autopkgtests failures on armhf of 7.1.4

[Rico Tzschichholz]

The autopkgtests for 7.1.4~rc2-0ubuntu1 on impish/armhf has started to fail reliably.
https://autopkgtest.ubuntu.com/results/autopkgtest-impish/impish/armhf/libr/libreoffice/20210610_121847_0ab56@/log.gz

========

"/usr/lib/libreoffice/program/unopkg" add -f "/tmp/tmp.oOlP0i1pzN/out/sdk/LINUXexample.out/bin/SampleDialog.oxt"
#
# A fatal error has been detected by the Java Runtime Environment:
#
# SIGSEGV (0xb) at pc=0xf47474b8, pid=35616, tid=35643
#
# JRE version: OpenJDK Runtime Environment (11.0.12+4) (build 11.0.12-ea+4-Ubuntu-0ubuntu2)
# Java VM: OpenJDK Server VM (11.0.12-ea+4-Ubuntu-0ubuntu2, mixed mode, g1 gc, linux-)
# Problematic frame:
# C [libbootstraplo.so+0x1c4b8] stoc_impreg::mergeKeys(com::sun::star::uno::Reference<com::sun::star::registry::XRegistryKey> const&, com::sun::star::uno::Reference<com::sun::star::registry::XRegistryKey> const&, std::vector<stoc_impreg::(anonymous namespace)::Link, std::allocator<stoc_impreg::(anonymous namespace)::Link> >&)+0xb
#
# Core dump will be written. Default location: Core dumps may be processed with "/usr/share/apport/apport %p %s %c %d %P %E" (or dumping to /usr/lib/libreoffice/sdk/examples/DevelopersGuide/BasicAndDialogs/CreatingDialogs/core.35616)
#
# An error report file with more information is saved as:
# /tmp/hs_err_pid35616.log
#
# If you would like to submit a bug report, please visit:
# https://bugs.launchpad.net/ubuntu/+source/openjdk-lts
#
ERROR: Exception occurred: Binary URP bridge disposed during call ./binaryurp/source/bridge.cxx:613

ERROR: unopkg failed.

make[1]: Leaving directory '/usr/lib/libreoffice/sdk/examples/DevelopersGuide/BasicAndDialogs/CreatingDialogs'
make[1]: *** [Makefile:112: /tmp/tmp.oOlP0i1pzN/out/sdk/LINUXexample.out/misc/devguide_basicanddialogs_creatingdialogs_register_component.flag] Error 1
make[1]: Target 'ALL' not remade because of errors.
make: *** [/tmp/autopkgtest.U74TIg/build.myd/src/odk/CustomTarget_build-examples_java.mk:62: /tmp/autopkgtest.U74TIg/build.myd/src/workdir/CustomTarget/odk/build-examples_java.done] Error 2
make: Target 'CustomTarget_odk/build-examples_java' not remade because of errors.

========

While a re-run of the test for 7.1.3-0ubuntu1 were successful it suggests it is binary problem.
https://autopkgtest.ubuntu.com/results/autopkgtest-impish/impish/armhf/libr/libreoffice/20210609_234830_1265b@/log.gz

liberoffice is built with gcc-9/g++-9 on armhf to avoid https://bugs.launchpad.net/ubuntu/+source/gcc-10/+bug/1891623

A significant difference in the used toolchain points to the update of gcc-9

7.1.3-0ubuntu1 was built with gcc 9.3.0-24ubuntu1
https://launchpadlibrarian.net/537628130/buildlog_ubuntu-impish-armhf.libreoffice_1%3A7.1.3-0ubuntu1_BUILDING.txt.gz

7.1.4~rc2-0ubuntu1 was built with gcc 9.4.0-1ubuntu1
https://launchpadlibrarian.net/542492348/buildlog_ubuntu-impish-armhf.libreoffice_1%3A7.1.4~rc2-0ubuntu1_BUILDING.txt.gz

--------

Regarding 21.04/Hirsute

1:7.1.4~rc2-0ubuntu0.21.04.1~lo1 built and successfully ran autopkgtests, see https://bugs.launchpad.net/ubuntu/+source/libreoffice/+bug/1931272
If was using the gcc-9 9.3.0-23ubuntu2 available at the time.

The no-change rebuild against 9.4.0-1ubuntu1~21.04 results in the same failure now.
https://launchpadlibrarian.net/543392061/buildlog_ubuntu-hirsute-armhf.libreoffice_1%3A7.1.4~rc2-0ubuntu0.21.04.1~lo1_BUILDING.txt.gz
https://autopkgtest.ubuntu.com/results/autopkgtest-hirsute-libreoffice-experimental/hirsute/armhf/libr/libreoffice/20210612_113218_f7f69@/log.gz

[Matthias Klose (doko)]

why are you using GCC 9? We default to GCC 10 for a while, and are about to switch to GCC 11.

[Rico Tzschichholz (ricotz)]

@doko As mentioned above, liberoffice is built with gcc-9/g++-9 on armhf to avoid https://bugs.launchpad.net/ubuntu/+source/gcc-10/+bug/1891623

[Launchpad Janitor (janitor)]

This bug was fixed in the package libreoffice - 1:7.1.4-0ubuntu1

---------------
libreoffice (1:7.1.4-0ubuntu1) impish; urgency=medium

  * New upstream release
  * Use gcc 8 on armhf to avoid autopkgtest failure with gcc 9
    (LP: #1931701)

 -- Rico Tzschichholz <email address hidden> Thu, 17 Jun 2021 15:35:43 +0200

[Matthias Klose (doko)]

Using a three year old compiler is not a fix, but a work around. Please start investigating this properly.

https://bugs.launchpad.net/ubuntu/+source/gcc-10/+bug/1891623/comments/8
suggests to do a proper debug build at least. This was filed ten months ago ...

[Rico Tzschichholz (ricotz)]

A better workaround was applied to the hirsute SRU

https://git.launchpad.net/~libreoffice/ubuntu/+source/libreoffice/commit/?id=c832999071e9582ca7e4405cb7a7cfcbd656279d

[Sebastien Bacher (seb128)]

@Matthias, since that's a regression following a gcc update maybe you could help with the debugging?

[Matthias Klose (doko)]

@Sebastien, even identifying the exact change in gcc-9 won't be enough. It boils down to a miscompiled issue in two million lines of code. Already ten months ago, I asked to do proper debugging in LO, see LP: #1891623. However every bug I filed (comment #8) to get the debugging easier, or even possible was triaged as not affecting the user, and apparently delayed. There are two sides of the issue, one is to produce a test case below the two million line code mark, and below the 48 hour build mark. Helping yes, but doing that on my own, no.

[Rico Tzschichholz (ricotz)]

Bisecting this in gcc-9 points to https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=49cc1253d079bbefc18275f29adc526679422176

Preparing a patched gcc-9 based on 9.4.0-1ubuntu1 by reverting this specific change has confirmed it. libreoffice 7.1.4-0ubuntu1, using gcc-9 on armhf again, built and passed the failing test.

[Matthias Klose (doko)]

please don't revert this commit yet. Does this resolve the issue for gcc-10 and/or -11 as well?

[Matthias Klose (doko)]

the test case pr96796.c added in this commit passes on armhf. same for gcc-10. Please could you provide a test case?

[Rico Tzschichholz (ricotz)]

The problems with gcc-10 occurred way before 10.3.0/10.4.0, so this commit might cause the same regression in addition to the existing one. Note that the original PR96796 patch is different and was altered for gcc-9.

Of course the test-case for PR96796 passes, but the change itself causes the regression in libreoffice. As of now I still can't provide a basic reproducer.

[Rico Tzschichholz (ricotz)]

The problem is actually located in libbootstraplo.so as the trace already suggested.
Replacing the library with a "good" compilation confirmed it.

# Problematic frame:
# C [libbootstraplo.so+0x1c4b8] stoc_impreg::mergeKeys(com::sun::star::uno::Reference<com::sun::star::registry::XRegistryKey> const&, com::sun::star::uno::Reference<com::sun::star::registry::XRegistryKey> const&, std::vector<stoc_impreg::(anonymous namespace)::Link, std::allocator<stoc_impreg::(anonymous namespace)::Link> >&)+0xb

https://cgit.freedesktop.org/libreoffice/core/tree/stoc/source/implementationregistration/mergekeys.cxx?h=libreoffice-7-1#n54

[Launchpad Janitor (janitor)]

This bug was fixed in the package libreoffice - 1:7.1.5~rc2-0ubuntu1

---------------
libreoffice (1:7.1.5~rc2-0ubuntu1) impish; urgency=medium

  * New upstream release candidate
  * Bump yaru source to "2021-07-02"
  * Force noopt (thus also build with --disable-optimized) on armhf and
    build with default gcc (LP: #1931701)

 -- Rico Tzschichholz <email address hidden> Sun, 18 Jul 2021 20:52:33 +0200

[Matthias Klose (doko)]

this is still missing a test case

[Brian Murray (brian-murray)]

The Hirsute Hippo has reached End of Life, so this bug will not be fixed for that release.