Created attachment 423018 [details] full build log TL;DR: A missing ruby executable aborts the configure, but fails to abort the ebuild, resulting in a merged package with zero(!!) files. So after the qt-5.5.1-r1 bumps I did my usual revdep-rebuild, and it said a couple packages needed rebuilt, but they failed to build as they couldn't find qtwebkit5-related cmake files. Further investigation showed I had a qtwebkit-5.5.1-r1 package installed... WITH NO FILES!! (Yes, I yelled a bit there, but IMO it's appropriate.) The binpkg I always build had no files in it either, but the package installed without error. Attempting a rebuild resulted in (last few lines of src_configure and src_compile): XHR Timeout ................... yes XSLT .......................... yes The WebKit build was disabled for the following reasons: * Missing ruby from PATH >>> Source configured. >>> Compiling source in /tmp/portage/dev-qt/qtwebkit-5.5.1-r1/work/qtwebkit-opensource-src-5.5.1 ... * Running emake make -j10 -l8 make: Nothing to be done for 'first'. >>> Source compiled. But equery list ruby says I have ruby-2.3.0 installed. However, there is indeed no "ruby" executable on the path, only "ruby23". ruby-2.3.0 was installed on Dec 28, based on the date, so presumably it upgraded but didn't run eselect-ruby. Running that and selecting the only ruby installed, ruby23, does indeed give me an on-path ruby executable, and qtwebkit:5 is rebuilding (and way past configure into a proper build, qtwebkit always takes foorrreeevverrrr to build... when it's actually building, at least) as I type this, but the bug here is that a missing ruby executable is indeed flagged as an error by the configure, but the ebuild doesn't consider that a /fatal/ error and thus goes on to blindly merge an empty package with absolutely no files! An error from configure should abort the ebuild entirely, so it never gets to the point of merging in the first place. Given the situation, emerge --info shouldn't be necessary as the problem is easily duplicated by simply removing the ruby executable from the path. I'm attaching the full build log, however. FWIW, I found the "checking 0 files for package collisions" bit rather... amusing, in a fatalistic sense. =:^p (The rebuild just finished... with files this time! =:^) And a quick check after renaming the ruby symlink, ctl-C-aborted in the fake-install step to avoid another real rebuild, confirms the above non-fatal ruby missing configure error behavior once again. With a qtwebkit:5 containing actual files merged, revdep-rebuild says everything's fine again, so no other rebuilds after all. =:^)
It could be tricky to detect a failed configure: !isEmpty(skipBuildReason) { log("$${EOL}The WebKit build was disabled for the following reasons: $$skipBuildReason $${EOL}$${EOL}") SUBDIRS= export(SUBDIRS) } else { log("$${EOL}WebKit is now configured for building. Just run 'make'.$${EOL}$${EOL}") }
Hmm, this guy ran into the same issue, empty package due to config failing to find the ruby executable, when testing to see if ruby was actually a required dep. He didn't realize he had a "successfully merged" empty package until later, and was thus fooled into believing ruby wasn't required (would that were so, as it's an expensive dep for only qtwebkit, here!). https://bugs.gentoo.org/show_bug.cgi?id=538450#c2
@ruby: I'd argue that either eselect-ruby isn't doing its job properly, or ruby ebuilds aren't calling it to update the symlink after an older ruby version is unmerged. Also, bug 564252 seems to be caused by the same root problem.
@ruby: ruby-2.3.0 does not depend on the latest version of eselect-ruby which added support for ruby23. Please fix ASAP.
so the issue with ruby was fixed in [1]... not sure if we want to do anything about the "silent failure". [1] https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=0cbbf13c21a144ceae18716b4f4647add409d081
*** Bug 572220 has been marked as a duplicate of this bug. ***
*** Bug 618036 has been marked as a duplicate of this bug. ***
Added a check in the overlay to die if nothing is built: https://gitweb.gentoo.org/proj/qt.git/commit/?id=604de1a5b113ef3b6c168f093b02d73e176ca57e
(In reply to Michael Palimaka (kensington) from comment #8) > Added a check in the overlay to die if nothing is built: > https://gitweb.gentoo.org/proj/qt.git/commit/ > ?id=604de1a5b113ef3b6c168f093b02d73e176ca57e you should use ${D%/}${QT5_LIBDIR} (and quoting is unnecessary)
Thanks, improved: https://gitweb.gentoo.org/proj/qt.git/commit/?id=5820fbc293c9ab58424bd4284d4823b881454581
Thanks, fixed in git. https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=7f211f960a4b79ce23572e2325c758b788bfdc45
Actually, that should have been "${ED%/}/...". Prefix got broken because of that. As you can see below, the library was installed, but the check failed. -- Set runtime path of "/var/cache/portage/dev-qt/qtwebkit-5.212.0_pre20180120/image//cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/lib64/libQt5WebKitWidgets.so.5.212.0" to "/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/x86_64-pc-linux-gnu/lib/gcc:/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/x86_64-pc-linux-gnu/lib:/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/lib64:/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/lib64" -- Installing: /var/cache/portage/dev-qt/qtwebkit-5.212.0_pre20180120/image/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/lib64/libQt5WebKit.so.5.212.0 -- Installing: /var/cache/portage/dev-qt/qtwebkit-5.212.0_pre20180120/image/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/lib64/libQt5WebKit.so.5 -- Installing: /var/cache/portage/dev-qt/qtwebkit-5.212.0_pre20180120/image/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/lib64/libQt5WebKit.so -- Set runtime path of "/var/cache/portage/dev-qt/qtwebkit-5.212.0_pre20180120/image//cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/lib64/libQt5WebKit.so.5.212.0" to "/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/x86_64-pc-linux-gnu/lib/gcc:/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/x86_64-pc-linux-gnu/lib:/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/usr/lib64:/cvmfs/sft.cern.ch/lcg/contrib/gentoo/linux/lib64" * dev-qt/qtwebkit-5.212.0_pre20180120 could not build due to a broken ruby environment. * ERROR: dev-qt/qtwebkit-5.212.0_pre20180120::gentoo failed (install phase): * Check "eselect ruby" and ensure you have a working ruby in your $PATH * * Call stack: * ebuild.sh, line 124: Called src_install * environment, line 4405: Called die * The specific snippet of code: * die 'Check "eselect ruby" and ensure you have a working ruby in your $PATH';
(In reply to Guilherme Amadio from comment #12) > Actually, that should have been "${ED%/}/...". Prefix got broken because of > that. As you can see below, the library was installed, but the check failed. No, ${D%/}${QT5_LIBDIR} is/was correct because QT5_LIBDIR already contains EPREFIX. The regression was introduced in the recently committed 5.212 version, which changed ${QT5_LIBDIR} into $(qt5_get_libdir), and the latter does not contain EPREFIX.
The bug has been referenced in the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=86b411ceb1cc8df7b359c05b8d3dd807a42051b6 commit 86b411ceb1cc8df7b359c05b8d3dd807a42051b6 Author: Davide Pesavento <pesa@gentoo.org> AuthorDate: 2018-07-05 16:46:37 +0000 Commit: Davide Pesavento <pesa@gentoo.org> CommitDate: 2018-07-05 17:06:11 +0000 dev-qt/qtwebkit: fix src_install sanity check on prefix Bug: https://bugs.gentoo.org/572056 Package-Manager: Portage-2.3.41, Repoman-2.3.9 dev-qt/qtwebkit/qtwebkit-5.212.0_pre20180120.ebuild | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
