Bug#1077181: src:gtk4: Multiple test failures on RISC-V

[Simon McVittie]

Control: severity -1 serious
Control: tags -1 + experimental
Control: block 1072395 by -1
Control: retitle -1 gtk4: FTBFS on riscv64: a11y/text, a11y/textview tests segfault

This is RC for experimental due to the test failure causing a FTBFS.
We can downgrade it to a non-RC severity if a workaround is found.

On Fri, 26 Jul 2024 at 14:53:03 +0200, Matthias Geiger wrote:
> 603/668 gtk:reftest / reftest label-shadows.ui                                       FAIL              7.64s   0/1 subtests passed

This seems to be the same as #1077178 on mips64el, so let's use #1077178 to
represent that failure on both of the affected architectures, and #1077181
to represent only the gtk/a11y / text{,view} crashes discussed below.

> The a11y tests just seem to segfault:
> 
> 478/668 gtk:a11y / text                                                              ERROR             0.73s   killed by signal 11 SIGSEGV
> >>> GTK_A11Y=test GSETTINGS_BACKEND=memory MESON_TEST_ITERATION=1 ASAN_OPTIONS=halt_on_error=1:abort_on_error=1:print_summary=1 TEST_OUTPUT_SUBDIR=wayland G_TEST_BUILDDIR=/<<PKGBUILDDIR>>/debian/build/deb/testsuite/a11y GSETTINGS_SCHEMA_DIR=/<<PKGBUILDDIR>>/debian/build/deb/gtk UBSAN_OPTIONS=halt_on_error=1:abort_on_error=1:print_summary=1:print_stacktrace=1 MSAN_OPTIONS=halt_on_error=1:abort_on_error=1:print_summary=1:print_stacktrace=1 G_ENABLE_DIAGNOSTIC=0 GDK_DEBUG=default-settings GTK_CSD=1 G_TEST_SRCDIR=/<<PKGBUILDDIR>>/testsuite/a11y GDK_BACKEND=wayland MALLOC_PERTURB_=222 GSK_RENDERER=cairo /<<PKGBUILDDIR>>/debian/build/deb/testsuite/a11y/text --tap -k
> 
> 479/668 gtk:a11y / textview                                                          ERROR             0.71s   killed by signal 11 SIGSEGV
> >>> GTK_A11Y=test GSETTINGS_BACKEND=memory MESON_TEST_ITERATION=1 ASAN_OPTIONS=halt_on_error=1:abort_on_error=1:print_summary=1 TEST_OUTPUT_SUBDIR=wayland G_TEST_BUILDDIR=/<<PKGBUILDDIR>>/debian/build/deb/testsuite/a11y GSETTINGS_SCHEMA_DIR=/<<PKGBUILDDIR>>/debian/build/deb/gtk MALLOC_PERTURB_=80 UBSAN_OPTIONS=halt_on_error=1:abort_on_error=1:print_summary=1:print_stacktrace=1 MSAN_OPTIONS=halt_on_error=1:abort_on_error=1:print_summary=1:print_stacktrace=1 G_ENABLE_DIAGNOSTIC=0 GDK_DEBUG=default-settings GTK_CSD=1 G_TEST_SRCDIR=/<<PKGBUILDDIR>>/testsuite/a11y GDK_BACKEND=wayland GSK_RENDERER=cairo /<<PKGBUILDDIR>>/debian/build/deb/testsuite/a11y/textview --tap -k

A crash like this seems a lot more serious than some mis-rendering.

Can someone reproduce this crash on a riscv64 porterbox and get a
backtrace from these two crashes?

If we switch from llvmpipe to softpipe:

------
-ifneq ($(filter mips%,$(DEB_HOST_ARCH_CPU)),)
-$(info Disabling use of llvmpipe due to mips family CPU, see https://bugs.debian.org/993550)
+ifneq ($(filter mips% riscv64,$(DEB_HOST_ARCH_CPU)),)
+$(info Disabling use of llvmpipe on this CPU, see https://bugs.debian.org/993550)
 export GALLIUM_DRIVER=softpipe
 endif
------

does that maybe work around it?

Another relevant question is: if a riscv64 porter tries compiling gtk4
from experimental (with DEB_BUILD_OPTIONS=nocheck or with the affected
tests otherwise skipped), do GTK4 apps like gnome-control-center run
successfully on practical end-user riscv64 hardware?

(in other words, is this crash specific to the buildd environment,gor
does it affect real apps running in a real desktop environment?)

    smcv