Bugzilla Bug 230205

Tracker bug for all packages that need to be fixed before python 2.6 leaves
hard mask.

I had some issues with lxml and a couple of others, which I think is related to
distutils.  It seems the new one calls test.test_support (which seems to be a
valid python module), but lxml and at least one other package come with a
test.py (which then seems to take precedence in the namespace) and the
installation comes tumbling down.

The deprecation warnings are also a bit irritating, since pycrypto (which
portage doesn't actually need, but will use if present) uses md5 and sha1. 
They're both PEP 247 compliant, but hashlib (which is what's supposed to be
used instead of md5 and sha1) isn't, so it's not a simple patch.  The upstream
bug is at: https://bugs.launchpad.net/pycrypto/+bug/269429.

I'll start posting full bugs and patches once I've had a crack at fixing
them...

(In reply to comment #1)
> I'll start posting full bugs and patches once I've had a crack at fixing
> them...
> 

Thanks, your help is appreciated :-]

 * Starting Python Updater from 2.5 to 2.6 :
 *   Adding to list: =dev-python/gdata-1.0.8
 *   Adding to list: =dev-python/numeric-24.2-r6
 *   Adding to list: =dev-python/pyrex-0.9.8.5
 *   Adding to list: =dev-python/pycairo-1.4.12
 *   Adding to list: =x11-libs/vte-0.16.14
 *   Adding to list: =sys-apps/file-4.26

All of these packages, for me, all fail with attempting to write to the live
filesystem during install phase.  Except for vte - vte always shows up in
python-updater's output even though it successfully installs.

(In reply to comment #3)
> Except for vte - vte always shows up in
> python-updater's output even though it successfully installs.
> 

vte is in python-updater's manual list, use -dmanual to disable manual packages
from being reinstalled. I'll have a look at the other packages now.

All of these fail to merge on my system using python-2.6 because of access
violations during the installation phase:
dev-python/pyxml-0.8.4-r1
dev-python/setuptools-0.6_rc8-r1
dev-python/pyrex-0.9.8.5
dev-python/imaging-1.1.6
dev-python/pexpect-2.3
dev-java/java-config-2.1.6-r1
dev-java/java-config-1.3.7
sys-apps/file-4.26
www-client/pybugz-0.7.3
dev-python/ipython-0.8.4-r1
dev-python/django-1.0
dev-util/mercurial-1.0.2


Should I disable the sandbox as a workaround, is this a bug or is there
something I've missed which will make me look like an idiot?

Please ignore my previous comment.
All I had to do was a little eselect magic and the access violations
disappeared.
I'm deeply sorry for spamming and will report if I find any real breakage.

Don't disable the sandbox, that should be an absolutely last resort.

I found this a couple of times too, directly after the install.  Could you
please try running "eselect python list" to get a list of the python
environments you have installed, and remember the number in front of python-2.6
(in my example I'll use 2), and then run "eselect python set 2".  I've found
that this generally made the access violation issues and a few other issues go
away.  I haven't tracked down exactly why it helps, but I think it's some
crossed wires when the installer tries not to update symlinks...

ACCESS VIOLATION also with

sci-libs/vtk-5.2.0-r1

Please add depends on 240188 and 240149.  The latter contains an explanation
for the reason the bad symlinks cause distutils-based packages to fail to
build, and a workaround to fix those packages without disabling sandboxing.  It
does not yet have any info on why the symlinks are created improperly, tho.

I was unable to manually install zopeinterface 3.0.1 into Python 2.6.  However,
3.0.3 manually installed fine.  So it appears the problem has already been
fixed upstream and we just need a version bump.  There's already a bug
requesting a version bump of net-zope/zopeinterface (#191250); someone please
add it as a dependency of this bug.

Well, zopeinterface 3.0.1-r1 did install.

Rafal, I agree that ZopeInterface 3.0.1-r1 does work with Python 2.6-r2.  I
must have done something wrong before.  Sorry for the noise.

sci-libs/scipy does not compile, see bug #245100

dev-python/pywebkitgtk-1.0.2 doesn't merge either. 

Python 2.6 has been unmasked.

exaile-0.2.x works but throws warnings due to use of "import md5" instead of
hashlib.

There's a 0.3-rc in the tree but it's not even nearly feature-complete and so
no real replacement for the stable versions.

Want me to file a bug or are warnings not "enough"? 

(In reply to comment #16)
> Want me to file a bug or are warnings not "enough"? 

You can file such bugs and make them block this bug, but they won't block
stabilization of Python 2.6.*.