ROS - GDB

Introduction

GDB is EXTREMELY handy. I used to doing a lot of print statements, but now I use GDB a lot. The reason is simple: I can check multiple things at once without recompiling everytime, especially my compilation time is long.

Concepts

Debug Symbols in C or C++ are mappings from machine code to function names, variable names, line numbers, etc.

E.g., in a binary, i is just a memory location. When built with a Debug flag (see the Usage section), we know the variable’s name.

1
int i = 1;

Debugging symbols do not impact the binary execution logic. It’s purely informational and consumed by gdb. The binary filesize though would increase.

Compilation

• In CMakeLists.txt

1
set(CMAKE_BUILD_TYPE Debug)

1
- Or alternatively, `catkin_make -DCMAKE_BUILD_TYPE=Debug`

• In gcc:

1
gcc -g -o myprog myprogram.c

Run GDB

• In a ROS Launch file:

1
<node pkg="rgbd_slam_rico" type="rgbd_rico_slam_mini" name="rgbd_rico_slam_mini" output="screen" launch-prefix="gdb -ex run --args"/>

• Alternatively, you can run the binary using the command gdb:

1
gdb ./devel/lib/<PACKAGE_NAME>/binary

GDB Usage

Some common usages include:

• b <FILE_NAME>:<LINE_NUM> set a breakpoint
• r: run
• c: continue
• p <var_name>: print a variable name
• info breakpoints: check which breakpoints are available
• delete <BREAKPOINT_NUM>: to delete a break point’s number

Some cautions:

• When debug printing an element in a cv::Mat, we need to specify its data type. To print it as a:
  • double (CV_64F), do (gdb) x/4g var. 4 means the first 4 values, g means double precision.
  • float (CV_32F): (gdb) x/4f var
  • int: (gdb) x/4d var

With ROS2

1
2
3
4
5
6
7
8
9
10
colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=Debug
gdb --args ./test_direct_3d_ndt_lo --gtest_catch_exceptions=0

# 5 In gdb:
(gdb) catch throw   # break on any C++ throw
(gdb) run
# gdb will now break inside std::_Deque_base<...>::_M_range_check
# you can then:
(gdb) bt           # to see the backtrace
(gdb) info locals  # to inspect variables

Core Dump

Assume we have multiple C++ nodes that crash after running a long time (so they are hard to reproduce). What would be a good way to start debugging?

• Core dump

A “core dump” is a snapshot of a computer program’s memory state, typically taken when the program crashes or terminates abnormally, and it’s used to help diagnose the cause of the crash.

• How to get it working:
  1. ulimit -c unlimited
  2. Adjust kernel settings to store it in /tmp:

     1
      cat /proc/sys/kernel/core_pattern
     

  3. To change it: sudo sysctl -w kernel.core_pattern=/tmp/core.%e.%p.%h.%t
  4. /etc/systemd/system.conf

     1
      DefaultLimitCORE=infinity
     

  5. sudo systemctl daemon-reload
  6. Create a crash program:

     1
     2
     3
     4
     5
     6
     7
      #include <stdio.h>
      int main() {
          // Intentionally dereference a NULL pointer to cause a segmentation fault.
          int *ptr = NULL;
          *ptr = 42;
          return 0;
      }
     

  7. Compile the executable with debugging symbols:

     1
      gcc -o0 crash_example crash_example.c
     

  8. Temporarily Enable Core Dumps: ulimit -c unlimited
  9. Verify that the current core dump file name pattern with: cat /proc/sys/kernel/core_pattern
  10. Run the program: ./crash_example
  11. Launch gdb: gdb ./crash_example /tmp/core.<executable_name>.<pid>.<hostname>.<timestamp>: and run bt at gdb. There will be backtrace saved

• Previous
  
• Next
  C++ - GTest