Log from #gentoo-qa.

irc.txt (text/plain), 7.38 KB, created by Samuli Suominen (RETIRED) on 2009-10-15 20:26:33 UTC

(I feel so stupid repeating myself here.)

22:26 < Arfrever> ssuominen: You probably failed to rebuild qscintilla-python's dependencies (sip, qscintilla, PyQt4) with Python 2 and
                  Python 3.
22:27 < ssuominen> I didn't see any message from PM or eclass that it's required
22:28 < ssuominen> so I didn't fail... the design did
22:29 < Arfrever> ssuominen: ABI dependencies haven't been implemented yet in Portage.
22:29 < ssuominen> meanwhile it's OK to let python3 break e.g. kde4 installation? perhaps i'd be wise to p.mask it until it's fixed
22:30 < Arfrever> ssuominen: How does it break KDE 4 installation? I have successfully installed KDE 4.3.2.
22:30 < ssuominen> kdebindings doesn't build
22:30 < ssuominen> for the same reason
22:30 < ssuominen> if python is updated in-between
22:31 < ssuominen> s/updated/python 3 is installed/
22:31 < Arfrever> ssuominen: What package name?
22:34 < ssuominen> hrm, sec...
22:34 < Arfrever> ssuominen: The only possible workaround for absence of ABI dependencies would be unconditional requirement of Python 2
                  and Python 3, and revbumping of some packages which are dependencies of some packages which support both Python 2 and
                  Python 3.
22:34 < Arfrever> ssuominen: I don't see 'kdebindings' package in the tree.
22:35 < ssuominen> krosspython I believe.
22:35 < Arfrever> (I don't have objections to unconditional requirement of Python 2 and Python 3 :) .)
22:36 < ssuominen> If the alternatives are forcing python-3, or masking it, i'd go for the masking
22:36 < ssuominen> pyqt4, pykde4, scintilla-python and krosspython are the 4 pkgs that failed, IIRC
22:36 < ssuominen> all for the same reason, the one you just explained
22:37 < Arfrever> ssuominen: Does kde-base/krosspython support Python 3?
22:37 < ssuominen> unsure
22:38 < ssuominen> and that's not the point.
22:38 < ssuominen> 22:27 < ssuominen> I didn't see any message from PM or eclass that it's required
22:38 < ssuominen> still stands
22:38 < Arfrever> ssuominen: kde-base/krosspython doesn't support installation for multiple Python versions, so the problem with
                  kde-base/krosspython probably wasn't related to Python 3.
22:39 < ssuominen> it can't be left the way it is now, a consensus is required. fix the pm/eclass, force it, or mask it. and seeing how
                   forcing isn't an option, there's only one alternative
22:39 < ssuominen> ...until the first option is fulfilled
22:40 < Arfrever> ssuominen: Why do you think that forcing isn't an option?
22:40 < ssuominen> Because it's not required by any of the packages.
22:41 < ssuominen> Except by broken design
22:41 < ssuominen> and that's not a valid reason for a dep
22:41 < Arfrever> Design isn't broken.
22:42 < Arfrever> Only ABI dependencies are missing.
22:42 < ssuominen> You just explained it is:
22:42 < ssuominen> 22:26 < Arfrever> ssuominen: You probably failed to rebuild qscintilla-python's dependencies (sip, qscintilla, PyQt4)
                   with Python 2 and Python 3.
22:42 < ssuominen> I believe you got my point by now, this is pointless.
22:43 < ssuominen> Fix the pm or mask it
22:43 < ssuominen> It's causing random build failures as is
22:44 < ssuominen> I don't have anything against the current way, long as the ABI deps you explained are introduced first
22:44 < ssuominen> So it's ONLY a matter of the order you are introducing this change in tree
22:47 < Arfrever> ssuominen: I will fix the pm (Portage).
22:47 < ssuominen> Do you have any ETA for that? My point is simple: Mask it while working on it.
22:50 < ssuominen> Don't get me wrong. I'm not trying to offend you. It's just clearly broken as-is.
22:50 < Arfrever> ssuominen: I can also quicklier implement a function similar to built_with_use(), which would check if dependencies
                  have been built with the same versions of Python.
22:50  * ssuominen has nothing against that
22:51 < ssuominen> long as there's a message, instead of failure without a message.
22:51 < ssuominen> can't expect people to "just know about it", even though, i do now :)
22:54 < Arfrever> ssuominen: Also note that it's a problem with less than 5 packages.
22:55 < Arfrever> ssuominen: x11-libs/qscintilla doesn't need to be rebuilt. Only sip and PyQt4.
22:55 < ssuominen> and with current status, is there a way to fix this in those ebuilds?
22:56 < ssuominen> and of how many packages are those reverse deps of?
22:57 < Arfrever> ssuominen: To implement python_check_dependencies(), and add pkg_setup() { python_check_dependencies dev-python/sip
                  dev-python/PyQt4; }.
22:58 < Arfrever> ssuominen: 1 (which supports Python 2 and Python 3)
22:59 < Arfrever> ssuominen: So qscintilla-python has to check for sip and PyQt4, and PyQt4 has to check for sip.
22:59 < ssuominen> yeah, I believe it can be done
22:59 < ssuominen> that's why I asked for the ETA and masking it while working on that.
22:59 < Arfrever> Portage has optional dependency on Python 3.
23:00 < ssuominen> and?
23:00 < ssuominen> optional. can be package.use.masked.
23:00 < ssuominen> people who want to test it, can still do so
23:00 < ssuominen> and rest of us won't break.
23:01 < Arfrever> It's not intended only for testing.
23:01 < ssuominen> it can't be anything else before this is fixed
23:03 < ssuominen> i hope you're not proposing we introduce this breakage to stable tree :)
23:05 < Arfrever> ssuominen: Python 3 itself doesn't cause this problem or any other.
23:06 < ssuominen> they way it's in tree does. that's called packaging.
23:08 < Arfrever> ssuominen: It's caused by new algorithm in python.eclass. The same algorithm could concern e.g. Python 2.5 and Python
                  2.6. (Currently it concerns Python 2.* and Python 3.*.)
23:08 < ssuominen> Yes, it's clear by now why it happens. No need to go there again.
23:09 < Arfrever> So it's clear that it doesn't happen due to Python 3, so Python 3 shouldn't be blamed for random problems.
23:10 < ssuominen> It happens because the way it's in tree without support from PM
23:10 < ssuominen> As in, tree-incompatible
23:11 < ssuominen> The way you put a package in tree, is packaging, and if that's broken, you mask the package
23:11 < ssuominen> It's pretty simple...
23:11 < Arfrever> `USE_PYTHON=2.6 emerge -1 sip; USE_PYTHON="2.5 2.6" emerge -1 PyQt4` can also reproduce this problem.
23:12 < Arfrever> But Python 2.5 or 2.6 rather shouldn't be masked :) .
23:14 < ssuominen> There's python-updater to move from 2.5 to 2.6 to which users are custom to. That's no excuse to keep the breakage.
23:15 < Arfrever> The breakage, if it all exists, isn't caused by Python 3.1 more than by any other version of Python.
23:18 < ssuominen> http://dpaste.com/107839/
23:18 < ssuominen> http://dpaste.com/107840/
23:19 < ssuominen> Directly caused by the issue you, yourself explained above
23:19 < ssuominen> False packaging without support from PM
23:20 < ssuominen> !rdep PyQt4
23:20 < Arfrever> ssuominen: And I also explained that it happens also with e.g. 2.5 and 2.6.
23:20 < Willikins> ssuominen: Too many packages have reverse RDEPEND on dev-python/PyQt4, go to
                   http://tinderbox.dev.gentoo.org/misc/rindex/dev-python/PyQt4 instead.
23:20 < Arfrever> ssuominen: IIRC only 1 reverse dependency of PyQt4 supports Python 3.
23:21 < Arfrever> I.e. qscintilla-python.
23:21 < ssuominen> If you can't get PyQt4 installed, how can you install any of the reverse deps?
23:21 < ssuominen> I know I can workaround it. Why do you expect users can?