When i make a stage1 build with catalyst, in the image tarball the /usr/bin/python file points to /tmp/stage1root/usr/bin/python2.6 This is wrong, it seems this command ends with broken symlinks: ROOT="/tmp/stage1root" emerge =python-2.6.2 It failed with python-2.6.1* also, but this worked with python-2.5* ! Reproducible: Always
Please post your `emerge --info' and attach the build log that ends up producing the bad symlink. Also, it's not a blocker.
As i mentioned, the problem is the symlink handling change in the ebuilds. In python-2.5.4.ebuild: pkg_postinst() { local myroot myroot=$(echo $ROOT | sed 's:/$::') local mansuffix=$(ecompress --suffix) python_makesym alternatives_auto_makesym "/usr/bin/idle" "idle[0-9].[0-9]" alternatives_auto_makesym "/usr/bin/pydoc" "pydoc[0-9].[0-9]" alternatives_auto_makesym "/usr/bin/python-config" \ "python-config-[0-9].[0-9]" and in the python.eclass: # @FUNCTION: python_makesym # @DESCRIPTION: # Run without arguments, it will create the /usr/bin/python symlinks # to the latest installed version python_makesym() { alternatives_auto_makesym "/usr/bin/python" "python[0-9].[0-9]" alternatives_auto_makesym "/usr/bin/python2" "python2.[0-9]" } But in the python-2.6.2.ebuild: pkg_postinst() { eselect python update --ignore 3.0 python_version python_mod_optimize -x "(site-packages|test)" \ /usr/$(get_libdir)/python${PYVER} } So, the 2.5.4's ebuild uses the methods in python.eclass, but the 2.6.2's ebuild uses the eselect. But eselect didnt know anythng about the ROOT environment variable. And i think this is a blocking, as i cannot make gentoo stage1 images..
Created attachment 189870 [details, diff] patch for /usr/share/eselect/modules/python.eselect patch for /usr/share/eselect/modules/python.eselect this fixes the symlinker ln commands arguments, just as the way it did in binutils.eselect, and in others.
Attached a patch what fixes this bug. Please use this with eselect-python (as i saw, the original developer hawking retired..).
I have no idea why I was CC'ed on this bug. In case I'm needed or have something to do, please CC me back.
This bug should be fixed in trunk of eselect-python.
Ok, i looked for the trunk version of eselect-python (i didnt know there is another repo for this), and it is fixed yes. But when will be this in portage? Portage have this bug, so i reopened the bug report.
I'm in python@gentoo.org alias, so you don't need to CC me. (In reply to comment #7) > But when will be this in portage? Soon...
I'm getting this bug in my stage1 builds. I've masked out python-3, so python-2.7.1 is the only version being installed. It doesn't seem like the logic should be so tricky for choosing where the python symlink should be pointing. :)
(In reply to comment #9) It's unrelated, different, known problem.