Function ros2_medkit_integration_tests::run_demo_node

• Defined in File demo_node_main.hpp

Function Documentation

inline int ros2_medkit_integration_tests::run_demo_node(int argc, char **argv, const std::function<std::shared_ptr<rclcpp::Node>()> &node_factory)

Run a demo node with a graceful, race-free shutdown.

Integration-test shutdowns used to flake on a rotating cast of demo binaries with errors like:

“Unexpected condition: string capacity was zero for allocated data!

Exiting.” (rosidl_runtime_c / exit 255)

Root cause: rclcpp’s default SIGINT handler calls rclcpp::shutdown() asynchronously from a signal context, which invalidates the shared rcl context mid-run. An in-flight timer callback can then be holding a rosidl message whose string fields point at allocator state that is already partially torn down; destruction aborts.

Pattern used here (following “signalfd / sigwait thread” best practice):

1. Block SIGINT / SIGTERM in the main thread before any rclcpp work.

2. Call rclcpp::init with SignalHandlerOptions::None so rclcpp does NOT install its own handlers on top.

3. Construct the node + executor with signals still blocked.

4. Spawn a dedicated sigwait thread that blocks on sigwait() for SIGINT / SIGTERM and, on arrival, calls executor.cancel() to wake spin() from a known-good thread context.

5. executor.spin() returns cleanly; the main thread then removes the node from the executor and resets the shared_ptr while the rcl context is still valid. Message / publisher destructors therefore run against a consistent allocator and cannot trip the rosidl sequence-invariant abort.

6. rclcpp::shutdown() is called last, AFTER everything is destroyed.

Usage:

int main(int argc, char ** argv) { return ros2_medkit_integration_tests::run_demo_node(argc, argv, [] { return std::make_shared<MyDemoNode>(); }); }

The factory is called with signals already blocked and rclcpp::init already complete, so node constructors can create publishers / timers freely.