libjxl: 0.8.1: test suite is failing in ~17% of the test units
Describe the bug
Looks lik esomething is wrong with test suite or I’m doing something wrongl;y. On test suite exetion I have I have ~17% of failing units.
To Reproduce
cmake settings (cmake -L output):
-- Cache values
ASCIIDOC:FILEPATH=/usr/bin/a2x
BASH_PROGRAM:FILEPATH=/usr/bin/bash
BUILD_SHARED_LIBS:BOOL=ON
BUILD_TESTING:BOOL=ON
CMAKE_BUILD_TYPE:STRING=RelWithDebInfo
CMAKE_INSTALL_PREFIX:PATH=/usr
GMOCK_INCLUDE_DIR:PATH=/usr/include
GTest_DIR:PATH=/usr/lib64/cmake/GTest
JPEGXL_BUNDLE_LIBPNG:BOOL=OFF
JPEGXL_BUNDLE_SKCMS:BOOL=ON
JPEGXL_DEP_LICENSE_DIR:STRING=
JPEGXL_ENABLE_BENCHMARK:BOOL=OFF
JPEGXL_ENABLE_BOXES:BOOL=ON
JPEGXL_ENABLE_COVERAGE:BOOL=OFF
JPEGXL_ENABLE_DEVTOOLS:BOOL=ON
JPEGXL_ENABLE_DOXYGEN:BOOL=ON
JPEGXL_ENABLE_EXAMPLES:BOOL=ON
JPEGXL_ENABLE_FUZZERS:BOOL=
JPEGXL_ENABLE_JNI:BOOL=ON
JPEGXL_ENABLE_JPEGLI:BOOL=ON
JPEGXL_ENABLE_JPEGLI_LIBJPEG:BOOL=ON
JPEGXL_ENABLE_MANPAGES:BOOL=ON
JPEGXL_ENABLE_OPENEXR:BOOL=ON
JPEGXL_ENABLE_PLUGINS:BOOL=OFF
JPEGXL_ENABLE_PROFILER:BOOL=OFF
JPEGXL_ENABLE_SIZELESS_VECTORS:BOOL=OFF
JPEGXL_ENABLE_SJPEG:BOOL=OFF
JPEGXL_ENABLE_SKCMS:BOOL=OFF
JPEGXL_ENABLE_TCMALLOC:BOOL=ON
JPEGXL_ENABLE_TOOLS:BOOL=ON
JPEGXL_ENABLE_TRANSCODE_JPEG:BOOL=ON
JPEGXL_ENABLE_VIEWERS:BOOL=OFF
JPEGXL_FORCE_NEON:BOOL=false
JPEGXL_FORCE_SYSTEM_BROTLI:BOOL=ON
JPEGXL_FORCE_SYSTEM_GTEST:BOOL=ON
JPEGXL_FORCE_SYSTEM_HWY:BOOL=ON
JPEGXL_FORCE_SYSTEM_LCMS2:BOOL=ON
JPEGXL_STATIC:BOOL=false
JPEGXL_WARNINGS_AS_ERRORS:BOOL=OFF
SPHINX_BUILD_PROGRAM:FILEPATH=SPHINX_BUILD_PROGRAM-NOTFOUND
benchmark_DIR:PATH=benchmark_DIR-NOTFOUND
Expected behavior Test suite should not fail.
Environment
- OS: Linux x86/64
- Compiler version: gcc 13.0.1
- CPU type: Intel® Xeon® Silver 4116
- cjxl/djxl version string: [e.g. cjxl [v0.3.7 | SIMD supported: SSE4,Scalar]]
Additional context I’m not sure what more I can add 🤔
About this issue
- Original URL
- State: open
- Created a year ago
- Comments: 22 (14 by maintainers)
Commits related to this issue
- media-libs/libjxl: wire up tests These are CURRENTLY FAILING: https://github.com/libjxl/libjxl/issues/2433 Bug: https://bugs.gentoo.org/908210 Signed-off-by: Matoro Mahri <matoro@users.noreply.gith... — committed to matoro/gentoo by matoro a year ago
- media-libs/libjxl: wire up tests These are CURRENTLY FAILING: https://github.com/libjxl/libjxl/issues/2433 Bug: https://bugs.gentoo.org/908210 Signed-off-by: Matoro Mahri <matoro@users.noreply.gith... — committed to matoro/gentoo by matoro a year ago
- media-libs/libjxl: wire up tests These are CURRENTLY FAILING: https://github.com/libjxl/libjxl/issues/2433 Bug: https://bugs.gentoo.org/908210 Closes: https://bugs.gentoo.org/907890 Signed-off-by: ... — committed to matoro/gentoo by matoro a year ago
- media-libs/libjxl: wire up tests These are CURRENTLY FAILING: https://github.com/libjxl/libjxl/issues/2433 Bug: https://bugs.gentoo.org/908210 Closes: https://bugs.gentoo.org/907890 Signed-off-by: ... — committed to matoro/gentoo by matoro a year ago
- media-libs/libjxl: fix tests Tests were broken with lcms2. Backporting this pair of patches from https://github.com/libjxl/libjxl/issues/2433 fixes the tests on little-endian. Closes: https://bugs.... — committed to matoro/gentoo by matoro a year ago
- media-libs/libjxl: fix tests Tests were broken with lcms2. Backporting this pair of patches from https://github.com/libjxl/libjxl/issues/2433 fixes the tests on little-endian. Closes: https://bugs.... — committed to matoro/gentoo by matoro a year ago
- media-libs/libjxl: fix tests Tests were broken with lcms2. Backporting this pair of patches from https://github.com/libjxl/libjxl/issues/2433 fixes the tests on little-endian. Closes: https://bugs.... — committed to gentoo/gentoo by matoro a year ago
I suppose we should add a github workflow testing a build with LCMS2
Current list of failing test on s390x:
Confirming that when applying https://github.com/libjxl/libjxl/pull/2596 and https://github.com/libjxl/libjxl/pull/2617 I get 100% pass on little-endian with lcms2.
Unfortunately on big-endian I get
94% tests passed, 217 tests failed out of 3488. This is comparing ppc64 BE to ppc64 LE on two VM’s configured identically except for endianness, so all of these failures are guaranteed to be endian-related and not some other cause.Some of these look like pretty simple byteswaps, others not so much. Here’s the full log: LastTest.log
Cannot apply that commit on top of of the last release and I see that it is already +300 commits since last release 😞 if it is only test suite issue it can wait until next release.
PS. IMO you should consider to release new versions with smaller batches of commits 😋