[Python] Deterministic Segmentation Fault on Python Exit

luojw-dwr commented 1 month ago

A deterministic behavior of segmentation fault on python exit is observed on my machine. Here is a minimal example on my machine:

import mtkahypar as mtkhp
mtkhp.initializeThreadPool(4) # any thread count
ctx = mtkhp.Context()
ctx.loadPreset(mtkhp.PresetType.DEFAULT)
ctx.setPartitioningParameters(2, 0.03, mtkhp.Objective.KM1)
HG = mtkhp.Hypergraph("a.hmetis", mtkhp.FileFormat.HMETIS) # any legal input file
# any following code
# any way of exit python

The input file is confirmed to be correct with standalone MtKaHyPar.
The hypergraph can be partitioned normally.
The segmentation fault does not happen if the hypergraph creation line is removed.
The program does not crash instantly if del HG even if call gc.collect() after that, however the segmentation fault still happens at exit in these two cases.

My dependency versions are: -1: mt-kahypar: git hash 035e595b762bb490f71ad1668ad00547e64653b4

python: 3.8.20
g++: 13.2
cmake: 3.28.3
boost: 1.86
TBB: ~~2024.2.1.105~~ 2021.13 (seems my mkl and tbb differs in version)
hwloc: 2.11.1

Please help to see if it is reproducible on developers' side.

luojw-dwr commented 1 month ago

The segmentation fault is now observed to be random but very frequent, as an accidental no-fault exit happens in my machine.

N-Maas commented 1 month ago

Hi @luojw-dwr,

thanks for the detailed report! I tried to reproduce the issue, but unfortunately without success yet.

Some followup questions that could help for reproduction:

Which operating system are you using?
Are you sure the input hypergraph doesn't matter? How large is the input roughly? Perhaps you could provide the input file?

It could also help if you try running MtKaHyPar in debug mode and report the output. I.e., call cmake in the build folder as follows (note: disabling address sanitizer since it doesn't seem to work with python):

cmake .. -DCMAKE_BUILD_TYPE=DEBUG -DKAHYPAR_ADD_ADDRESS_SANITIZER=Off

Then build the python library again and see whether the output changes (e.g., if we are lucky there might be a failing assertion instead of the segmentation fault).

Note: I didn't build with the same gcc version yet, I will probably try this next week.

luojw-dwr commented 1 month ago

Hi @N-Maas I am working on Ubuntu 24.04. The problem exists both in normal execution and in REPL. The minimal reproducible input a.hmetis is:

1 2
1 2

Larger inputs with ~1e7 vertices ~1e7 hyperedges also reproduce, no matter w/ or w/o weights.

The segmentation fault seems to disappear if the program after this code piece runs long enough (e.g. tens of minutes).

Sorry that I am working in a shared enviroment with fragile dependencies. Thank you for your local reproduction. I will try figure it out locally.

Thank you for your time and reply.

luojw-dwr commented 1 month ago

During make after cmake .. -DCMAKE_BUILD_TYPE=DEBUG in an empty build folder, I get the following log (where $HOME is due to my handle replacement in the log):

Running main() from $HOME/kahypar_home/mt-kahypar/external_tools/googletest/googletest/src/gtest_main.cc
[==========] Running 67 tests from 2 test suites.
[----------] Global test environment set-up.
[----------] 13 tests from MtKaHyPar
[ RUN      ] MtKaHyPar.ReadHypergraphFile
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:40:5: runtime error: reference binding to misaligned address 0x7eb45b00a460 for type 'struct __as_base ', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:40:5: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:33: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:33: runtime error: constructor call on misaligned address 0x7eb45b00a460 for type 'struct task', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/detail/_task.h:214:31: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct task', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/detail/_task.h:214:31: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct task', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/detail/_task.h:214:31: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct task', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  88 4c 1c c6 60 62 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/detail/_task.h:214:31: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct task', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  88 4c 1c c6 60 62 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:33: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  88 4c 1c c6 60 62 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:41:9: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  70 19 1c 7f b4 7e 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:41:9: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  70 19 1c 7f b4 7e 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:9: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  70 19 1c 7f b4 7e 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:9: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  70 19 1c 7f b4 7e 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:40:5: runtime error: reference binding to misaligned address 0x7eb45b00a520 for type 'struct __as_base ', which requires 64 byte alignment
0x7eb45b00a520: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:40:5: runtime error: member access within misaligned address 0x7eb45b00a520 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a520: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:33: runtime error: member access within misaligned address 0x7eb45b00a520 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a520: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00

... and many other such lines, and at last:

Assertion is_aligned(p, alignment == 0 ? alignof(T) : alignment) failed (located in the assert_pointer_valid function, line in file: 217)

This looks strange to me, as for my first build where I compiled as RELEASE, such logs didn't appear. I will try another DEBUG build with sanitizer disabled.

luojw-dwr commented 1 month ago

I noticed that my mkl and tbb have different version string, which may be the cause. I will try and report.

Edit: I reinstalled Intel oneAPI and re-compile from scratch. The version string remains the same. The behavior has not changed.

luojw-dwr commented 1 month ago

FYI, with a DEBUG build with sanitizer disabled in another empty folder, during make, I get the following log:

[ RUN      ] APartitioner.PartitionsAGraphInTwoBlocksWithQualityPreset
[multilevel_coarsener.h:387:matchVertices]: Assertion `_matching_state[rep] == static_cast<uint8_t>(MatchingState::MATCHED)` failed:
Aborted (core dumped)

which seems to be irrelevant, and more strangely, the alignment assertion is now silent.

With this build, with the new python so, the segmentation fault reproduces still silently.

larsgottesbueren commented 1 month ago

Can you try cmake .. -DKAHYPAR_DOWNLOAD_TBB=On -DCMAKE_BUILD_TYPE=Debug -DKAHYPAR_ADD_ADDRESS_SANITIZER=Off and let me know if this changes anything? This downloads a specified TBB version and builds mt-kahypar with it. Don't worry, that TBB version is not installed globally. We've had some trouble with different TBB versions in the past.

luojw-dwr commented 1 month ago

Can you try cmake .. -DKAHYPAR_DOWNLOAD_TBB=On -DCMAKE_BUILD_TYPE=Debug -DKAHYPAR_ADD_ADDRESS_SANITIZER=Off and let me know if this changes anything? This downloads a specified TBB version and builds mt-kahypar with it. Don't worry, that TBB version is not installed globally. We've had some trouble with different TBB versions in the past.

Hi @larsgottesbueren Thank you for your reply. I want to make sure in advance: what if the TBB searched by LD_LIBRARY_PATH differs from the downloaded TBB?

luojw-dwr commented 1 month ago

Hi @larsgottesbueren I tried your cmake in another empty build folder. The behavior is the same as cmake .. -DCMAKE_BUILD_TYPE=DEBUG -DKAHYPAR_ADD_ADDRESS_SANITIZER=Off, as in:

FYI, with a DEBUG build with sanitizer disabled in another empty folder, during make, I get the following log:
[ RUN      ] APartitioner.PartitionsAGraphInTwoBlocksWithQualityPreset
[multilevel_coarsener.h:387:matchVertices]: Assertion `_matching_state[rep] == static_cast<uint8_t>(MatchingState::MATCHED)` failed:
Aborted (core dumped)
which seems to be irrelevant, and more strangely, the alignment assertion is now silent.

With this build, with the new python so, the segmentation fault reproduces still silently.

larsgottesbueren commented 1 month ago

Given that we are not able to reproduce the issue on our side, let's discuss some possible steps to debug this issue.

1) Have you tried getting this to work on a different machine? Laptop or university lab machine...whatever is available 2) Would you be open to debugging this in C++ yourself?

N-Maas commented 1 month ago

Can you try cmake .. -DKAHYPAR_DOWNLOAD_TBB=On -DCMAKE_BUILD_TYPE=Debug -DKAHYPAR_ADD_ADDRESS_SANITIZER=Off and let me know if this changes anything? This downloads a specified TBB version and builds mt-kahypar with it. Don't worry, that TBB version is not installed globally. We've had some trouble with different TBB versions in the past.

Hi @larsgottesbueren Thank you for your reply. I want to make sure in advance: what if the TBB searched by LD_LIBRARY_PATH differs from the downloaded TBB?

With a clean build, it should automatically use the downloaded version. The -DKAHYPAR_DOWNLOAD_TBB doesn't work well with a partial build, so a clean build is most likely required. You can check by running ldd on the created .so file and see whether it points into the build directory for libtbb.

N-Maas commented 1 month ago

Given that we are not able to reproduce the issue on our side, let's discuss some possible steps to debug this issue.
1. Have you tried getting this to work on a different machine? Laptop or university lab machine...whatever is available

2. Would you be open to debugging this in C++ yourself?

I also can't reproduce the issue using the same gcc version. However, I now get the mentioned alignment errors in the tests. So the latter might be related to the compiler version.

Some additional questions:

Is your processor architecture something else than x86_64? E.g. ARM?
Could you try building with clang instead of gcc and see whether the issue reproduces?

luojw-dwr commented 1 month ago

Hi @larsgottesbueren and @N-Maas

I am using x86_64. My environment do not have reliable clang.

Sorry that I am hurrying in some project. Currently, debugging this is not of my highest priority. I will try debugging this later. Thank you for your time and patience.

By the way, I'd like to highlight the following phenomenon during the build, hope this will give some hint.

During make after cmake .. -DCMAKE_BUILD_TYPE=DEBUG in an empty build folder, I get the following log (where $HOME is due to my handle replacement in the log):

Running main() from $HOME/kahypar_home/mt-kahypar/external_tools/googletest/googletest/src/gtest_main.cc
[==========] Running 67 tests from 2 test suites.
[----------] Global test environment set-up.
[----------] 13 tests from MtKaHyPar
[ RUN      ] MtKaHyPar.ReadHypergraphFile
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:40:5: runtime error: reference binding to misaligned address 0x7eb45b00a460 for type 'struct __as_base ', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:40:5: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:33: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:33: runtime error: constructor call on misaligned address 0x7eb45b00a460 for type 'struct task', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/detail/_task.h:214:31: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct task', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/detail/_task.h:214:31: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct task', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/detail/_task.h:214:31: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct task', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  88 4c 1c c6 60 62 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/detail/_task.h:214:31: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct task', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  88 4c 1c c6 60 62 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:33: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  88 4c 1c c6 60 62 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:41:9: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  70 19 1c 7f b4 7e 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:41:9: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  70 19 1c 7f b4 7e 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:9: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  70 19 1c 7f b4 7e 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:9: runtime error: member access within misaligned address 0x7eb45b00a460 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a460: note: pointer points here
 00 00 00 00  70 19 1c 7f b4 7e 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:40:5: runtime error: reference binding to misaligned address 0x7eb45b00a520 for type 'struct __as_base ', which requires 64 byte alignment
0x7eb45b00a520: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:40:5: runtime error: member access within misaligned address 0x7eb45b00a520 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a520: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00
              ^
$HOME/intel/oneapi/tbb/2021.13/include/oneapi/tbb/parallel_invoke.h:42:33: runtime error: member access within misaligned address 0x7eb45b00a520 for type 'struct function_invoker', which requires 64 byte alignment
0x7eb45b00a520: note: pointer points here
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00

... and many other such lines, and at last:

Assertion is_aligned(p, alignment == 0 ? alignof(T) : alignment) failed (located in the assert_pointer_valid function, line in file: 217)

This looks strange to me, as for my first build where I compiled as RELEASE, such logs didn't appear. I will try another DEBUG build with sanitizer disabled.

luojw-dwr commented 1 month ago

Hi @N-Maas My trial with downloaded cmake as cmake .. -DKAHYPAR_DOWNLOAD_TBB=On -DCMAKE_BUILD_TYPE=Debug -DKAHYPAR_ADD_ADDRESS_SANITIZER=Off is here:

Hi @larsgottesbueren I tried your cmake in another empty build folder. The behavior is the same as cmake .. -DCMAKE_BUILD_TYPE=DEBUG -DKAHYPAR_ADD_ADDRESS_SANITIZER=Off, as in:
FYI, with a DEBUG build with sanitizer disabled in another empty folder, during make, I get the following log:
[ RUN      ] APartitioner.PartitionsAGraphInTwoBlocksWithQualityPreset
[multilevel_coarsener.h:387:matchVertices]: Assertion `_matching_state[rep] == static_cast<uint8_t>(MatchingState::MATCHED)` failed:
Aborted (core dumped)
which seems to be irrelevant, and more strangely, the alignment assertion is now silent. With this build, with the new python so, the segmentation fault reproduces still silently.

luojw-dwr commented 1 month ago

Unluckily, the adminstrator of my server refuses to provide other compilers. Fortunately, mtkhp.initializeThreadPool(1) avoids the segmentation fault. Thanks for all your time and patience.

edit: For those who are concerned with g++-13, -fno-ipa-stack-alignment does not help. Please refer to #192 .

edit: The problem is now confirmed to be strongly coupled with the TBB version.

luojw-dwr commented 1 month ago

Testing with clang-18 and clang++-18 in Release mode replicates the segmentation fault.

edit: Debug (with ASAN) build in clang++ does not report misalignment. Instead, memory leaks are reported by ASAN, which seems to be irrelevant. edit: Debug (without ASAN) build in clang++ passes the tests during make without errors. I'm trying with the python binding, but struggling with undefined symbol to ubsan in python3 REPL.

-2: Ubuntu 24.04, x86_64 -1: mt-kahypar: git hash 39e06c755cb909387a66eed24ee385ba5ad393d4

python: tested with 3.8.20 and 3.12.3
compiler: tested with g++-13 and clang++-18
cmake: 3.28.3
boost: 1.86
TBB: 2021.13
hwloc: 2.11.1

I will try -DKAHYPAR_DOWNLOAD_TBB=On. Notice that cmake prioritizes the previously installed TBB over the downloaded TBB, which is confirmed by ldd.

luojw-dwr commented 1 month ago

With the downloaded TBB, python3 now correctly exits. Thank you for all your time and patience :-D

-2: Ubuntu 24.04, x86_64 -1: mt-kahypar: git hash 39e06c755cb909387a66eed24ee385ba5ad393d4

python: 3.8.20
compiler: clang++-18
cmake: 3.28.3
boost: 1.86
TBB: 2021.7, by -DKAHYPAR_DOWNLOAD_TBB=On
hwloc: 2.11.1

kahypar / mt-kahypar

[Python] Deterministic Segmentation Fault on Python Exit #188