cf-natali opened a new pull request #395:
URL: https://github.com/apache/mesos/pull/395
Mesos is currently unusable on ARM64 - and probably other architectures.
Here's an example test failure:
```
[ RUN ] JsonTest.ParseError
terminate called after throwing an instance of 'std::overflow_error'
terminate called recursively
*** Aborted at 1622796321 (unix time) try "date -d @1622796321" if you
are using GNU date ***
PC: @ 0x0 (unknown)
*** SIGABRT (@0x3e80000090c) received by PID 2316 (TID 0xffff918cf010)
from PID 2316; stack trace: ***
@ 0xffff918dd7fc ([vdso]+0x7fb)
@ 0xffff91367188 gsignal
@ 0xffff91353dac abort
@ 0xffff91569848 __gnu_cxx::__verbose_terminate_handler()
@ 0xffff915671ec (unknown)
@ 0xffff91567250 std::terminate()
@ 0xffff915675b0 __cxa_rethrow
@ 0xffff915697e4 __gnu_cxx::__verbose_terminate_handler()
@ 0xffff915671ec (unknown)
@ 0xffff91567250 std::terminate()
@ 0xffff91567544 __cxa_throw
@ 0xaaaab4e3d324 picojson::value::value()
@ 0xaaaab4e421d8 JSON::internal::ParseContext::set_number()
@ 0xaaaab4e52140 picojson::_parse<>()
@ 0xaaaab4e63e74
JSON::internal::ParseContext::parse_object_item<>()
@ 0xaaaab4e5ae34 picojson::_parse_object<>()
@ 0xaaaab4e51fd4 picojson::_parse<>()
@ 0xaaaab4e49e18 picojson::_parse<>()
@ 0xaaaab4e426e4 JSON::parse()
@ 0xaaaab4e4a38c JSON::parse<>()
@ 0xaaaab4ecce28 JsonTest_ParseError_Test::TestBody()
@ 0xaaaab5186fec
testing::internal::HandleSehExceptionsInMethodIfSupported<>()
@ 0xaaaab517f1d4
testing::internal::HandleExceptionsInMethodIfSupported<>()
@ 0xaaaab515a9d0 testing::Test::Run()
@ 0xaaaab515b258 testing::TestInfo::Run()
@ 0xaaaab515b8d0 testing::TestCase::Run()
@ 0xaaaab5162344 testing::internal::UnitTestImpl::RunAllTests()
@ 0xaaaab5188440
testing::internal::HandleSehExceptionsInMethodIfSupported<>()
@ 0xaaaab517ffd4
testing::internal::HandleExceptionsInMethodIfSupported<>()
@ 0xaaaab516100c testing::UnitTest::Run()
@ 0xaaaab4f30950 RUN_ALL_TESTS()
@ 0xaaaab4f30418 main
```
Looking at the code shows that exceptions which are supposed to be
caught are actually not caught, causing `std::terminate` to be called,
and `std::terminate` itself breaks and is called recursively.
It can be reproduced with the following code:
```
charles-natali@martin-arm64:~$ cat test.cpp
int main()
{
try {
throw std::overflow_error("");
} catch (const std::overflow_error&) {
}
}
```
This works, as expected:
```
charles-natali@martin-arm64:~$ g++ -o test test.cpp && ./test
charles-natali@martin-arm64:~$
```
However linking with `-lunwind` - which is used by glog to get
tracebacks in case of fatal error - breaks exception handling:
```
charles-natali@martin-arm64:~$ g++ -o test test.cpp -lunwind && ./test
terminate called after throwing an instance of 'std::overflow_error'
terminate called recursively
Aborted (core dumped)
charles-natali@martin-arm64:~$
```
The reasons are explained in this bug report - for IA-64 but the problem
is the same on ARM64 - but basically it breaks libgcc unwinding code:
https://bugzilla.redhat.com/show_bug.cgi?id=480412
The solution is to make sure we pass `-lgcc_s` before `-lunwind`:
```
charles-natali@martin-arm64:~$ g++ -o test test.cpp -lgcc_s -lunwind &&
./test;
charles-natali@martin-arm64:~$
```
Tested on both x86-64 and ARM64, with gcc (whole Mesos test suite) and
clang (only standalone reproducer).
Closes MESOS-10223.
@qianzhangxa
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
