Design the ROS2 implementation for Clock class

[GiulioRomualdi]

This PR develops the ROS2 implementation for the Clock class in the system component. This PR uses plain CMake instead of ament_cmake (provided by ROS2). The required ROS version is humble.

The clock can be used as follows:

1. In c++

#include <BipedalLocomotion/System/RosClock.h>
#include <BipedalLocomotion/System/Clock.h>

#include <chrono>
#include <memory>

using namespace std::chrono_literals;

int main(int argc, char *argv[])
{
    BipedalLocomotion::System::ClockBuilder::setFactory(std::make_shared<BipedalLocomotion::System::RosClockFactory>(argc, argv));

    BipedalLocomotion::clock().sleepFor(2000ms);

    return 0;
}

2. In python

import bipedal_locomotion_framework as blf
import sys
import datetime

# Call this function to set the ros clock
blf.system.ClockBuilder.set_factory(blf.system.RosClockFactory(args=sys.argv))

start = blf.clock().now()
blf.clock().sleep_for(datetime.timedelta(seconds=7))

[traversaro]

@GiulioRomualdi GiulioRomualdi requested a review from traversaro

I am supposed to wait for it being not Draft.

[GiulioRomualdi]

Actually yes, there is a problem related to

m_clock.sleep_for(...)

function.
There seems that the context is not valid.
Looking online there seems that I should call rclcpp::init() before calling the clock functions, however this method should be called only once for each ROS node. So a possibility is to check if this has been already called by someone else and if not call it

[GiulioRomualdi]

The PR is ready

[S-Dafarra]

Does it make sense to mix ROS2 and std calls? Is it better to simply do anything?

[GiulioRomualdi]

You're right. Removed

[traversaro]

Not a big problem, but it could make sense to have std::this_thread::yield() in this context. In the end, sleep/sleepUntil are functions that in the end are waiting on some kind of OS synchronization primitive (mutex/condition variables), so it could make sense that you want to yield even if you are waiting on those. Anyhow, I do not think it is a big problem.

[S-Dafarra]

I just thought that the effect could have not been very predictable given the sleep implementation of ROS2, but maybe I am being paranoid.

[GiulioRomualdi]

In ros we should use the command spin so in theory yield is not needed. Shall we add a debug print that remember the user that this implementation of yield does nothing?

[traversaro]

Shall we add a debug print that remember the user that this implementation of yield does nothing?

I would avoid the print, as it would make writing generic code quite difficult (you would need to not call yield if you use the ROS2 clock). I would say let's merge it empty (safest choice), and then let's see if problems emerge.

[GiulioRomualdi]

Ok merged thank you all as usual ❤️

[traversaro]

@GiulioRomualdi GiulioRomualdi requested a review from traversaro

I am supposed to wait for it being not Draft.

Sorry, that was originally meant to be a question, I put a . just for a typo.