Fix crash for missing urdf in resource manager

[saikishor]

Fixes #1299

These changes are needed after the merging of #1271 and #1272. It seems like the gazebo_ros2_control doesn't load the URDF into the resource manager, and this is causing the CM services not to start.

https://github.com/saikishor/gazebo_ros2_control/blob/60e4b75f4a786c9563d7dd97c6d08a31ca69c5b1/gazebo_ros2_control/src/gazebo_ros2_control_plugin.cpp#L366-L371

The gazebo_ros2_control doesn't load_urdf into the resource manager instead, it sets up the components and initializes it. Loading the URDF with the load_urdf method will probably fail, because this method originally loads and initializes the hardware interfaces and this step might probably fail.

Solution 1

I thought it would be wiser to utilize the size of the components to check if the URDF is loaded. This way without any changes from the gazebo_ros2_control side, we can have this issue fixed.

Solution 2

We can either go with this fix or add a new default argument to the load_urdf method to not load_and_initialize_components

ros2_control/hardware_interface/src/resource_manager.cpp

Lines 759 to 796 in 3309c6d

	void ResourceManager::load_urdf(const std::string & urdf, bool validate_interfaces)
	{
	is_urdf_loaded__ = true;
	const std::string system_type = "system";
	const std::string sensor_type = "sensor";
	const std::string actuator_type = "actuator";
	const auto hardware_info = hardware_interface::parse_control_resources_from_urdf(urdf);
	for (const auto & individual_hardware_info : hardware_info)
	{
	if (individual_hardware_info.type == actuator_type)
	{
	std::scoped_lock guard(resource_interfaces_lock_, claimed_command_interfaces_lock_);
	resource_storage_->load_and_initialize_actuator(individual_hardware_info);
	}
	if (individual_hardware_info.type == sensor_type)
	{
	std::lock_guard<std::recursive_mutex> guard(resource_interfaces_lock_);
	resource_storage_->load_and_initialize_sensor(individual_hardware_info);
	}
	if (individual_hardware_info.type == system_type)
	{
	std::scoped_lock guard(resource_interfaces_lock_, claimed_command_interfaces_lock_);
	resource_storage_->load_and_initialize_system(individual_hardware_info);
	}
	}
	// throw on missing state and command interfaces, not specified keys are being ignored
	if (validate_interfaces)
	{
	validate_storage(hardware_info);
	}
	std::lock_guard<std::recursive_mutex> guard(resources_lock_);
	read_write_status.failed_hardware_names.reserve(
	resource_storage_->actuators_.size() + resource_storage_->sensors_.size() +
	resource_storage_->systems_.size());
	}

This is also interesting as in the future, if we want to get the URDF information into the resource manager to parse things like limits or other entities

[codecov]

Codecov Report

Attention: 6 lines in your changes are missing coverage. Please review.

Comparison is base (3309c6d) 48.02% compared to head (89ec65d) 48.01%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1301      +/-   ##
==========================================
- Coverage   48.02%   48.01%   -0.02%     
==========================================
  Files          41       41              
  Lines        3525     3526       +1     
  Branches     1912     1913       +1     
==========================================
  Hits         1693     1693              
  Misses        442      442              
- Partials     1390     1391       +1     

Flag	Coverage Δ
unittests	48.01% <60.00%> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
hardware_interface/src/resource_manager.cpp	48.83% <60.00%> (-0.08%)	⬇️

[christophfroehlich]

I confirm that it works with ros2_control_demo_example_9 without the PR to gazebo_ros2_control.

This is also interesting as in the future, if we want to get the URDF information into the resource manager to parse things like limits or other entities

This PR does not break the use of the URDF in the ressource_manager?

A small note: For semantic reasons, maybe a change to load_urdf instead of check for the component sizes would be better.

But this is a quick fix for the broken simulation now. (I guess ign_ros2_control is affected too?)

[saikishor]

A small note: For semantic reasons, maybe a change to load_urdf instead of check for the component sizes would be better.

Thanks for the feedback. Tomorrow, I'll work on the changes to load_urdf method

[destogl]

Looks good! Thanks!

[christophfroehlich]

LGTM