.codehosting.puller.tests.test_acceptance.TestBranchPuller): LayerIsolationError: Test left new live threads: [<Thread(Thread-3, started daemon 47618947405584)>]

ZopelessDatabaseLayer: OK

BugsWindmillLayer: OK (12 tests)

CodeWindmillLayer: OK (11 tests)

ForeignBranchPluginLayer: OK

SSHServerLayer: OK (46 tests)

RegistryWindmillLayer: OK (17 tests)

SoyuzWindmillLayer: OK (2 tests)

TestZopeTestInSubProcessLayer: OK (1 test)

TranslationsWindmillLayer: OK (7 tests)

UnitTests

562734 RuntimeError: The wrong site.py is imported. Scripts should usually be started with Launchpad's bin/py, or with a Python invoked with the -S flag. breaking some Rosetta scripts run from tests, seemingly only if lp.services.scripts.test_all_scripts ran earlier in the same test run:
- test_script (lp.translations.scripts.tests.test_validate_translations_file.TestValidateTranslationsFile) (says AssertionError: 0 != 1, I added some print statements locally to see the cause of the failure)
- lib/lp/translations/tests/../doc/remove-upstream-translations-script.txt

Shipit

Might be ok now, I've fixed the errors that were listed here, but I don't have access to shipit --maxb.

launchpad development

PythonMigrationStatus