ArduPilot SITL : SIGSEGV Sending message 76

[depauwjimmy]

Hi,

I am trying the SDK again starting a new project.
I have a simulation environment running gazebo/ardu sitl/ros2 and i have an issue when creating System.

Ardu version : Copter-4.5.7
MAVSDK : 2.12.10

Running logs are

[03:15:53|Info ] MAVSDK version: v2.12.10 (mavsdk_impl.cpp:26)
[03:15:53|Info ] New system on: 127.0.0.1:33815 (with system ID: 1) (udp_connection.cpp:178)
[03:15:53|Debug] New system ID: 1 Comp ID: 1 (mavsdk_impl.cpp:750)
[03:15:53|Debug] Component Autopilot (1) added. (system_impl.cpp:366)
[03:15:53|Warn ] Vehicle type changed (new type: 2, old type: 0) (system_impl.cpp:217)
[03:15:53|Debug] Discovered 1 component(s) (system_impl.cpp:509)

Sometimes it stops there and no crashes happens.
But other times i have

[03:15:53|Debug] Sending message 76 from 176/224 to 1/1 (mavsdk_impl.cpp:444)

and it crash immediately.
The backtrace is :

#0  0x00007ffff621afe0 in main_arena () from /lib/x86_64-linux-gnu/libc.so.6
#1  0x00007ffff73ac8bb in std::function<bool (__mavlink_message&)>::operator()(__mavlink_message&) const (this=0x7ffff621afd8 <main_arena+856>, __args#0=...) at /usr/include/c++/11/bits/std_function.h:590
#2  0x00007ffff73a1949 in mavsdk::MavsdkImpl::send_message (this=0x7ffff621ace0 <main_arena+96>, message=...) at MAVSDK/src/mavsdk/core/mavsdk_impl.cpp:453
#3  0x00007ffff7476e0b in mavsdk::ServerComponentImpl::queue_message(std::function<__mavlink_message (MavlinkAddress, unsigned char)>) (this=0x555555a084c0, fun=...) at MAVSDK/src/mavsdk/core/server_component_impl.cpp:170
#4  0x00007ffff7375878 in mavsdk::SystemImpl::queue_message(std::function<__mavlink_message (MavlinkAddress, unsigned char)>) (this=0x7fffbc000e50, fun=...) at MAVSDK/src/mavsdk/core/system_impl.cpp:469
#5  0x00007ffff73c9849 in mavsdk::MavlinkCommandSender::send_mavlink_message (this=0x7fffbc001020, command=std::variant [index 1] = {...}) at MAVSDK/src/mavsdk/core/mavlink_command_sender.cpp:461
#6  0x00007ffff73c8ff2 in mavsdk::MavlinkCommandSender::do_work (this=0x7fffbc001020) at MAVSDK/src/mavsdk/core/mavlink_command_sender.cpp:401
#7  0x00007ffff737483d in mavsdk::SystemImpl::system_thread (this=0x7fffbc000e50) at MAVSDK/src/mavsdk/core/system_impl.cpp:280
#8  0x00007ffff739a7ca in std::__invoke_impl<void, void (mavsdk::SystemImpl::*)(), mavsdk::SystemImpl*> (__f=@0x7fffbc001dd0: (void (mavsdk::SystemImpl::*)(mavsdk::SystemImpl * const)) 0x7ffff7374762 <mavsdk::SystemImpl::system_thread()>, __t=@0x7fffbc001dc8: 0x7fffbc000e50) at /usr/include/c++/11/bits/invoke.h:74
#9  0x00007ffff739a71d in std::__invoke<void (mavsdk::SystemImpl::*)(), mavsdk::SystemImpl*> (__fn=@0x7fffbc001dd0: (void (mavsdk::SystemImpl::*)(mavsdk::SystemImpl * const)) 0x7ffff7374762 <mavsdk::SystemImpl::system_thread()>) at /usr/include/c++/11/bits/invoke.h:96
#10 0x00007ffff739a67d in std::thread::_Invoker<std::tuple<void (mavsdk::SystemImpl::*)(), mavsdk::SystemImpl*> >::_M_invoke<0ul, 1ul> (this=0x7fffbc001dc8) at /usr/include/c++/11/bits/std_thread.h:259
#11 0x00007ffff739a632 in std::thread::_Invoker<std::tuple<void (mavsdk::SystemImpl::*)(), mavsdk::SystemImpl*> >::operator() (this=0x7fffbc001dc8) at /usr/include/c++/11/bits/std_thread.h:266
#12 0x00007ffff739a612 in std::thread::_State_impl<std::thread::_Invoker<std::tuple<void (mavsdk::SystemImpl::*)(), mavsdk::SystemImpl*> > >::_M_run (this=0x7fffbc001dc0) at /usr/include/c++/11/bits/std_thread.h:211
#13 0x00007ffff64dc253 in ?? () from /lib/x86_64-linux-gnu/libstdc++.so.6
#14 0x00007ffff6094ac3 in start_thread (arg=<optimized out>) at ./nptl/pthread_create.c:442
#15 0x00007ffff6126850 in clone3 () at ../sysdeps/unix/sysv/linux/x86_64/clone3.S:81

Code is basic

auto conf = Mavsdk::Configuration{Mavsdk::ComponentType::CompanionComputer};
Mavsdk mavsdk{conf};
auto con_result = mavsdk.add_udp_connection("127.0.0.1", 14550);
if (con_result != ConnectionResult::Success) {
  return false;
}

auto systemOpt = mavsdk.first_autopilot(3.0);
if (!systemOpt.has_value()) {
  return false;
}

Not sure what to do at this point.
Thanks for any advices

[julianoes]

Can you share the example, so I can reproduce it myself?

And maybe you can also share the stacktrace of all threads, in gdb it's thread apply all bt.

And you could also run it using valgrind, that sometimes tells you more.

[depauwjimmy]

Of course.
For some reason, doing an example with only code inside main never triggers the message sending. Doing it in a class method does it.
I also have a runloop (ros2 in my case) but it happens with anything that waits so i just put a sleep.

#include <mavsdk/mavsdk.h>
#include <unistd.h>

using namespace mavsdk;

class Runtime {
private:
    std::shared_ptr<mavsdk::System> system;

public:
    bool init() {
        auto conf = Mavsdk::Configuration{Mavsdk::ComponentType::CompanionComputer};

        Mavsdk mavsdk{conf};
        auto con_result = mavsdk.add_udp_connection("127.0.0.1", 14550);

        if (con_result != ConnectionResult::Success) {
            return false;
        }

        auto systemOpt = mavsdk.first_autopilot(3.0);
        if (!systemOpt.has_value()) {
            return false;
        }

        this->system = systemOpt.value();
        return true;
    }
};

int main(int argc, char **argv) {
    Runtime runtime;
    runtime.init();
    sleep(10);
    return EXIT_SUCCESS;
}

Full stack does not give anything more. There's no other thread in the stacktrace (other than the sleep)

[julianoes]

Ok, I see the problem. You need to keep Mavsdk alive. You can put it in a std::shared_ptr or std::unique_ptr. In your current implementation it goes out of scope and is destructed.

Also see this comment:
https://github.com/mavlink/MAVSDK/blob/v2.12/src/mavsdk/core/include/mavsdk/mavsdk.h#L35-L36

It might not be the best wording. Probably should be something like:

 * An instance of this class must be created and kept alive in order to use the library.
 * The instance can be destroyed after use in order to break connections and release all resources.