Bug 924038

Created attachment 713575 [details]
log from second run, where install works

It really gets down into the weeds of setuptools, but I have an inkling the problem comes from the dependency in pytest-runner [1], which has a file ".hgtags" in it's zip.

[1] https://pypi.python.org/packages/source/p/pytest-runner/pytest-runner-1.2.zip

So here's the problem in a nutshell

keyring -> pytest-runner -> hgtools

hgtools is a "setup_requires" dependency for pytest-runner.  This
means the following code is run:

/usr/lib/python2.6/site-packages/setuptools/dist.py

---
class Distribution(_Distribution):
    def __init__():
     ...
        if attrs and 'setup_requires' in attrs:
            self.fetch_build_eggs(attrs.pop('setup_requires'))

    def fetch_build_eggs(self, requires):
        """Resolve pre-setup requirements"""
        from pkg_resources import working_set, parse_requirements
        for dist in working_set.resolve(
            parse_requirements(requires), installer=self.fetch_build_egg
        ):
            working_set.add(dist)
---

That goes fine, hgtools gets installed

---
  Running setup.py egg_info for package keyring
    zip_safe flag not set; analyzing archive contents...
    
    Installed /tmp/easy_install-QgHJjS/pytest-runner-1.2/hgtools-3.0.2-py2.6.egg
    
    Installed /root/build/keyring/pytest_runner-1.2-py2.6.egg
---

The temporary hgtools (via easy_install) has registered some entry
points with its distribution object in working_set.  However, it then
disappears but doesn't remove itself.  Later on in sdist.py we have

---
def walk_revctrl(dirname=''):
    """Find all files under revision control"""
    for ep in pkg_resources.iter_entry_points('setuptools.file_finders'):
        for item in ep.load()(dirname):
            yield item
----

pkg_resources.iter_entry_points walks working_set, finds hgtools'
entry points for this file_finders stuff, tries to call it and a
backtrace ensues because that temporary directory with hgtools has now
gone.

This also explains why it works when run the second time around;
pytest-runner is already there, so we don't need to looking at its
setup_requires and the problematic distribution doesn't get added.

Somehow or other, hgtools needs to get removed from the working_set in
dist.py before it is removed?

I have reported this upstream with 

http://bugs.python.org/setuptools/issue148

wrong component, changing to python-setuptools

Thanks to some help from the upstream bug, I now realise that rhel6.4 python-seutptools is distribute 0.6.10.  There are a series of patches attached to [1] which may be the solution to this.  Upstream bug mentions 0.6.19 was about the version it was fixed.

I confirmed that removing python-setuptools, installing upstream distribute (0.6.36) and then installing python-pip from epel with --nodeps does install keyring properly.  so it has been fixed at some point between 0.6.10 -> 0.6.36...

[1] https://bitbucket.org/tarek/distribute/issue/205/distributes-sandboxing-doesnt-preserve

This request was evaluated by Red Hat Product Management for
inclusion in the current release of Red Hat Enterprise Linux.
Because the affected component is not scheduled to be updated
in the current release, Red Hat is unable to address this
request at this time.

Red Hat invites you to ask your support representative to
propose this request, if appropriate, in the next release of
Red Hat Enterprise Linux.

Note: Link to the patch that fixes the issue: https://bitbucket.org/tarek/distribute/commits/856cd8712941d37257f8ef1d768564ad0e49e90a?at=fix-sdist