2021.1.1: Test failures

[Lastique]

When building oneTBB 2021.1.1 release on Kubuntu 20.10 x86_64, the following tests fail:

• 43 - test_limiter_node (INTERRUPT)
• 69 - conformance_arena_constraints (Failed)
• 125 - test_malloc_whitebox (SEGFAULT)
• 126 - test_malloc_atexit (SEGFAULT)
• 127 - test_malloc_overload (SEGFAULT)
• 128 - test_malloc_overload_disable (SEGFAULT)
• 129 - test_malloc_new_handler (SEGFAULT)

The test_limiter_node test is hanging indefinitely and in this case was interrupted with SIGINT.

The libtbbbind library CMake file was patched to be able to compile it with libhwloc 2.2.0, which is the version of the library that is available in Kubuntu 20.10 repos. This patch is here: fix_hwloc_detection.patch.txt

The library was also patched to work around #312. This patch is here: fix_inheriting_members.patch.txt (#318)

The full log is here: tbb_build_fail.txt

This was run on a Core i7 2600K.

[alexey-katranov]

Thank you for the report.

Can you please clarify what issues did you have with libtbbbind? Notify: @AlexVeprev , @ivankochin

[Lastique]

libtbbbind is simply excluded from the build because CMake script does not find neither 1.11 nor 2.0 version of libhwloc on the system (because the installed version is 2.2.0). This is described in #319.

[alexey-katranov]

@Lastique , do you use release build (not relwithdebinfo)? If so, can you try to add CXXFLAGS="-fno-ipa-cp-clone" to check if the issue relates to #342 (comment). Thank you in advance.

[Lastique]

do you use release build (not relwithdebinfo)?

Yes.

If so, can you try to add CXXFLAGS="-fno-ipa-cp-clone"

It fixed some tests but not conformance_arena_constraints and conformance_enumerable_thread_specific. The latter one does not fail 100% of the runs, only on some runs. I'm attaching a full build log.

tbb_build.txt

Having to disable compiler optimizations makes me suspect that something is not right with the code.

[anton-potapov]

@Lastique is this still reproducible with 2021.4 or latest master ?

[Lastique]

2021.4 does not compile:

In member function ‘set’,
    inlined from ‘set_my_tls_end_of_input’ at ./src/tbb/parallel_pipeline.cpp:246:29,
    inlined from ‘set_end_of_input’ at ./src/tbb/parallel_pipeline.cpp:465:52:
./src/tbb/tls.h:44:46: warning: ‘pthread_setspecific’ expecting 1 byte in a region of size 0 [-Wstringop-overread]
./src/tbb/tls.h: In function ‘set_end_of_input’:
/usr/include/pthread.h:1308:12: note: in a call to function ‘pthread_setspecific’ declared with attribute ‘access (none, 2)’
 1308 | extern int pthread_setspecific (pthread_key_t __key,
      |            ^
In file included from /build/tbb/test/tbbmalloc/test_malloc_new_handler.cpp:36:
In member function ‘void tbb::detail::r1::basic_tls<T>::set(T) [with T = bool]’,
    inlined from ‘T tbb::detail::r1::tls<T>::operator=(T) [with T = bool]’ at /build/tbb/test/tbbmalloc/../../src/tbb/tls.h:85:37,
    inlined from ‘void customNewHandler()’ at /build/tbb/test/tbbmalloc/test_malloc_new_handler.cpp:40:24:
/build/tbb/test/tbbmalloc/../../src/tbb/tls.h:44:46: error: ‘int pthread_setspecific(pthread_key_t, const void*)’ expecting 1 byte in a region of size 0 [-Werror=stringop-overread]
   44 |     void set( T value ) { pthread_setspecific(my_key, (void*)value); }
      |                           ~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~
compilation terminated due to -Wfatal-errors.
cc1plus: all warnings being treated as errors
make[4]: *** [test/CMakeFiles/test_malloc_new_handler.dir/build.make:85: test/CMakeFiles/test_malloc_new_handler.dir/tbbmalloc/test_malloc_new_handler.cpp.o] Error 1

/build/tbb/test/tbbmalloc/test_malloc_init_shutdown.cpp: In member function ‘void TestThread::operator()(std::size_t) const’:
/build/tbb/test/tbbmalloc/test_malloc_init_shutdown.cpp:145:28: error: ‘int pthread_setspecific(pthread_key_t, const void*)’ expecting 1 byte in a region of size 0 [-Werror=stringop-overread]
  145 |         pthread_setspecific(key, (const void*)42);
      |         ~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~
compilation terminated due to -Wfatal-errors.
cc1plus: all warnings being treated as errors
make[4]: *** [test/CMakeFiles/test_malloc_init_shutdown.dir/build.make:85: test/CMakeFiles/test_malloc_init_shutdown.dir/tbbmalloc/test_malloc_init_shutdown.cpp.o] Error 1

This is on Kubuntu 21.10 with gcc 11.2.

[Lastique]

After applying #519 to 2021.4, test_eh_algorithms fails:

 22/133 Test  #20: test_eh_algorithms .......................***Failed    4.65 sec
[doctest] doctest version is "2.4.6"
[doctest] run with "--help" for options
===============================================================================
./test/tbb/test_eh_algorithms.cpp:462:
TEST CASE:  parallel_for and parallel_reduce exception handling test #4

./test/tbb/test_eh_algorithms.cpp:374: FATAL ERROR: REQUIRE( g_CurExecuted <= minExecuted + (g_NumThreads-1)*g_NumThreads/2 ) is NOT correct!
  values: REQUIRE( 2048 <= 1941 )
  logged: Too many tasks survived exception

===============================================================================
[doctest] test cases:    22 |    21 passed | 1 failed | 0 skipped
[doctest] assertions: 71691 | 71690 passed | 1 failed |
[doctest] Status: FAILURE!

[Lastique]

After running the build a second time, test_eh_thread failed:

 54/133 Test  #61: test_eh_thread ...........................Child aborted***Exception:   0.19 sec
terminate called after throwing an instance of 'std::runtime_error'
terminate called recursively
[doctest] doctest version is "2.4.6"
[doctest] run with "--help" for options
===============================================================================
./test/tbb/test_eh_thread.cpp:88:
TEST CASE:  Too many threads

./test/tbb/test_eh_thread.cpp:133: FATAL ERROR: No exception was caught

./test/tbb/test_eh_thread.cpp:88: FATAL ERROR: test case CRASHED: SIGABRT - Abort (abnormal termination) signal

===============================================================================
[doctest] test cases: 1 | 0 passed | 1 failed | 0 skipped
[doctest] assertions: 3 | 2 passed | 1 failed |
[doctest] Status: FAILURE!

[alexey-katranov]

I believe test_eh_thread is not really stable. @ivankochin @Iliamish , do you have any ideas how to improve it?

@Lastique , is test_eh_thread the only failing test in your environment?

[Lastique]

I mentioned test_eh_algorithms as well above. I only ran the tests twice.

[anton-potapov]

@Lastique, does #697 helped with the test_eh_thread ?

[Lastique]

I built 2021.5.0 with #697 applied a couple times and got these failures:

The following tests FAILED:
        130 - test_malloc_atexit (SEGFAULT)
        131 - test_malloc_overload (SEGFAULT)
        132 - test_malloc_overload_disable (SEGFAULT)
        133 - test_malloc_new_handler (SEGFAULT)

Not test_eh_thread though, so I suppose it helped.

[sbonfert]

test_eh_thread seems to fail, if more than 1024 threads are already running when the test is started, since the outermost thread cannot be created. Maybe the thread limit should be chosen dynamically.

[nofuturre]

@Lastique is this issue still relevant?

[Lastique]

I don't know, I haven't been running TBB tests for a while.

[nofuturre]

If anyone encounter this issue in the future please open new issue with a link to this one