[KDE Itinerary] [Bug 442821] New: TimelineModelTest::testWeatherElements fails at 23:00
https://bugs.kde.org/show_bug.cgi?id=442821 Bug ID: 442821 Summary: TimelineModelTest::testWeatherElements fails at 23:00 Product: KDE Itinerary Version: unspecified Platform: Other OS: Linux Status: REPORTED Severity: normal Priority: NOR Component: general Assignee: vkra...@kde.org Reporter: kdebugs...@lsmod.de Target Milestone: --- SUMMARY timelinemodeltest.cpp fails reproducibly 4% of the time STEPS TO REPRODUCE 1. run itinerary-21.08.1 tests with the system clock set between 23:00 and 23:59 UTC (in case it matters: we use TZ=UTC) on openSUSE, you can do that with osc co openSUSE:Factory/itinerary && cd $_ && osc build --vm-type=kvm --noservice --clean --build-opt=--vm-custom-opt="-rtc base=2021-11-01T23:29:00" standard OBSERVED RESULT QDEBUG : TimelineModelTest::testWeatherElements() weather recomputation done^M FAIL! : TimelineModelTest::testWeatherElements() Compared values are not the sa me^M Actual (fc.dateTime().date()): 2021/09/05^M Expected (QDate::currentDate()): 2021/09/04^M Loc: [/home/abuild/rpmbuild/BUILD/itinerary-21.08.1/autotests/timelinemodelte st.cpp(284)]^M QWARN : TimelineModelTest::testMultiTraveller() Comparators already registered for type KItinerary::BusTrip^M EXPECTED RESULT tests should not fail like this SOFTWARE/OS VERSIONS Linux/KDE Plasma: openSUSE Tumbleweed 20210916 KDE Plasma Version: ? KDE Frameworks Version: 5.86 Qt Version: 5.15.2 ADDITIONAL INFORMATION This bug was found while working on reproducible builds for openSUSE. -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 401742] unreproducible .a files should not be built with LTO
https://bugs.kde.org/show_bug.cgi?id=401742 --- Comment #12 from Bernhard M. Wiedemann --- According to our compiler guys, such .a files should then be created with -ffat-lto-objects so that they contain actual machine code that can be used without having the same gcc version and lto-plugins. In openSUSE, header files and .a files are mixed together in a valgrind-devel package and that is used in builds of anjuta ceph gperftools libqt5-qtwebengine mpich rdma-core ruby2.5 wine xmms2 But maybe they only use the .h files. I quickly tested xmms2, rdma-core, gperftools, anjuta and none of the 4 uses .a files. Anyway, such a patch seems useful: valgrind-3.14.0/configure.ac: -TEST_LTO_CFLAGS="-flto -flto-partition=one -fuse-linker-plugin" +TEST_LTO_CFLAGS="-flto -flto-partition=one -fuse-linker-plugin -ffat-lto-objects" That would allow us to strip the LTO sections from the .a files and still have the normal machine code. Or we just drop the .a files from our rpms since nobody seems to use them. -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 401742] unreproducible .a files should not be built with LTO
https://bugs.kde.org/show_bug.cgi?id=401742 --- Comment #9 from Bernhard M. Wiedemann --- According to our (SUSE) compiler guys, the .o files created with -flto are only intermediate artifacts and not meant to be shipped to users. So maybe the better fix would then be to not ship those .a files containing them? It would also be nice if gcc could produce deterministic results there, but apparently they need some unique ID and it is hard (from code or design) to hash inputs or such. -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 401742] unreproducible .a files should not be built with LTO
https://bugs.kde.org/show_bug.cgi?id=401742 --- Comment #6 from Bernhard M. Wiedemann --- The patch only disables LTO for .a file creation where it does not make sense to have it. Shared objects and such still use LTO with the patch. grep -- "-flto " finds 134 occurrences in the build log with applied patch. -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 401742] unreproducible .a files should not be built with LTO
https://bugs.kde.org/show_bug.cgi?id=401742 --- Comment #4 from Bernhard M. Wiedemann --- Created attachment 116673 --> https://bugs.kde.org/attachment.cgi?id=116673=edit fixing patch -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 401742] unreproducible .a files should not be built with LTO
https://bugs.kde.org/show_bug.cgi?id=401742 Bernhard M. Wiedemann changed: What|Removed |Added CC||philippe.waroquiers@skynet. ||be -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 401742] unreproducible .a files should not be built with LTO
https://bugs.kde.org/show_bug.cgi?id=401742 --- Comment #1 from Bernhard M. Wiedemann --- Submitted a fix at https://phabricator.kde.org/D17349 -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 401742] New: unreproducible .a files should not be built with LTO
https://bugs.kde.org/show_bug.cgi?id=401742 Bug ID: 401742 Summary: unreproducible .a files should not be built with LTO Product: valgrind Version: 3.14.0 Platform: Other OS: Linux Status: REPORTED Severity: normal Priority: NOR Component: general Assignee: jsew...@acm.org Reporter: kdebugs...@lsmod.de Target Milestone: --- SUMMARY While working on reproducible builds for openSUSE, I found that the valgrind package varied for every build because its .a files contain .o files created by gcc -flto -c Such .o files are gcc version specific and not very useful. Also they always differ from randomness introduced by gcc, making it impossible to verify that the build was correct. See also https://bugzilla.opensuse.org/show_bug.cgi?id=1118163 STEPS TO REPRODUCE 1. build openSUSE Factory valgrind package 2 times 2. compare results OBSERVED RESULT .a files differ EXPECTED RESULT It should be possible to generate identical build results multiple times. ADDITIONAL INFORMATION I got a patch to fix this and will upload it to Phabricator. -- You are receiving this mail because: You are watching all bug changes.
[frameworks-kdoctools] [Bug 398263] kdoctools man page build should depend on all-l10n.xml
https://bugs.kde.org/show_bug.cgi?id=398263 --- Comment #2 from Bernhard M. Wiedemann --- Yes, that patch fixes the race. -- You are receiving this mail because: You are watching all bug changes.
[frameworks-kdoctools] [Bug 398263] New: kdoctools man page build should depend on all-l10n.xml
https://bugs.kde.org/show_bug.cgi?id=398263 Bug ID: 398263 Summary: kdoctools man page build should depend on all-l10n.xml Product: frameworks-kdoctools Version: 5.49.0 Platform: Other OS: Linux Status: UNCONFIRMED Severity: normal Priority: NOR Component: general Assignee: kde-doc-engl...@kde.org Reporter: kdebugs...@lsmod.de CC: kdelibs-b...@kde.org Target Milestone: --- While working on reproducible builds for openSUSE, I found that there is a parallelism-related race: When building openSUSE's kdoctools package with a 1-core-VM using osc build -j1 it creates different man-pages than with a 4-core-VM osc build -j4 diffs include: --- old//usr/share/man/man1/checkXML5.1 2018-08-26 03:25:41.086296212 + +++ new//usr/share/man/man1/checkXML5.1 2018-08-26 03:25:41.098296086 + @@ -5,9 +5,9 @@ .\" Date: 2014-03-04 .\"Manual: checkXML5 User's Manual .\"Source: KDE Frameworks Frameworks 5.0 -.\" Language: English +.\" Language: -.SH "NAME" +.SH "" checkXML5 \- An XML lint tool for KDE DocBook XML documents\&. -.SH "SYNOPSIS" +.SH "" .HP \w'\fBcheckXML5\fR\ 'u \fBcheckXML5\fR [\fIfilename\fR] -.SH "DESCRIPTION" +.SH "Description" I think, this comes from a I/O warning : failed to load external entity "/home/abuild/rpmbuild/BUILD/kdoctools-5.49.0/build/src/customization/xsl/all-l10n.xml" that only happens in the -j1 case in the output of kdoctools-5.49.0/build/bin/meinproc5 Looking through the source, all-l10n.xml seems to be generated by docbookl10n but there is nothing telling cmake about this dependency, so it is scheduled at different times before/after the man pages get generated: difflog | grep "Built target docbookl10n" + [ 8%] Built target docbookl10nhelper_autogen + [ 25%] Built target docbookl10nhelper - [ 94%] Built target docbookl10nhelper_autogen - [100%] Built target docbookl10nhelper -- You are receiving this mail because: You are watching all bug changes.
