The solution proposal for panic in the program termination

[evshary]

I would like to raise a discussion about the possible solution for https://github.com/ros2/rmw_zenoh/tree/8306a63313cd9e21883344deff9abc782636a464?tab=readme-ov-file#crash-when-program-terminates
I thought we had reached an agreement on this in the meeting before. Nevertheless, it might be worth writing it down here to make it clearer.

I know we expect users to shutdown the ROS Context explicitly, but it would be good if there is a way to detect this, or we might need to add shutdown to all the packages without it. For example, in ros_gz used by nav2
ZettaScaleLabs#26 (comment)

The proposal:

The Zenoh session can't be closed while terminating the program due to the reason described in README. However, while calling rmw_shutdown, there is no available way to know if we can close the Zenoh session. Therefore, we come up with a mechanism to detect the termination of the program. We can register the atexit function, and while the entering the termination stage, it will toggle a flag. If the flag is set, we shouldn't touch the Zenoh session to avoid panic.
For more information, please check the PR here: https://github.com/ZettaScaleLabs/rmw_zenoh/pull/37/files

Some questions:

1. Is this logic only necessary with 1.0.0?
   • Not really. I can do the porting to rolling if we all agree with the changes.
2. How do we ensure two or more threads do not access is_exiting concurrently?
   • I thought it didn't cause any issue since this is a one-way flag, which means we don't toggle it back to false. However, I can add atomic here to avoid possible concurrency issues.
3. When the program terminates, static ordering is unknown so there is no guarantee that is_exiting is valid.
   • I'm not sure whether I misunderstood the spec or not. According to the spec here, the destructor will be called in a reverse order. I think that means the global variable static bool is_exiting should be valid before the Context is shutdown. Feel free to correct me if I'm wrong.

Let me know your thoughts. Thank you!

Also cc @JEnoch here

[evshary]

After the PR, we might not need this workaround
cc @yellowhatter here

It should be related to the PR eclipse-zenoh/zenoh#1632

[YuanYuYuan]

This should fix https://github.com/Yadunund/rmw_zenoh_shutdown_test