ros_realtime: Publisher is crashing during destruction

[ablasdel]

Migrated from code.ros.org, issue #2877
Reported by: sglaser

https://code.ros.org/trac/ros/ticket/2877

I'm bringing up a nodelet which brings up a rosrt publisher, and then bringing down the nodelet and it crashes:

Program received signal SIGTRAP, Trace/breakpoint trap.
[Switching to Thread 0x7f167cff9950 (LWP 9727)]
ros::Publisher::publish<tirt_controller::JointSetpoint_<std::allocator<void> > const> (this=0x7f16940ef458, message=@0x0)
    at /u/sglaser/ros/ros/core/roscpp/include/ros/publisher.h:85
85        }
Current language:  auto; currently c++
(gdb) bt
#0  ros::Publisher::publish<tirt_controller::JointSetpoint_<std::allocator<void> > const> (this=0x7f16940ef458, message=@0x0)
    at /u/sglaser/ros/ros/core/roscpp/include/ros/publisher.h:85
#1  0x00007f168fdf2c32 in rosrt::detail::publishMessage<tirt_controller::JointSetpoint_<std::allocator<void> > > (pub=@0x7f16bb532720,
    msg=<value optimized out>) at /u/sglaser/ros/stacks/ros_realtime/rosrt/include/rosrt/publisher.h:59
#2  0x00007f1684f8f146 in rosrt::detail::PublishQueue::publishAll (this=<value optimized out>)
    at /u/sglaser/ros/stacks/ros_realtime/rosrt/src/publisher.cpp:80
#3  0x00007f1684f8f373 in rosrt::detail::PublisherManager::publishThread (this=0x7f1680066ce0)
    at /u/sglaser/ros/stacks/ros_realtime/rosrt/src/publisher.cpp:126
#4  0x00007f16bb102a0e in thread_proxy () from /opt/ros/lib/libboost_thread.so.1.40.0
#5  0x00007f16b98ef3ba in start_thread () from /lib/libpthread.so.0
#6  0x00007f16b965c02d in clone () from /lib/libc.so.6
#7  0x0000000000000000 in ?? ()

Not sure why this is dying, but it is consistent.

[ablasdel]

Changed 3 years ago by sglaser

Here's the code doing the publishing:

  void update(ros::Time time)
  {
    tirt_controller::JointSetpoint::Ptr setpoint;
    if (setpoint = pub_setpoint_.allocate())
    {
      setpoint->mode = tirt_controller::JointSetpoint::POSITION;
      double scale = 2 * M_PI / period_;
      setpoint->position = amplitude_ * sin(time.toSec() * scale) + offset_;
      setpoint->velocity = amplitude_ * scale * cos(time.toSec() * scale);
      pub_setpoint_.publish(setpoint);
    }
    else
      ROS_ERROR_THROTTLE(5.0, "Runout");
  }

Looks like the publisher is invalid:

(gdb) frame 1
#1  0x00007f82666cdc32 in rosrt::detail::publishMessage<tirt_controller::JointSetpoint_<std::allocator<void> > > (pub=@0x7f829535e720, 
    msg=<value optimized out>) at /u/sglaser/ros/stacks/ros_realtime/rosrt/include/rosrt/publisher.h:59
59        pub.publish(m);
(gdb) p m
$2 = {px = 0x7f825c000da0, pn = {pi_ = 0x7f825c064bc0}}
(gdb) p pub
$3 = (const ros::Publisher &) @0x7f829535e720: {impl_ = {px = 0x0, pn = {pi_ = 0x0}}}

Other things for curiosity's sake:

(gdb) frame 2
#2  0x00007f826b2f2146 in rosrt::detail::PublishQueue::publishAll (this=<value optimized out>)
    at /u/sglaser/ros/stacks/ros_realtime/rosrt/src/publisher.cpp:80
80          it->val.pub_func(it->val.pub, clone);
(gdb) p count
$4 = 34

What's particularly odd is that this publisher was successfully publishing for a while, and it is only on shutdown that it suddenly goes NULL.
Changed 3 years ago by jfaust

My guess is it's because you've removed the code segment for that publisher (and what it's referencing) from memory. Does it happen if the publisher is just destroyed, but the nodelet is not unloaded?