bzr crashed with AttributeError in stopTest(): '_TypeEqualityDict' object has no attribute 'clear' - breaks bzr selftest on oneiric

[Jean-Louis Dupond]

Getting this error when trying to compile cia-clients

https://bugs.launchpad.net/bugs/808749

ProblemType: Crash
DistroRelease: Ubuntu 11.10
Package: bzr 2.4.0~beta4-4ubuntu2
ProcVersionSignature: Ubuntu 3.0.0-4.5-generic 3.0.0-rc6
Uname: Linux 3.0.0-4-generic x86_64
Architecture: amd64
BzrDebugFlags: set()
BzrVersion: 2.4b4
CommandLine: ['/usr/bin/bzr', 'selftest', '-v', '-s', 'bp.cia']
CrashDb: bzr
Date: Tue Jul 12 00:29:07 2011
ExecutablePath: /usr/bin/bzr
FileSystemEncoding: UTF-8
InstallationMedia: Ubuntu 11.04 "Natty Narwhal" - Alpha amd64 (20110119)
InterpreterPath: /usr/bin/python2.7
Locale: nl_BE.UTF-8
PackageArchitecture: all
Platform: Linux-3.0.0-4-generic-x86_64-with-Ubuntu-11.10-oneiric
ProcCmdline: /usr/bin/python /usr/bin/bzr selftest -v -s bp.cia
ProcEnviron:
 SHELL=/bin/bash
 LC_MESSAGES=nl_BE.utf8
 LANG=nl_BE.UTF-8
 LANGUAGE=nl_BE:nl:en_GB:en
PythonVersion: 2.7.2
SourcePackage: bzr
Title: bzr crashed with AttributeError in stopTest(): '_TypeEqualityDict' object has no attribute 'clear'
UpgradeStatus: Upgraded to oneiric on 2011-06-18 (23 days ago)
UserEncoding: UTF-8
UserGroups: adm admin audio cdrom dialout fuse libvirtd lpadmin netdev plugdev sambashare sbuild video

[Andrew Bennetts (spiv)]

oneiric's Python 2.7 is using an (unreleased) version of unittest which changes the behaviour of TestCase._type_equality_funcs a little. That's fine, in that it's supposed to be a private variable, but bzr is poking at it to try stop used TestCase instances from wasting memory.

bzr needs to become a little more careful in how it pokes at the private variable, and also be adjusted to poke it slightly differently so it will work with Python 2.7.3 (and oneiric's patched 2.7.2).

[Andrew Bennetts (spiv)]

I suppose alternatively we could try to get upstream to add a 'clear' method to the _TypeEqualityDict class before 2.7.3 is released.

[Martin Packman (gz)]

Upstream change is:
<http://hg.python.org/cpython/rev/9c337a4c650d/>

At least with that unittest stops storing bound methods in the dict, so the best upstream change would be to also rename the variable from `_type_equality_funcs` to `_type_equality_func_names` or similar. Perhaps Michael Foord could be persuaded to land such a change without me having to go through the pain of being yelled at for breaking the underscore taboo?

[Martin Packman (gz)]

Actually, it's not quite that easy as there's now a straight _TypeEqualityDict<->TestCase reference link. I'll see if I can get upstream to accept some kind of fix.

[Andrew Bennetts (spiv)]

Martin [gz] wrote:
> Actually, it's not quite that easy as there's now a straight
> _TypeEqualityDict<->TestCase reference link. I'll see if I can get
> upstream to accept some kind of fix.

In the interim we could add some ugly code to handle both versions:

    tef_clear = getattr(type_equality_funcs, "clear", None):
    if tef_clear is None:
        tef_instance_dict = getattr(type_equality_funcs, "__dict__", None)
        if tef_instance_dict is not None:
     tef_clear = tef_instance_dict.clear
    if tef_clear is not None:
        tef_clear()

Or perhaps we should just set the entire attribute to None, if it exists, and
rely on the assumption that by the time we do this self.assertEqual isn't going
to be invoked again:

    sentinel = object()
    if getattr(self, '_type_equality_funcs', sentinel) is not sentinel:
        self._type_equality_funcs = None

-Andrew.

[Jelmer Vernooij (jelmer)]

This is preventing the upload of bzr 2.4b5 to oneiric at the moment.

[Launchpad Janitor (janitor)]

This bug was fixed in the package bzr - 2.4.0~beta5-2ubuntu1

---------------
bzr (2.4.0~beta5-2ubuntu1) oneiric; urgency=low

  * Merge from Debian. Remaining changes:
   + Drop build dependencies on python-{subunit,meliae,lzma,medusa}, which are
     not in main.
   + Disable support for running selftest in parallel (requires subunit).
  * Fix compatibility with newer versions of Python 2.7. LP: #809048
 -- Jelmer Vernooij <email address hidden> Thu, 14 Jul 2011 18:18:11 +0200

[Martin Pool (mbp)]

escalating to critical as a blocker for 2.4.0

could we merge the workaround for this into bzr 2.4?